Itk Software Guide 5.0
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 1001
| Download | |
| Open PDF In Browser | View PDF |
The ITK Software Guide
The Insight Toolkit (ITK) is an open-source, cross-platform system for medical image processing. It
provides medical imaging researchers with an extensive suite of leading-edge algorithms for
registering, segmenting, analyzing, and quantifying medical data. It is used in thousands of research
and commercial applications, from major labs to individual innovators.
This ITK Software Guide is the handbook for developing software with ITK. It is divided into two
companion books.
The first book covers building and installation, general architecture and design, as well as the process
of contributing in the ITK community.
The second book covers detailed design and functionality for reading and writing images, filtering,
registration, segmentation, and performing statistical analysis.
NOTICE: This PDF is a concatenation of both Book 1 and Book 2 of the ITK Software Guide into a
single document.
The ITK Software Guide
Book 1: Introduction and Development Guidelines ……..…Pages 2- 424
Book 2: Design and Functionality ………………………….Pages 425-1001
The ITK Software Guide
Book 1: Introduction and Development
Guidelines
Fourth Edition
Updated for ITK version 5.0.0
Hans J. Johnson, Matthew M. McCormick, Luis Ibáñez,
and the Insight Software Consortium
January 18, 2019
https://itk.org
https://discourse.itk.org/
The purpose of computing is Insight, not numbers.
Richard Hamming
ABSTRACT
The National Library of Medicine Insight Segmentation and Registration Toolkit, shortened as the
Insight Toolkit (ITK), is an open-source software toolkit for performing registration and segmentation. Segmentation is the process of identifying and classifying data found in a digitally sampled
representation. Typically the sampled representation is an image acquired from such medical instrumentation as CT or MRI scanners. Registration is the task of aligning or developing correspondences
between data. For example, in the medical environment, a CT scan may be aligned with a MRI scan
in order to combine the information contained in both.
ITK is a cross-platform software. It uses a build environment known as CMake to manage platformspecific project generation and compilation process in a platform-independent way. ITK is implemented in C++. ITK’s implementation style employs generic programming, which involves the
use of templates to generate, at compile-time, code that can be applied generically to any class or
data-type that supports the operations used by the template. The use of C++ templating means that
the code is highly efficient and many issues are discovered at compile-time, rather than at run-time
during program execution. It also means that many of ITK’s algorithms can be applied to arbitrary
spatial dimensions and pixel types.
An automated wrapping system integrated with ITK generates an interface between C++ and a highlevel programming language Python. This enables rapid prototyping and faster exploration of ideas
by shortening the edit-compile-execute cycle. In addition to automated wrapping, the SimpleITK
project provides a streamlined interface to ITK that is available for C++, Python, Java, CSharp, R,
Tcl and Ruby.
Developers from around the world can use, debug, maintain, and extend the software because ITK
is an open-source project. ITK uses a model of software development known as Extreme Programming. Extreme Programming collapses the usual software development methodology into a simultaneous iterative process of design-implement-test-release. The key features of Extreme Programming
are communication and testing. Communication among the members of the ITK community is what
helps manage the rapid evolution of the software. Testing is what keeps the software stable. An
extensive testing process supported by the system known as CDash measures the quality of ITK
code on a daily basis. The ITK Testing Dashboard is updated continuously, reflecting the quality of
the code at any moment.
The
most
recent
version
of
this
document
is
available
online
at
https://itk.org/ItkSoftwareGuide.pdf.
This book is a guide for developing software
with ITK; it is the first of two companion books. This book covers building and installation, general
architecture and design, as well as the process of contributing in the ITK community. The second
book covers detailed design and functionality for reading and writing images, filtering, registration,
segmentation, and performing statistical analysis.
CONTRIBUTORS
The Insight Toolkit (ITK) has been created by the efforts of many talented individuals and prestigious organizations. It is also due in great part to the vision of the program established by Dr. Terry
Yoo and Dr. Michael Ackerman at the National Library of Medicine.
This book lists a few of these contributors in the following paragraphs. Not all developers of ITK are
credited here, so please visit the Web pages at https://itk.org/ITK/project/parti.html for the names of
additional contributors, as well as checking the GIT source logs for code contributions.
The following is a brief description of the contributors to this software guide and their contributions.
Luis Ibáñez is principal author of this text. He assisted in the design and layout of the text, implemented the bulk of the LATEX and CMake build process, and was responsible for the bulk of the
content. He also developed most of the example code found in the Insight/Examples directory.
Will Schroeder helped design and establish the organization of this text and the Insight/Examples
directory. He is principal content editor, and has authored several chapters.
Lydia Ng authored the description for the registration framework and its components, the section
on the multiresolution framework, and the section on deformable registration methods. She also
edited the section on the resampling image filter and the sections on various level set segmentation
algorithms.
Joshua Cates authored the iterators chapter and the text and examples describing watershed segmentation. He also co-authored the level-set segmentation material.
Jisung Kim authored the chapter on the statistics framework.
Julien Jomier contributed the chapter on spatial objects and examples on model-based registration
using spatial objects.
Karthik Krishnan reconfigured the process for automatically generating images from all the examples. Added a large number of new examples and updated the Filtering and Segmentation chapters
vi
for the second edition.
Stephen Aylward contributed material describing spatial objects and their application.
Tessa Sundaram contributed the section on deformable registration using the finite element method.
Mark Foskey contributed the examples on the itk::AutomaticTopologyMeshSource class.
Mathieu Malaterre contributed the entire section on the description and use of DICOM readers and
writers based on the GDCM library. He also contributed an example on the use of the VTKImageIO
class.
Gavin Baker contributed the section on how to write composite filters. Also known as minipipeline
filters.
Since the software guide is generated in part from the ITK source code itself, many ITK developers
have been involved in updating and extending the ITK documentation. These include David Doria,
Bradley Lowekamp, Mark Foskey, Gaëtan Lehmann, Andreas Schuh, Tom Vercauteren, Cory
Quammen, Daniel Blezek, Paul Hughett, Matthew McCormick, Josh Cates, Arnaud Gelas,
Jim Miller, Brad King, Gabe Hart, Hans Johnson.
Hans Johnson, Kent Williams, Constantine Zakkaroff, Xiaoxiao Liu, Ali Ghayoor, and
Matthew McCormick updated the documentation for the initial ITK Version 4 release.
Luis Ibáñez and Sébastien Barré designed the original Book 1 cover. Matthew McCormick and
Brad King updated the code to produce the Book 1 cover for ITK 4 and VTK 6. Xiaoxiao Liu, Bill
Lorensen, Luis Ibáñez, and Matthew McCormick created the 3D printed anatomical objects that
were photographed by Sébastien Barré for the Book 2 cover. Steve Jordan designed the layout of
the covers.
Lisa Avila, Hans Johnson, Matthew McCormick, Sandy McKenzie, Christopher Mullins,
Katie Osterdahl, and Michka Popoff prepared the book for the 4.7 print release.
CONTENTS
I Introduction
1
1 Welcome
3
1.1
Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.2
How to Learn ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.3
Software Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.4
The Insight Community and Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
1.5
A Brief History of ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
2 Configuring and Building ITK
2.1
2.2
9
Obtaining the Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
2.1.1
Downloading Packaged Releases . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
2.1.2
Downloading From Git . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
2.1.3
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
Using CMake for Configuring and Building ITK . . . . . . . . . . . . . . . . . . . . . . . .
12
2.2.1
Preparing CMake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
2.2.2
Configuring ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14
2.2.3
Advanced Module Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
2.2.4
Static and Shared Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
2.2.5
Compiling ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
2.2.6
Installing ITK on Your System . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
viii
CONTENTS
2.3
Cross compiling ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.4
Getting Started With ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.5
Using ITK as an External Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
2.5.1
21
Hello World! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
II Architecture
3 System Overview
25
27
3.1
System Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
3.2
Essential System Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
3.2.1
Generic Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
3.2.2
Include Files and Class Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
3.2.3
Object Factories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
3.2.4
Smart Pointers and Memory Management . . . . . . . . . . . . . . . . . . . . . . .
30
3.2.5
Error Handling and Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
3.2.6
Event Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
3.2.7
Multi-Threading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
3.3
Numerics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
3.4
Data Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
3.5
Data Processing Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
3.6
Spatial Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
3.7
Wrapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
3.7.1
Python Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
Install Stable Python Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
Install Latest Python Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
Build Python Packages from Source . . . . . . . . . . . . . . . . . . . . . . . . . . .
41
4 Data Representation
4.1
43
Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
4.1.1
Creating an Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
4.1.2
Reading an Image from a File . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
4.1.3
Accessing Pixel Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
CONTENTS
4.2
4.3
4.4
4.1.4
Defining Origin and Spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
4.1.5
RGB Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
4.1.6
Vector Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
4.1.7
Importing Image Data from a Buffer . . . . . . . . . . . . . . . . . . . . . . . . . .
56
PointSet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
4.2.1
Creating a PointSet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
4.2.2
Getting Access to Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
61
4.2.3
Getting Access to Data in Points . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
4.2.4
RGB as Pixel Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
66
4.2.5
Vectors as Pixel Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
4.2.6
Normals as Pixel Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
4.3.1
Creating a Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
4.3.2
Inserting Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
4.3.3
Managing Data in Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
77
4.3.4
Customizing the Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
4.3.5
Topology and the K-Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
4.3.6
Representing a PolyLine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
4.3.7
Simplifying Mesh Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
93
4.3.8
Iterating Through Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
4.3.9
Visiting Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
98
4.3.10
More on Visiting Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.4.1
4.5
ix
Creating a PolyLineParametricPath . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Containers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5 Spatial Objects
111
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.2
Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.3
SpatialObject Tree Container . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.4
Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.5
Types of Spatial Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
x
CONTENTS
5.5.1
ArrowSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.5.2
BlobSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.5.3
CylinderSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.5.4
EllipseSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.5.5
GaussianSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.5.6
GroupSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.5.7
ImageSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.5.8
ImageMaskSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.5.9
LandmarkSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.5.10
LineSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.5.11
MeshSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.5.12
SurfaceSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.5.13
TubeSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
VesselTubeSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
DTITubeSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.6
SceneSpatialObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.7
Read/Write SpatialObjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.8
Statistics Computation via SpatialObjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6 Iterators
149
6.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.2
Programming Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.3
6.2.1
Creating Iterators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.2.2
Moving Iterators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.2.3
Accessing Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.2.4
Iteration Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Image Iterators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.3.1
ImageRegionIterator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.3.2
ImageRegionIteratorWithIndex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.3.3
ImageLinearIteratorWithIndex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.3.4
ImageSliceIteratorWithIndex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3.5
ImageRandomConstIteratorWithIndex . . . . . . . . . . . . . . . . . . . . . . . . . 166
CONTENTS
6.4
xi
Neighborhood Iterators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.4.1
NeighborhoodIterator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Basic neighborhood techniques: edge detection . . . . . . . . . . . . . . . . . . . . . 174
Convolution filtering: Sobel operator . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Optimizing iteration speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Separable convolution: Gaussian filtering . . . . . . . . . . . . . . . . . . . . . . . . 180
Slicing the neighborhood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Random access iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.4.2
ShapedNeighborhoodIterator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Shaped neighborhoods: morphological operations . . . . . . . . . . . . . . . . . . . . 187
7 Image Adaptors
191
7.1
Image Casting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.2
Adapting RGB Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
7.3
Adapting Vector Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.4
Adaptors for Simple Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.5
Adaptors and Writers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
III Development Guidelines
8 How To Write A Filter
203
205
8.1
Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.2
Overview of Filter Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
8.3
Streaming Large Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.3.1
Overview of Pipeline Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8.3.2
Details of Pipeline Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
UpdateOutputInformation() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
PropagateRequestedRegion() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
UpdateOutputData() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.4
Threaded Filter Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
8.5
Filter Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.5.1
Optional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
xii
CONTENTS
8.5.2
8.6
Useful Macros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
How To Write A Composite Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.6.1
Implementing a Composite Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.6.2
A Simple Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9 How To Create A Module
9.1
221
Name and dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.1.1
CMakeLists.txt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
9.1.2
itk-module.cmake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
9.2
Headers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
9.3
Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
9.4
Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
9.5
Wrapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
9.5.1
CMakeLists.txt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
9.5.2
Class wrap files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Wrapping Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Wrapping Macros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
9.6
9.7
Third-Party Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
9.6.1
itk-module-init.cmake
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
9.6.2
CMakeList.txt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Contributing with a Remote Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
9.7.1
Policy for Adding and Removing Remote Modules . . . . . . . . . . . . . . . . . . 239
9.7.2
Procedure for Adding a Remote Module . . . . . . . . . . . . . . . . . . . . . . . . 240
10 Software Process
243
10.1 Git Source Code Repository . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
10.2 CDash Regression Testing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
10.2.1
Developing tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
10.3 Working The Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
10.4 The Effectiveness of the Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Appendices
249
CONTENTS
xiii
A Licenses
251
A.1
Insight Toolkit License . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
A.2
Third Party Licenses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
A.2.1
DICOM Parser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
A.2.2
Double Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
A.2.3
Expat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
A.2.4
GDCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
A.2.5
GIFTI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
A.2.6
HDF5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
A.2.7
JPEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
A.2.8
KWSys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
A.2.9
MetaIO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
A.2.10 Netlib’s SLATEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
A.2.11 NIFTI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
A.2.12 NrrdIO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
A.2.13 OpenJPEG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
A.2.14 PNG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
A.2.15 TIFF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
A.2.16 VNL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
A.2.17 ZLIB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
B ITK Git Workflow
B.1
277
Git Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
B.1.1
Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Git for Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Cygwin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
B.1.2
macOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Xcode 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
OS X Installer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
MacPorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
B.1.3
B.2
Linux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
xiv
CONTENTS
B.2.1
A Primer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
B.2.2
A Topic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Published Branches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Conflicts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
B.2.3
Publish . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
Push Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
Patches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
B.2.4
Hooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Local . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
B.2.5
TipsAndTricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Editor support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Shell Customization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
C Coding Style Guide
321
C.1
Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
C.2
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
C.3
System Overview & Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
C.3.1
Kitware Style . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
C.3.2
Implementation Language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
C.3.3
Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
C.3.4
Generic Programming and the STL . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
C.3.5
Portability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
C.3.6
Multi-Layer Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
C.3.7
CMake Build Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
CONTENTS
xv
C.3.8
Doxygen Documentation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
C.3.9
vnl Math Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
C.3.10 Reference Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
C.4
Copyright . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
C.5
Citations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
C.6
Naming Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.6.1
ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.6.2
Naming Namespaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.6.3
Naming Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
C.6.4
Naming Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
Naming Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
C.6.5
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
C.6.6
Naming Methods and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
C.6.7
Naming Class Data Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
C.6.8
Naming Enums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
C.6.9
Naming Local Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Temporary Variable Naming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Variable Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Control Statement Variable Naming . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
Variable Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
C.6.10 Naming Template Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
C.6.11 Naming Typedefs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
C.6.12 Naming Constants
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
C.6.13 Using Operators to Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
C.6.14 Using Operators to Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
C.6.15 Using Underscores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
C.6.16 Include Guards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
C.6.17 Preprocessor Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
C.6.18 Header Includes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
C.6.19 Const Correctness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
C.7
Namespaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
xvi
CONTENTS
C.8
Aliasing Template Parameter Typenames . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
C.9
Pipelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
C.10 The auto Keyword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
C.11 Initialization and Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
C.12 Accessing Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
C.13 Code Layout and Indentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
C.13.1 General Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
C.13.2 Class Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
C.13.3 Method Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
C.13.4 Use of Braces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Braces in Control Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Braces in Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
C.13.5 Indentation and Tabs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
C.13.6 White Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
C.13.7 Grouping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Conditional Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Return Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
C.13.8 Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
C.13.9 Line Splitting Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
C.13.10 Empty Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
C.13.11 New Line Character . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
C.13.12 End Of File Character . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
C.14 Increment/decrement Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
C.15 Empty Arguments in Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
C.16 Ternary Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
C.17 Using Standard Macros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
C.18 Exception Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
C.18.1 Errors in Pipelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
C.19 Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
C.19.1 Messages in Macros
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
CONTENTS
xvii
C.19.2 Messages in Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
C.20 Concept Checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
C.21 Printing Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
C.22 Checking for Null . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
C.23 Writing Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
C.23.1 Code Layout in Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
C.23.2 Regressions in Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
C.23.3 Arguments in Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
C.23.4 Test Return Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
C.24 Writing Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
C.25 Doxygen Documentation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
C.25.1 General Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
C.25.2 Documenting Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
C.25.3 Documenting Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
C.25.4 Documenting Data Members . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
C.25.5 Documenting Macros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
C.25.6 Documenting Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
C.26 CMake Style . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
C.27 Documentation Style . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
LIST OF FIGURES
2.1
CMake user interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
2.2
ITK Group Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
2.3
Default ITK Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
4.1
ITK Image Geometrical Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
4.2
PointSet with Vectors as PixelType . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
5.1
SpatialObject Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.2
SpatialObject Transform Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.1
ITK image iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.2
Copying an image subregion using ImageRegionIterator . . . . . . . . . . . . . . . . . . . . 157
6.3
Using the ImageRegionIteratorWithIndex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.4
Maximum intensity projection using ImageSliceIteratorWithIndex . . . . . . . . . . . . . . . 166
6.5
Neighborhood iterator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.6
Some possible neighborhood iterator shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.7
Sobel edge detection results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.8
Gaussian blurring by convolution filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.9
Finding local minima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.10 Binary image morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
xx
List of Figures
7.1
ImageAdaptor concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.2
Image Adaptor to RGB Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.3
Image Adaptor to Vector Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
7.4
Image Adaptor for performing computations . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
8.1
Relationship between DataObjects and ProcessObjects . . . . . . . . . . . . . . . . . . . . . 206
8.2
The Data Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8.3
Sequence of the Data Pipeline updating mechanism . . . . . . . . . . . . . . . . . . . . . . . 209
8.4
Composite Filter Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
8.5
Composite Filter Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
10.1 CDash Quality Dashboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
LIST OF TABLES
2.1
ITK Compiler Support Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
6.1
ImageRandomConstIteratorWithIndex usage . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9.1
Wrapping Configuration Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
9.2
Wrapping CMake Mangling Variables for PODs . . . . . . . . . . . . . . . . . . . . . . . . . 233
9.3
Wrapping CMake Mangling Variables for other ITK pixel types. . . . . . . . . . . . . . . . . 234
9.4
Wrapping CMake Mangling Variables for Basic ITK types. . . . . . . . . . . . . . . . . . . . 235
B.1 Git DAG notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Part I
Introduction
CHAPTER
ONE
WELCOME
Welcome to the Insight Segmentation and Registration Toolkit (ITK) Software Guide. This book has
been updated for ITK 5.0.0 and later versions of the Insight Toolkit software.
ITK is an open-source, object-oriented software system for image processing, segmentation, and
registration. Although it is large and complex, ITK is designed to be easy to use once you learn
about its basic object-oriented and implementation methodology. The purpose of this Software
Guide is to help you learn just this, plus to familiarize you with the important algorithms and data
representations found throughout the toolkit.
ITK is a large system. As a result, it is not possible to completely document all ITK objects and
their methods in this text. Instead, this guide will introduce you to important system concepts and
lead you up the learning curve as fast and efficiently as possible. Once you master the basics, take
advantage of the many resources available 1 , including example materials, which provide cookbook
recipes that concisely demonstrate how to achieve a given task, the Doxygen pages, which document
the specific algorithm parameters, and the knowledge of the many ITK community members (see
Section 1.4 on page 7.)
The Insight Toolkit is an open-source software system. This means that the community surrounding ITK has a great impact on the evolution of the software. The community can make significant
contributions to ITK by providing code reviews, bug patches, feature patches, new classes, documentation, and discussions. Please feel free to contribute your ideas through the ITK community
discussion.
The Insight Toolkit is built on the principle that patents are undesirable in an open-source software.
Thus, the community strives to keep the Insight Toolkit free from any patented code, algorithm or
method.
1
https://www.itk.org/ITK/help/documentation.html
4
Chapter 1. Welcome
1.1 Organization
This software guide is divided into three parts. Part I is a general introduction to ITK, with a
description of how to install the Insight Toolkit on your computer. This includes how to build the
library from its source code. Part II introduces basic system concepts such as an overview of the
system architecture, and how to build applications in the C++ and Python programming languages.
Part II also describes the design of data structures and application of analysis methods within the
system. Part III is for the ITK contributor and explains how to create your own classes, extend the
system, and be an active participant in the project.
1.2 How to Learn ITK
The key to learning how to use ITK is to become familiar with its palette of objects and the ways to
combine them. There are three categories of documentation to help with the learning process: high
level guidance material (the Software Guide), ”cookbook” demonstrations on how to achieve concrete objectives (the examples), and detailed descriptions of the application programming interface
(the Doxygen2 documentation). These resources are combined in the three recommended stages for
learning ITK.
In the first stage, thoroughly read this introduction, which provides an overview of some of the key
concepts of the system. It also provides guidance on how to build and install the software. After
running your first ”hello world” program, you are well on your way to advanced computational
image analysis!
The next stage is to execute a few examples and gain familiarity with the available documentation. By running the examples, one can gain confidence in achieving results and is introduced the
mechanics of the software system. There are three example resources,
1. the Examples directory of the ITK source code repository 3 .
2. the Examples pages on the ITK Wiki 4
3. the Sphinx documented ITK Examples 5
To gain familiarity with the available documentation, browse the sections available in Part II and Part
III of this guide. Also, browse the Doxygen application programming interface (API) documentation
for the classes applied in the examples.
Finally, mastery of ITK involves integration of information from multiple sources. the second companion book is a reference to algorithms available, and Part III introduces how to extend them to your
2 https://itk.org/Doxygen/index.html
3
See Section Obtaining the Software on page 10)
4 https://itk.org/Wiki/ITK/Examples
5 https://itk.org/ITKExamples
注意：
1.3. Software Organization
5
needs and participate in the community. Individual examples are a detailed starting point to achieve
certain tasks. In practice, the Doxygen documentation becomes a frequent reference as an index of
the classes available, their descriptions, and the syntax and descriptions of their methods. When examples and Doxygen documentation are insufficient, the software unit tests thoroughly demonstrate
how the code is utilized. Last, but not least, the source code itself is an extremely valuable resource.
The code is the most detailed, up-to-date, and definitive description of the software. A great deal of
attention and effort is directed to the code’s readability, and its value cannot be understated.
The following sections describe how to obtain the software, summarize the software functionality in
each directory, and how to locate data.
1.3 Software Organization
To begin your ITK odyssey, you will first need to know something about ITK’s software organization
and directory structure. It is helpful to know enough to navigate through the code base to find
examples, code, and documentation.
注意：ITK多个代码资源存储的仓库
ITK resources are organized into multiple Git repositories. The ITK library source code are in the
ITK6 Git repository. The Sphinx Examples are in the ITKExamples7 repository. The sources for this
guide are in the ITKSoftwareGuide8 repository.
The ITK repository contains the following subdirectories:
• ITK/Modules — the heart of the software; the location of the majority of the source code.
• ITK/Documentation — migration guides and Doxygen infrastructure.
• ITK/Examples — a suite of simple, well-documented examples used by this guide, illustrating important ITK concepts.
• ITK/Testing — a collection of the MD5 files, which are used to link with the ITK data
servers to download test data. This test data is used by tests in ITK/Modules to produce the
ITK Quality Dashboard using CDash. (see Section 10.2 on page 244.)
• Insight/Utilities — the scripts that support source code development. For example,
CTest and Doxygen support.
• Insight/Wrapping — the wrapping code to build interfaces between the C++ library and
various interpreted languages (currently Python is supported).
The source code directory structure—found in ITK/Modules—is the most important to understand.
6 https://github.com/InsightSoftwareConsortium/ITK.git
7
https://github.com/InsightSoftwareConsortium/ITKExamples.git
8 https://github.com/InsightSoftwareConsortium/ITKSoftwareGuide.git
6
Chapter 1. Welcome
• ITK/Modules/Core — core classes, macro definitions, type aliases, and other software constructs central to ITK. The classes in Core are the only ones always compiled as part of ITK.
• ITK/Modules/ThirdParty — various third-party libraries that are used to implement image
file I/O and mathematical algorithms. (Note: ITK’s mathematical library is based on the
VXL/VNL software package9.) 注意：
• ITK/Modules/Filtering — image processing filters.
• ITK/Modules/IO — classes that support the reading and writing of images, transforms, and
geometry.
• ITK/Modules/Bridge — classes used to connect with the other analysis libraries or visualization libraries, such as OpenCV10 and VTK11 .
• ITK/Modules/Registration — classes for registration of images or other data structures to
each other.
• ITK/Modules/Segmentation — classes for segmentation of images or other data structures.
• ITK/Modules/Video — classes for input, output and processing of static and real-time data
with temporal components.
• ITK/Modules/Compatibility — collects together classes for backwards compatibility with
ITK Version 3, and classes that are deprecated – i.e. scheduled for removal from future versions of ITK.
• ITK/Modules/Remote — a group of modules distributed outside of the main ITK source
repository (most of them are hosted on github.com) whose source code can be downloaded
via CMake when configuring ITK.
• ITK/Modules/External — a directory to place in development or non-publicized modules.
• ITK/Modules/Numerics — a collection of numeric modules, including FEM, Optimization,
Statistics, Neural Networks, etc.
注意：
The Doxygen documentation is an essential resource when working with ITK, but it is not contained
in a separate repository. Each ITK class is implemented with a .h and .cxx/.hxx file (.hxx file for
templated classes). All methods found in the .h header files are documented and provide a quick
way to find documentation for a particular method. Doxygen uses this header documentation to
produce its HTML output.
The extensive Doxygen web pages describe in detail every class and method in the system. It also
contains inheritance and collaboration diagrams, listing of event invocations, and data members.
9 http://vxl.sourceforge.net
10 http://opencv.org
11
http://www.vtk.org
1.4. The Insight Community and Support
7
heavily hyper-linked to other classes and to the source code. The nightly generated Doxygen documentation is online at https://itk.org/Doxygen/html/. Archived versions for each feature
release are also available online; for example, the documentation for the 4.4.0 release are available
at https://itk.org/Doxygen44/html/.
1.4 The Insight Community and Support
Joining the community discussion is strongly recommended. This is one of the primary resources
for guidance and help regarding the use of the toolkit. You can subscribe to the community list
online at
https://discourse.itk.org/
ITK transitioned to Discourse on September 2017. Discourse is a next generation, open source
discussion platform that functions as a mailing list, discussion forum, and long-form chat room.
Discourse is a simple, modern, and fun platform that facilitates civilized discussions.
ITK maintainers developed a Getting Started Guide to help people joining the discussion, subscribing to updates, or setting their preferences.
The previous mailing list resources can be reached at https://itk.org/ITK/help/mailing.html.
ITK was created from its inception as a collaborative, community effort. Research, teaching, and
commercial uses of the toolkit are expected. If you would like to participate in the community, there
are a number of possibilities. For details on participation, see Part III of this book.
• Interaction with other community members is encouraged on the ITK discussion by both asking as answering questions. When issues are discovered, patches submitted to the code review
system are welcome. Performing code reviews, even by novice members, is encouraged. Improvements and extensions to the documentation are also welcome.
• Research partnerships with members of the Insight Software Consortium are encouraged.
Both NIH and NLM will likely provide limited funding over the next few years and will
encourage the use of ITK in proposed work.
• For those developing commercial applications with ITK, support and consulting are available
from Kitware 12 . Kitware also offers short ITK courses either at a site of your choice or
periodically at Kitware offices.
• Educators may wish to use ITK in courses. Materials are being developed for this purpose,
e.g., a one-day, conference course and semester-long graduate courses. Check the Wiki13 for
a listing.
12 http://www.kitware.com
13 https://itk.org/Wiki/ITK/Documentation
8
Chapter 1. Welcome
1.5 A Brief History of ITK
In 1999 the US National Library of Medicine of the National Institutes of Health awarded six
three-year contracts to develop an open-source registration and segmentation toolkit, that eventually came to be known as the Insight Toolkit (ITK) and formed the basis of the Insight Software
Consortium. ITK’s NIH/NLM Project Manager was Dr. Terry Yoo, who coordinated the six prime
contractors composing the Insight consortium. These consortium members included three commercial partners—GE Corporate R&D, Kitware, Inc., and MathSoft (the company name is now
Insightful)—and three academic partners—University of North Carolina (UNC), University of Tennessee (UT) (Ross Whitaker subsequently moved to University of Utah), and University of Pennsylvania (UPenn). The Principle Investigators for these partners were, respectively, Bill Lorensen
at GE CRD, Will Schroeder at Kitware, Vikram Chalana at Insightful, Stephen Aylward with Luis
Ibáñez at UNC (Luis is now at Google), Ross Whitaker with Josh Cates at UT (both now at Utah),
and Dimitri Metaxas at UPenn (now at Rutgers). In addition, several subcontractors rounded out the
consortium including Peter Raitu at Brigham & Women’s Hospital, Celina Imielinska and Pat Molholt at Columbia University, Jim Gee at UPenn’s Grasp Lab, and George Stetten at the University of
Pittsburgh.
In 2002 the first official public release of ITK was made available. In addition, the National Library
of Medicine awarded thirteen contracts to several organizations to extend ITK’s capabilities. The
NLM has funded maintenance of the toolkit over the years, and a major funding effort was started in
July 2010 that culminated with the release of ITK 4.0.0 in December 2011. If you are interested in
potential funding opportunities, we suggest that you contact Dr. Terry Yoo at the National Library
of Medicine for more information.
CHAPTER
TWO
CONFIGURING AND BUILDING ITK
This chapter describes the process for configuring and compiling ITK on your system. Keep in
mind that ITK is a toolkit, and as such, once it is installed on your computer it does not provide an
application to run. What ITK does provide is a large set of libraries which can be used to create
your own applications. Besides the toolkit proper, ITK also includes an extensive set of examples
and tests that introduce ITK concepts and show how to use ITK in your own projects.
Some of the examples distributed with ITK depend on third party libraries, some of which may need
to be installed separately. For the initial build of ITK, you may want to ignore these extra libraries
and just compile the toolkit itself.
ITK has been developed and tested across different combinations of operating systems, compilers, and hardware platforms including Microsoft Windows, Linux on various architectures, UNIX,
macOS, and Cygwin. Dedicated community members and Kitware are committed to providing longterm support of the most prevalent development environments (Visual Studio, macOS, and Linux)
for building ITK:
Compiler variants will be supported for the duration that the associated operating system vendors
commit to in their long-term stable platforms. For example the gcc compilers supported will mirror
the compiler support in the RedHat lifecycle, the apple clang compilers will mirror the support lifecycle of the compiler by Apple, and the Visual Studio series support will follow lifecycle deprecation
of the compiler versions.
For example as of 2018 the following time schedule is expected for supporting these compiler environments.
• GCC
– 4.8.2 (From 2015-until 2020)
• Visual Studio
– 2010 [v10.0] (From 2010 - until 2020)
– 2012 [v11.0] (From 2012 - until 2022)
10
Chapter 2. Configuring and Building ITK
– 2013 [v12.0] (From 2013 - until 2023)
• Apple Clang
– Apple clang-600.0.56 (From 2013 - until 2019)
– Apple LLVM version 8.1.0 (clang-802.0.42) (From 2016 - until 2021)
• Clang
– 3.3 (From 2013 - until 2018)
Table 2.1 prints the compiler support timeline in ITK at the time of writing this guide.
Compiler
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
Visual Studio 10
GCC 4.2
GCC 4.4
GCC 4.9
Table 2.1: ITK Compiler Support Timeline.
Legend:
Fully supported
Phase out
If community supported
If you are currently using an outdated compiler this may be an excellent excuse for upgrading this
old piece of software! Support for different platforms is evident on the ITK quality dashboard (see
Section 10.2 on page 244).
2.1 Obtaining the Software
There are two different ways to access the ITK source code:
Periodic releases Official releases are available on the ITK web site1 . They are released twice a
year, and announced on the ITK web pages and discussion. However, they may not provide
the latest and greatest features of the toolkit.
Continuous repository checkout Direct access to the Git source code repository2 provides immediate availability to the latest toolkit additions. But, on any given day the source code may not
be stable as compared to the official releases.
1 https://itk.org/ITK/resources/software.html
2 https://itk.org/ITK.git
2.1. Obtaining the Software
11
This software guide assumes that you are using the current released version of ITK, available on the
ITK web site. If you are a new user, we recommend the released version of the software. It is more
consistent than the code available from the Git repository (see Section 2.1.2). When working from
the repository, please be aware of the ITK quality testing dashboard. The Insight Toolkit is heavily
tested using the open-source CDash regression testing system3 . Before updating the repository,
make sure that the dashboard is green, indicating stable code. (Learn more about the ITK dashboard
and quality assurance process in Section 10.2 on page 244.)
2.1.1 Downloading Packaged Releases
ITK can be downloaded without cost from the following web site:
https://www.itk.org/ITK/resources/software.html
On the web page, choose the tarball that better fits your system. The options are .zip and .tar.gz
files. The first type is better suited for Microsoft-Windows, while the second one is the preferred
format for UNIX systems.
Once you unzip or untar the file a directory called InsightToolkit-5.0.0 will be created in your
disk and you will be ready to start the configuration process described in Section 2.2.1 on page 12.
2.1.2 Downloading From Git
Git is a free and open source distributed version control system. For more information about Git
please see Section 10.1 on page 243. (Note: please make sure that you access the software via Git
only when the ITK quality dashboard indicates that the code is stable.)
Access ITK via Git using the following commands (under a Git Bash shell):
git clone git://itk.org/ITK.git
This will trigger the download of the software into a directory named ITK. Any time you want to
update your version, it will be enough to change into this directory, ITK, and type:
git pull
Once you obtain the software you are ready to configure and compile it (see Section 2.2.1 on page
12). First, however, we recommend reading the following sections that describe the organization of
the software and joining the discussion.
3 http://open.cdash.org/index.php?project=Insight
12
Chapter 2. Configuring and Building ITK
2.1.3 Data
The Insight Toolkit was designed to support the Visible Human Project and its associated data.
This data is available from the National Library of Medicine at
http://www.nlm.nih.gov/research/visible/visible_human.html.
Another source of data can be obtained from the ITK Web site at either of the following:
https://www.itk.org/ITK/resources/links.html
ftp://public.kitware.com/pub/itk/Data/.
2.2 Using CMake for Configuring and Building ITK
The challenge of supporting ITK across platforms has been solved through the use of CMake4 , a
cross-platform, open-source build system. CMake controls the software compilation process with
simple platform and compiler-independent configuration files. CMake is quite sophisticated—it
supports complex environments requiring system introspection, compiler feature testing, and code
generation.
CMake generates native Makefiles or workspaces to be used with the corresponding development
environment of your choice. For example, on UNIX and Cygwin systems, CMake generates Makefiles; under Microsoft Windows CMake generates Visual Studio workspaces; CMake is also capable
of generating appropriate build files for other development environments, e.g., Eclipse. The information used by CMake is provided in CMakeLists.txt files that are present in every directory of
the ITK source tree. Along with the specification of project structure and code dependencies these
files specify the information that need to be provided to CMake by the user during project configuration stage. Typical configuration options specified by the user include paths to utilities installed
on your system and selection of software features to be included.
An ITK build requires only CMake and a C++ compiler. ITK ships with all the third party library
dependencies required, and these dependencies are used during compilation unless the use of a
system version is requested during CMake configuration.
2.2.1 Preparing CMake
CMake can be downloaded at no cost from
https://cmake.org/download/
You can download binary versions for most of the popular platforms including Microsoft Windows,
macOS, Linux, PowerPC and IRIX. Alternatively you can download the source code and build
4 www.cmake.org
2.2. Using CMake for Configuring and Building ITK
13
CMake on your system. Follow the instructions provided on the CMake web page for downloading
and installing the software. The minimum version of CMake has been evolving along with the version of ITK. For example, the current version of ITK (5.0.0) requires the minimum CMake version
to be 3.9.5.
CMake provides a terminal-based interface (Figure 2.1) on platforms support the curses library.
For most platforms CMake also provides a GUI based on the Qt library. Figure 2.1 shows the
terminal-based CMake interface for Linux and CMake GUI for Microsoft Windows.
Figure 2.1: CMake user interfaces: at the top is the interface based on the curses library supported by
UNIX/Linux systems, below is the Microsoft Windows version of the CMake GUI based on the Qt library (CMake
GUI is also available on UNIX/Linux systems).
14
Chapter 2. Configuring and Building ITK
Running CMake to configure and prepare for compilation a new project initially requires two pieces
of information: where the source code directory is located, and where the compiled code is to be
produced. These are referred to as the source directory and the binary directory respectively. We
recommend setting the binary directory to be different than the source directory in order to produce
an out-of-source build.
If you choose to use the terminal-based version of CMake (ccmake) the binary directory needs to
be created first and then CMake is invoked from the binary directory with the path to the source
directory. For example:
mkdir ITK-build
cd ITK-build
ccmake ../ITK
In the GUI version of CMake (cmake-gui) the source and binary directories are specified in the
appropriate input fields (Figure 2.1) and the application will request a confirmation to create a new
binary directory if it does not exist.
CMake runs in an interactive mode which allows iterative selection of options followed by configuration according to the updated options. This iterative process proceeds until no more options
remain to be specified. At this point, a generation step produces the appropriate build files for your
configuration.
This interactive configuration process can be better understood by imagining the traversal of a path
in a decision tree. Every selected option introduces the possibility that new, dependent options may
become relevant. These new options are presented by CMake at the top of the options list in its
interface. Only when no new options appear after a configuration iteration can you be sure that
the necessary decisions have all been made. At this point build files are generated for the current
configuration.
2.2.2 Configuring ITK
Start terminal-based CMake interface ccmake on Linux and UNIX, or the graphical user interface
cmake-gui on Microsoft Windows. Remember to run ccmake from the binary directory on Linux
and UNIX. On Windows, specify the source and binary directories in the GUI, then set and modify
the configuration and build option in the interface as necessary.
The examples distributed with the toolkit provide a helpful resource for learning how to use ITK
components but are not essential for compiling the toolkit itself. The testing section of the source
tree includes a large number of small programs that exercise the capabilities of ITK classes. Enabling
the compilation of the examples and unit tests will considerably increase the build time. In order to
speed up the build process, you can disable the compilation of the unit tests and examples. This is
done by setting the variables BUILD TESTING and BUILD EXAMPLES to OFF.
Most CMake variables in ITK have sensible default values. Each time a CMake variable is changed,
it is necessary to re-run the configuration step. In the terminal-based version of the interface the
2.2. Using CMake for Configuring and Building ITK
15
configuration step is triggered by hitting the “c” key. In the GUI version this is done by clicking on
the “Configure” button.
When no new options appear highlighted in CMake, you can proceed to generate Makefiles, a Visual
Studio workspace, or other appropriate build files depending on your preferred development environment. This is done in the GUI interface by clicking on the “Generate” button. In the terminal-based
version this is done by hitting the “g” key. After the generation process the terminal-based version
of CMake will quit silently. The GUI window of CMake can be left open for further refinement of
configuration options as described in the next section. With this scenario it is important to generate
new build files to reflect the latest configuration changes. In addition, the new build files need to be
reloaded if the project is open in the integrated development environment such as Visual Studio or
Eclipse.
2.2.3 Advanced Module Configuration
注意：编译中可以选择的是否要编译
的模块
Following the default configuration introduced in 2.2.2, the majority of the toolkit will be built. The
modern modular structure of the toolkit makes it possible to customize the ITK library by choosing
which modules to include in the build. ITK was officially modularized in version 4.0.0 released in
December of 2011. Developers have been testing and improving the modular structure since then.
The toolkit currently contains more than 100 regular/internal modules and many remote modules,
while new ITK modules are being developed.
ITK BUILD DEFAULT MODULES is the CMake option to build all default modules in the toolkit,
by default this option is ON as shown in Figure 2.1. The default modules include most internal
ITK modules except the ones that depend on external third party libraries (such as ITKVtkGlue,
ITKVideoBridgeOpenCV, ITKVideoBridgeVXL, etc.) and several modules containing legacy code
(ITKReview, ITKDeprecated and ITKV3Compatibility).
Apart from the default mode of selecting the modules for building the ITK library there are two
other approaches module selection: the group mode, and the advanced module mode. When ITK BUILD DEFAULT MODULES is set to OFF, the selection of modules to be included in the ITK library
can be customized by changing the variables enabling group and advanced module selection.
ITKGroup {group name} variables for group module selection are visible when ITK BUILD DEFAULT MODULES is OFF. The ITK source code tree is organized in such way that a group of modules characterised by close relationships or similar functionalities stay in one subdirectory. Currently
there are 11 groups (excluding the External and Remote groups). The CMake ITKGroup {group
name} options are created for the convenient enabling or disabling of multiple modules at once. The
ITKGroup Core group is selected by default as shown in Figure 2.2. When a group is selected, all
modules in the group and their depending modules are enabled. When a group variable is set to OFF,
all modules in the group, except the ones that are required by other enabled modules, are disabled.
If you are not sure about which groups to turn on, but you do have a list of specific modules to
be included in your ITK library, you can certainly skip the Group options and use the Module {module name} options only. Whatever modules you select, their dependent modules are automat-
16
Chapter 2. Configuring and Building ITK
Figure 2.2: CMake GUI shows the ITK Group options.
ically enabled. In the advanced mode of the CMake GUI, you can manually toggle the build of the
non-default modules via the Module {module name} variables. In Figure 2.3 all default modules’
Module {module name} variables are shown disabled for toggling since they are enabled via the
ITK BUILD DEFAULT MODULES set to ON variable.
Figure 2.3: CMake GUI for configuring ITK: the advanced mode shows options for non-default ITK Modules.
However, not all modules will be visible in the CMake GUI at all times due to the various levels
of controls in the previous two modes. If some modules are already enabled by other modes, these
modules are set as internal variables and are hidden in the CMake GUI. For example, Module ITKFoo variable is hidden when the module ITKFoo is enabled in either of the following scenarios:
2.2. Using CMake for Configuring and Building ITK
17
1. module ITKBar is enabled and depends on ITKFoo,
2. ITKFoo belongs to the group ITKGroup FooAndBar and the group is enabled
3. ITK BUILD DEFAULT MODULES is ON and ITKFoo is a default module.
To find out why a particular module is enabled, check the CMake configuration messages where the
information about enabling or disabling the modules is displayed (Figure 2.3); these messages are
sorted in alphabetical order by module names.
Those who prefer to build ITK using the command line are referred to the online cmake commandline tool documentation5. Only some typical use cases are shown here for reference.
• Example 1: Build all default modules.
cmake [-DITK_BUILD_DEFAULT_MODULES:BOOL=ON]
../ITK-build
As ITK BUILD DEFAULT MODULES is ON by default, the above can also be accomplished by
cmake ../ITK-build
• Example 2: Enable specific group(s) of modules.
cmake -DITK_BUILD_DEFAULT_MODULES:BOOL=OFF
-DBUILD_EXAMPLES:BOOL=OFF
-DITKGroup_{Group1}:BOOL=ON
[-DITKGroup_{Group2}:BOOL=ON]
../ITK-build
where ITKGroup GroupN could be, for example, ITKGroup Filtering or ITKGroup Registration for the Filtering and Registration groups, respectively.
• Example 3: Enable specific modules.
cmake -DITK_BUILD_DEFAULT_MODULES:BOOL=OFF
-DBUILD_EXAMPLES:BOOL=OFF
-DModule_{Module1}:BOOL=ON
[-DModule_{Module2}:BOOL=ON]
../ITK-build
where Module Module1 could be, for example, Module ITKFEM for the non-default, built-in
FEM module, or Module Cuberille for the Cuberille remote module.
• Example 4: Enable examples.
5 https://cmake.org/cmake/help/latest/manual/cmake.1.html
18
Chapter 2. Configuring and Building ITK
cmake -DITK_BUILD_DEFAULT_MODULES:BOOL=ON
-DBUILD_EXAMPLES:BOOL=ON
../ITK-build
Note that BUILD EXAMPLES is OFF by default, and BUILD EXAMPLES=ON requires ITK BUILD DEFAULT MODULES=ON.
2.2.4 Static and Shared Libraries
ITK libraries can be built as static libraries, i.e. files whose functions and variables are included in
a binary during the link phase of the build cycle. Alternatively, ITK libraries can be built as shared
libraries, where libraries are dynamically linked to a binary. In this case, functions and variables are
shared at runtime according to their symbols.
By enabling the standard CMake configuration variable, BUILD SHARED LIBS, ITK modules with
the ENABLE SHARED option (see Section 9.1) will be built as shared libraries.
Static libraries are preferred when creating a stand-alone executable. An application can be distributed as a single file when statically linked. Additional effort is not required to package library
dependencies, configure the system to find library dependencies at runtime, or define symbol export
specifications. However, care should be taken to only link static libraries once into the binaries used
by an application. Failure to due so can result in duplicated global variables and, consequently,
undefined or undesirable behavior.
Shared libraries should be used when ITK is linked to more than one binary in an application. This
reduces binary size and ensures that singleton variables are unique across the application.
An advanced CMake configuration variable, ITK TEMPLATE VISIBILITY DEFAULT defines the
symbol visibility attribute on template classes to default on systems that require it to perform
dynamic cast’s on pointers passed across binaries. The default value can be disabled only when it
is known that template classes are not implicitly instantiated and passed across binaries.
2.2.5 Compiling ITK
To initiate the build process after generating the build files on Linux or UNIX, simply type make
in the terminal if the current directory is set to the ITK binary directory. If using Visual Studio,
first load the workspace named ITK.sln from the binary directory specified in the CMake GUI and
then start the build by selecting “Build Solution” from the “Build” menu or right-clicking on the
ALL BUILD target in the Solution Explorer pane and selecting the “Build” context menu item.
The build process can take anywhere from 15 minutes to a couple of hours, depending on the build
configuration and the performance of your system. If testing is enabled as part of the normal build
process, about 2400 test programs will be compiled. In this case, you will then need to run ctest to
verify that all the components of ITK have been correctly built on your system.
2.3. Cross compiling ITK
19
2.2.6 Installing ITK on Your System
When the build process is complete an ITK binary distribution package can be generated for installation on your system or on a system with compatible specifications (such as hardware platform and
operating system) as well as suitable development environment components (such as C++ compiler
and CMake). The default prefix for installation destination directory needs to be specified during
CMake configuration process prior to compiling ITK. The installation destination prefix can to be
set through the CMake cache variable CMAKE INSTALL PREFIX.
Typically distribution packages are generated to provide a “clean” form of the software which is
isolated from the details of the build process (separate from the source and build trees). Due to
the intended use of ITK as a toolkit for software development the step of generating ITK binary
packages for installing ITK on other systems has limited application and thus it can be treated as
optional. However, the step for generating binary distribution packages has a much wide application
for distributing software developed with ITK. Further details on configuring and generating binary
packages with CMake can be found in the CMake tutorial6 .
2.3 Cross compiling ITK
This section describes the procedure to follow to cross compile ITK for another system. Cross
compiling involves a build system, the system where the executables are built, and the target system,
the system where the executables are intended to run.
Currently, the best way to cross-compile ITK is to use dockcross.
For example, the commands to build for Linux-ARMv7 are:
git clone https://github.com/InsightSoftwareConsortium/ITK
docker run --rm dockcross/linux-armv7 > ./dockcross-linux-armv7
chmod +x ./dockcross-linux-armv7
mkdir ITK-build
./dockcross-linux-armv7 cmake -BITK-build -HITK -GNinja
./dockcross-linux-armv7 ninja -CITK-build
2.4 Getting Started With ITK
The simplest way to create a new project with ITK is to create two new directories somewhere in
your disk, one to hold the source code and one to hold the binaries and other files that are created
in the build process. For this example, create a HelloWorldITK directory to hold the source and a
HelloWorldITK-build directory to hold the binaries. The first file to place in the source directory
is a CMakeLists.txt file that will be used by CMake to generate a Makefile (if you are using Linux
6 https://cmake.org/cmake-tutorial/
20
Chapter 2. Configuring and Building ITK
or UNIX) or a Visual Studio workspace (if you are using Microsoft Windows). The second source
file to be created is an actual C++ program that will exercise some of the large number of classes
available in ITK. The details of these files are described in the following section.
Once both files are in your directory you can run CMake in order to configure your project. Under UNIX/Linux, you can cd to your newly created binary directory and launch the terminal-based
version of CMake by entering “ccmake ../HelloWorldITK” in the terminal. Note the “../HelloWorldITK” in the command line to indicate that the CMakeLists.txt file is up one directory and
in HelloWorldITK. In CMake GUI which can be used under Microsoft Windows and UNIX/Linux,
the source and binary directories will have to be specified prior to the configuration and build file
generation process.
Both the terminal-based and GUI versions of CMake will require you to specify the directory where
ITK was built in the CMake variable ITK DIR. The ITK binary directory will contain a file named
ITKConfig.cmake generated during ITK configuration process with CMake. From this file, CMake
will recover all information required to configure your new ITK project.
After generating the build files, on UNIX/Linux systems the project can be compiled by typing
make in the terminal provided the current directory is set to the project’s binary directory. In
Visual Studio on Microsoft Windows the project can be built by loading the workspace named
HelloWorldITK.sln from the binary directory specified in the CMake GUI and selecting “Build
Solution” from the “Build” menu or by right-clicking on the ALL BUILD target in the Solution Explorer pane and selecting the “Build” context menu item.
The resulting executable, which will be called HelloWorld, can be executed on the command line.
If on Microsoft Windows, please note that double-clicking on the icon of the executable will quickly
launch a command line window, run the executable and close the window right away, not giving you
time to see the output. It is therefore preferable to run the executable from the DOS command line
by starting the cmd.exe shell first.
2.5 Using ITK as an External Library
注意：itk用作第三方库，CmakeList
设置
For a project that uses ITK as an external library, it is recommended to specify the individual ITK
modules in the COMPONENTS argument in the find package CMake command:
find_package(ITK REQUIRED COMPONENTS Module1 Module2)
include(\${ITK_USE_FILE})
e.g.
find_package(ITK REQUIRED
COMPONENTS
MorphologicalContourInterpolation
ITKSmoothing
ITKIOImageBase
2.5. Using ITK as an External Library
21
ITKIONRRD
)
include(\${ITK_USE_FILE})
If you would like to use the CMake ExternalProject Module7 to download ITK source code when
building your ITK application (a.k.a. Superbuild ITK), here is a basic CMake snippet for setting up
a Superbuild in an ITK application project using CMake:
ExternalProject_Add(ITK
GIT_REPOSITORY \${git_protocol}://github.com/InsightSoftwareConsortium/ITK.git"
GIT_TAG "" # specify the commit id or the tag id
SOURCE_DIR
BINARY_DIR
CMAKE_GENERATOR ${gen}
CMAKE_ARGS
${ep_common_args}
-DBUILD_SHARED_LIBS:BOOL=OFF
-DBUILD_EXAMPLES:BOOL=OFF
-DBUILD_TESTING:BOOL=OFF
-DITK_BUILD_DEFAULT_MODULES:BOOL=ON
[-DModule_LevelSetv4Visualization:BOOL=ON]
INSTALL_COMMAND ""
DEPENDS
[VTK] [DCMTK] # if some of the modules requested require extra third party libraries
)
More exemplary configurations for superbuild ITK projects can be found in: Slicer8 , BrainsTools9 ,
ITK Wiki Examples10, ITK Sphinx Examples11, and ITK Software Guide12 .
2.5.1 Hello World!
This section provides and explains the contents of the two files which need to be created for your
new project. These two files can be found in the ITK/Examples/Installation directory.
The CMakeLists.txt file contains the following lines:
project(HelloWorld)
find_package(ITK REQUIRED)
include(${ITK_USE_FILE})
7 https://cmake.org/cmake/help/latest/module/ExternalProject.html
8
https://github.com/Slicer/Slicer
9 https://github.com/BRAINSia/BRAINSTools
10 https://github.com/InsightSoftwareConsortium/ITKWikiExamples
11 https://github.com/InsightSoftwareConsortium/ITKExamples
12 https://github.com/InsightSoftwareConsortium/ITKSoftwareGuide
22
Chapter 2. Configuring and Building ITK
add_executable(HelloWorld HelloWorld.cxx)
target_link_libraries(HelloWorld ${ITK_LIBRARIES})
The first line defines the name of your project as it appears in Visual Studio or Eclipse; this line will
have no effect with UNIX/Linux Makefiles. The second line loads a CMake file with a predefined
strategy for finding ITK. If the strategy for finding ITK fails, CMake will report an error which
can be corrected by providing the location of the directory where ITK was compiled or installed on
your system. In this case the path to the ITK’s binary/installation directory needs to be specified
as the value of the ITK DIR CMake variable. The line include(${USE ITK FILE}) loads the
UseITK.cmake file which contains the configuration information about the specified ITK build. The
line starting with add executable call defines as its first argument the name of the executable
that will be produced as result of this project. The remaining argument(s) of add executable are
the names of the source files to be compiled. Finally, the target link libraries call specifies
which ITK libraries will be linked against this project. Further details on creating and configuring
CMake projects can be found in the CMake tutorial13 and CMake online documentation14.
The source code for this section can be found in the file
HelloWorld.cxx.
The following code is an implementation of a small ITK program. It tests including header files and
linking with ITK libraries.
#include "itkImage.h"
#include
int main()
{
using ImageType = itk::Image< unsigned short, 3 >;
ImageType::Pointer image = ImageType::New();
std::cout << "ITK Hello World !" << std::endl;
return EXIT_SUCCESS;
}
This code instantiates a 3D image15 whose pixels are represented with type unsigned short. The
image is then constructed and assigned to a itk::SmartPointer. Although later in the text we
will discuss SmartPointers in detail, for now think of it as a handle on an instance of an object (see
section 3.2.4 for more information). The itk::Image class will be described in Section 4.1.
By this point you have successfully configured and compiled ITK, and created your first simple
13
https://cmake.org/cmake-tutorial/
14 https://cmake.org/documentation/
15 Also
known as a volume.
2.5. Using ITK as an External Library
23
program! If you have experienced any difficulties while following the instructions provided in this
section, please join the community discussion (see Section 1.4 on page 7) and post questions there.
Part II
Architecture
CHAPTER
THREE
SYSTEM OVERVIEW
The purpose of this chapter is to provide you with an overview of the Insight Toolkit system. We
recommend that you read this chapter to gain an appreciation for the breadth and area of application
of ITK.
3.1 System Organization
The Insight Toolkit consists of several subsystems. A brief description of these subsystems follows.
Later sections in this chapter—and in some cases additional chapters—cover these concepts in more
detail.
Essential System Concepts. Like any software system, ITK is built around some core design concepts. Some of the more important concepts include generic programming, smart pointers for
memory management, object factories for adaptable object instantiation, event management
using the command/observer design paradigm, and multi-threading support.
Numerics. ITK uses VXL’s VNL numerics libraries. These are easy-to-use C++ wrappers around
the Netlib Fortran numerical analysis routines 1 .
Data Representation and Access. Two principal classes are used to represent data: the
itk::Image and itk::Mesh classes. In addition, various types of iterators and containers are used to hold and traverse the data. Other important but less popular classes are also
used to represent data such as itk::Histogram and itk::SpatialObject.
Data Processing Pipeline. The data representation classes (known as data objects) are operated on
by filters that in turn may be organized into data flow pipelines. These pipelines maintain
state and therefore execute only when necessary. They also support multi-threading, and are
streaming capable (i.e., can operate on pieces of data to minimize the memory footprint).
1 http://www.netlib.org
注意：
28
Chapter 3. System Overview
IO Framework. Associated with the data processing pipeline are sources, filters that initiate the
pipeline, and mappers, filters that terminate the pipeline. The standard examples of sources
and mappers are readers and writers respectively. Readers input data (typically from a file),
and writers output data from the pipeline.
Spatial Objects. Geometric shapes are represented in ITK using the spatial object hierarchy. These
classes are intended to support modeling of anatomical structures. Using a common basic
interface, the spatial objects are capable of representing regions of space in a variety of different ways. For example: mesh structures, image masks, and implicit equations may be used
as the underlying representation scheme. Spatial objects are a natural data structure for communicating the results of segmentation methods and for introducing anatomical priors in both
segmentation and registration methods.
Registration Framework. A flexible framework for registration supports four different types of
registration: image registration, multiresolution registration, PDE-based registration, and
FEM (finite element method) registration.
FEM Framework. ITK includes a subsystem for solving general FEM problems, in particular nonrigid registration. The FEM package includes mesh definition (nodes and elements), loads,
and boundary conditions.
Level Set Framework. The level set framework is a set of classes for creating filters to solve partial
differential equations on images using an iterative, finite difference update scheme. The level
set framework consists of finite difference solvers including a sparse level set solver, a generic
level set segmentation filter, and several specific subclasses including threshold, Canny, and
Laplacian based methods.
Wrapping. ITK uses a unique, powerful system for producing interfaces (i.e., “wrappers”) to interpreted languages such as Python. The CastXML2 tool is used to produce an XML description
of arbitrarily complex C++ code. An interface generator script is then used to transform the
XML description into wrappers using the SWIG3 package.
3.2 Essential System Concepts
This section describes some of the core concepts and implementation features found in ITK.
3.2.1 Generic Programming
Generic programming is a method of organizing libraries consisting of generic—or reusable—
software components [8]. The idea is to make software that is capable of “plugging together” in
2 https://github.com/CastXML/CastXML
3 http://www.swig.org/
3.2. Essential System Concepts
29
an efficient, adaptable manner. The essential ideas of generic programming are containers to hold
data, iterators to access the data, and generic algorithms that use containers and iterators to create
efficient, fundamental algorithms such as sorting. Generic programming is implemented in C++
with the template programming mechanism and the use of the STL Standard Template Library [1].
C++ templating is a programming technique allowing users to write software in terms of one or
more unknown types T. To create executable code, the user of the software must specify all types T
(known as template instantiation) and successfully process the code with the compiler. The T may
be a native type such as float or int, or T may be a user-defined type (e.g., a class). At compiletime, the compiler makes sure that the templated types are compatible with the instantiated code and
that the types are supported by the necessary methods and operators.
ITK uses the techniques of generic programming in its implementation. The advantage of this
approach is that an almost unlimited variety of data types are supported simply by defining the
appropriate template types. For example, in ITK it is possible to create images consisting of almost
any type of pixel. In addition, the type resolution is performed at compile time, so the compiler
can optimize the code to deliver maximal performance. The disadvantage of generic programming
is that the analysis performed at compile time increases the time to build an application. Also, the
increased complexity may produce difficult to decipher error messages due to even the simplest
syntax errors. For those unfamiliar with templated code and generic programming, we recommend
the two books cited above.
3.2.2 Include Files and Class Definitions
In ITK, classes are defined by a maximum of two files: a header file (.h) and an implementation file
(.cxx) if defining a non-templated class, and a .hxx file if defining a templated class. The header
files contain class declarations and formatted comments that are used by the Doxygen documentation
system to automatically produce HTML manual pages.
In addition to class headers, there are a few other important header files.
注意：基础操作的宏定义
itkMacro.h is found in the Modules/Core/Common/include directory and defines standard
system-wide macros (such as Set/Get, constants, and other parameters).
itkNumericTraits.h is found in the Modules/Core/Common/include directory and defines
numeric characteristics for native types such as its maximum and minimum possible values.
3.2.3 Object Factories
注意：itk中大部分类都是通过对象工
厂机制进行实例化的
Most classes in ITK are instantiated through an object factory mechanism. That is, rather than using
the standard C++ class constructor and destructor, instances of an ITK class are created with the
static class New() method. In fact, the constructor and destructor are protected: so it is generally
not possible to construct an ITK instance on the stack. (Note: this behavior pertains to classes
that are derived from itk::LightObject. In some cases the need for speed or reduced memory
注意：继承至itk::LightObject的派
生ITK类不能在stack上构建ITK实例
30
Chapter 3. System Overview
footprint dictates that a class is not derived from LightObject. In this case instances may be created
on the stack. An example of such a class is the itk::EventObject.)
The object factory enables users to control run-time instantiation of classes by registering one or
more factories with itk::ObjectFactoryBase. These registered factories support the method
CreateInstance(classname) which takes as input the name of a class to create. The factory can
choose to create the class based on a number of factors including the computer system configuration
and environment variables. For example, a particular application may wish to deploy its own class
implemented using specialized image processing hardware (i.e., to realize a performance gain). By
using the object factory mechanism, it is possible to replace the creation of a particular ITK filter at
run-time with such a custom class. (Of course, the class must provide the exact same API as the one
it is replacing.). For this, the user compiles his class (using the same compiler, build options, etc.)
and inserts the object code into a shared library or DLL. The library is then placed in a directory
referred to by the ITK AUTOLOAD PATH environment variable. On instantiation, the object factory
will locate the library, determine that it can create a class of a particular name with the factory, and
use the factory to create the instance. (Note: if the CreateInstance() method cannot find a factory
that can create the named class, then the instantiation of the class falls back to the usual constructor.)
In practice, object factories are used mainly (and generally transparently) by the ITK input/output
(IO) classes. For most users the greatest impact is on the use of the New() method to create a class.
Generally the New() method is declared and implemented via the macro itkNewMacro() found in
Modules/Core/Common/include/itkMacro.h.
3.2.4 Smart Pointers and Memory Management
By their nature, object-oriented systems represent and operate on data through a variety of object
types, or classes. When a particular class is instantiated, memory allocation occurs so that the instance can store data attribute values and method pointers (i.e., the vtable). This object may then
be referenced by other classes or data structures during normal operation of the program. Typically,
during program execution, all references to the instance may disappear at which point the instance
must be deleted to recover memory resources. Knowing when to delete an instance, however, is
difficult. Deleting the instance too soon results in program crashes; deleting it too late causes memory leaks (or excessive memory consumption). This process of allocating and releasing memory is
known as memory management.
注意：ITK运用的是引用计数
In ITK, memory management is implemented through reference counting. This compares to another
popular approach—garbage collection—used by many systems, including Java. In reference counting, a count of the number of references to each instance is kept. When the reference goes to zero,
the object destroys itself. In garbage collection, a background process sweeps the system identifying
instances no longer referenced in the system and deletes them. The problem with garbage collection
is that the actual point in time at which memory is deleted is variable. This is unacceptable when
an object size may be gigantic (think of a large 3D volume gigabytes in size). Reference counting
deletes memory immediately (once all references to an object disappear).
Reference counting is implemented through a Register()/Delete() member function interface.
注意：为了性能，使
用自定义的类将ITK中
的类进行替换；使用
的对象工厂机制
注意：创建对象实
例化的过程：类与
工厂对象绑定--由
工厂创建实例化对
象--再解除绑定
3.2. Essential System Concepts
31
All instances of an ITK object have a Register() method invoked on them by any other object
that references them. The Register() method increments the instances’ reference count. When the
reference to the instance disappears, a Delete() method is invoked on the instance that decrements
the reference count—this is equivalent to an UnRegister() method. When the reference count
returns to zero, the instance is destroyed.
This protocol is greatly simplified by using a helper class called a itk::SmartPointer. The smart
pointer acts like a regular pointer (e.g. supports operators -> and *) but automagically performs a
Register() when referring to an instance, and an UnRegister() when it no longer points to the
instance. Unlike most other instances in ITK, SmartPointers can be allocated on the program stack,
and are automatically deleted when the scope that the SmartPointer was created in is closed. As a
result, you should rarely if ever call Register() or Delete() in ITK. For example:
MyRegistrationFunction()
{ /* <----- Start of scope */
// here an interpolator is created and associated to the
// "interp" SmartPointer.
InterpolatorType::Pointer interp = InterpolatorType::New();
} /* <------ End of scope */
In this example, reference counted objects are created (with the New() method) with a reference
count of one. Assignment to the SmartPointer interp does not change the reference count. At the
end of scope, interp is destroyed, the reference count of the actual interpolator object (referred to
by interp) is decremented, and if it reaches zero, then the interpolator is also destroyed.
Note that in ITK SmartPointers are always used to refer to instances of classes derived from
itk::LightObject. Method invocations and function calls often return “real” pointers to instances,
but they are immediately assigned to a SmartPointer. Raw pointers are used for non-LightObject
classes when the need for speed and/or memory demands a smaller, faster class. Raw pointers are
注意：特殊需求时才会使用raw
preferred for multi-threaded sections of code.
pointers
3.2.5 Error Handling and Exceptions
In general, ITK uses exception handling to manage errors during program execution. Exception
handling is a standard part of the C++ language and generally takes the form as illustrated below:
try
{
//...try executing some code here...
}
catch ( itk::ExceptionObject & exp )
{
//...if an exception is thrown catch it here
}
注意：超出创建
SmartPointer的作
用域，其将自动删
除（即reference
count - 1）
32
Chapter 3. System Overview
A particular class may throw an exception as demonstrated below (this code snippet is taken from
itk::ByteSwapper:
switch ( sizeof(T) )
{
//non-error cases go here followed by error case
default:
ByteSwapperError e(__FILE__, __LINE__);
e.SetLocation("SwapBE");
e.SetDescription("Cannot swap number of bytes requested");
throw e;
}
注意：异常种类
Note that itk::ByteSwapperError is a subclass of itk::ExceptionObject. In fact, all ITK exceptions derive from ExceptionObject. In this example a special constructor and C++ preprocessor
variables FILE
and LINE
are used to instantiate the exception object and provide additional information to the user. You can choose to catch a particular exception and hence a specific
ITK error, or you can trap any ITK exception by catching ExceptionObject.
3.2.6 Event Handling
注意：：
注意：设计模式
Event handling in ITK is implemented using the Subject/Observer design pattern [3] (sometimes referred to as the Command/Observer design pattern). In this approach, objects indicate that they are
watching for a particular event—invoked by a particular instance—by registering with the instance
that they are watching. For example, filters in ITK periodically invoke the itk::ProgressEvent.
Objects that have registered their interest in this event are notified when the event occurs. The notification occurs via an invocation of a command (i.e., function callback, method invocation, etc.) that
is specified during the registration process. (Note that events in ITK are subclasses of EventObject;
look in itkEventObject.h to determine which events are available.)
To recap using an example: various objects in ITK will invoke specific events as they execute (from
ProcessObject):
this->InvokeEvent( ProgressEvent() );
To watch for such an event, registration is required that associates a command (e.g., callback function) with the event: Object::AddObserver() method:
unsigned long progressTag =
filter->AddObserver(ProgressEvent(), itk::Command*);
注意：比如按左键什么的
When the event occurs, all registered observers are notified via invocation of the associated Command::Execute() method. Note that several subclasses of Command are available
supporting const and non-const member functions as well as C-style functions. (Look in
3.2. Essential System Concepts
33
Modules/Core/Common/include/itkCommand.h to find pre-defined subclasses of Command. If
nothing suitable is found, derivation is another possibility.)
3.2.7 Multi-Threading
注意：进行高层封装
Multi-threading is handled in ITK through a high-level design abstraction. This approach provides portable multi-threading and hides the complexity of differing thread implementations on the
many systems supported by ITK. For example, the class itk::PlatformMultiThreader provides support for multi-threaded execution by directly using platform-specific primitives such as
pthread create. itk::TBBMultiThreader uses Intel’s Thread Building Blocks cross-platform
library, which can do dynamic workload balancing across multiple processes. This means that
outputRegionForThread might have different sizes which change over time, depending on overall
processor load. All multi-threader implementations derive from itk::MultiThreaderBase.
Multi-threading is typically employed by an algorithm during its execution phase. For example, in
the class itk::ImageSource (a superclass for most image processing filters) the GenerateData()
注意：大部分图像处理滤波器的
method uses the following methods:
superclass
this->GetMultiThreader()->template ParallelizeImageRegion(
this->GetOutput()->GetRequestedRegion(),
[this](const OutputImageRegionType & outputRegionForThread)
{ this->DynamicThreadedGenerateData(outputRegionForThread); }, this);
In this example each thread invokes DynamicThreadedGenerateData method of the derived filter. The ParallelizeImageRegion method takes care to divide the image into different regions that do not overlap for write operations. ImageSource’s GenerateData() passes this
pointer to ParallelizeImageRegion, which allows ParallelizeImageRegion to update the filter’s progress after each region has been processed.
If a filter has some serial part in the middle, in addition to initialization done in
BeforeThreadedGenerateData() and finalization done in AfterThreadedGenerateData(), it
can parallelize more than one method in its own version of GenerateData(), such as done by
itk::CannyEdgeDetectionImageFilter:
注意：其执行算法都是用的多线
::GenerateData()
程？？--其底层的实际实现都是
用的多线程
{
this->UpdateProgress(0.0f);
Superclass::AllocateOutputs();
// Small serial section
this->UpdateProgress(0.01f);
ProgressTransformer progress1( 0.01f, 0.45f, this );
// Calculate 2nd order directional derivative
this->GetMultiThreader()->template ParallelizeImageRegion(
this->GetOutput()->GetRequestedRegion(),
[this](const OutputImageRegionType & outputRegionForThread)
{ this->ThreadedCompute2ndDerivative(outputRegionForThread); },
34
Chapter 3. System Overview
progress1.GetProcessObject());
ProgressTransformer progress2( 0.45f, 0.9f, this );
// Calculate the gradient of the second derivative
this->GetMultiThreader()->template ParallelizeImageRegion(
this->GetOutput()->GetRequestedRegion(),
[this](const OutputImageRegionType & outputRegionForThread)
{ this->ThreadedCompute2ndDerivativePos(outputRegionForThread); },
progress2.GetProcessObject());
// More processing
this->UpdateProgress(1.0f);
}
注意：是为了不让进度条从0-100出
现超过一次
When invoking ParallelizeImageRegion multiple times from GenerateData(), either nullptr
or a itk::ProgressTransformer object should be passed instead of this, otherwise progress will
go from 0% to 100% more than once. And this will at least confuse any other class watching the
filter’s progress events, even if it does not cause a crash. So the filter’s author should estimate how
long each part of GenerateData() takes, and construct and pass ProgressTransformer objects as
in the example above.
With ITK version 5.0, the Multi-Threading mechanism has been refactored. What was previously
itk::MultiThreader, is now a hierarchy of classes. itk::PlatformMultiThreader is a slightly
cleaned-up version of the old class - MultipleMethodExecute and SpawnThread methods have
been deprecated. But much of its content has been moved to itk::MultiThreaderBase. And
classes should use the multi-threaders via MultiThreaderBase interface, to allow the end user the
flexibility to select the multi-threader at run time. This also allows the filter to benefit from future
improvements in threading such as addition of a new multi-threader implementation.
The backwards compatible ThreadedGenerateData(Region, ThreadId) method signature has
been kept, for use in filters that must know their thread number.
To use this signature, a filter must invoke this->DynamicMultiThreadingOff(); before Update(); is called
by the filter’s user or downstream filter in the pipeline.
The best place for invoking
this->DynamicMultiThreadingOff(); is the filter’s constructor.
In image filters and other descendants of ProcessObject, method SetNumberOfWorkUnits controls the level of parallelism. Load balancing is possible when NumberOfWorkUnits is greater than
the number of threads. In most places where developer would like to restrict number of threads,
work units should be changed instead. itk::MultiThreaderBase’s MaximumNumberOfThreads
should not generally be changed, except when testing performance and scalability, profiling and
sometimes debugging code.
The general philosophy in ITK regarding thread safety is that accessing different instances of a class
(and its methods) is a thread-safe operation. Invoking methods on the same instance in different
threads is to be avoided.
注意：
3.3. Numerics
35
3.3 Numerics
ITK uses the VNL numerics library to provide resources for numerical programming combining the
ease of use of packages like Mathematica and Matlab with the speed of C and the elegance of C++.
It provides a C++ interface to the high-quality Fortran routines made available in the public domain
by numerical analysis researchers. ITK extends the functionality of VNL by including interface
classes between VNL and ITK proper.
注意：ITK的数值计算是封装的VNL数
值库--是不是一些矩阵运算？？？
The VNL numerics library includes classes for:
Matrices and vectors. Standard matrix and vector support and operations on these types.
Specialized matrix and vector classes. Several special matrix and vector classes with special numerical properties are available. Class vnl diagonal matrix provides a fast and convenient
diagonal matrix, while fixed size matrices and vectors allow “fast-as-C” computations (see
vnl matrix fixed and example subclasses vnl double 3x3 and vnl double 3).
Matrix decompositions. Classes vnl svd, vnl symmetric eigensystem, and vnl generalized eigensystem.
Real polynomials. Class vnl real polynomial stores the coefficients of a real polynomial, and
provides methods of evaluation of the polynomial at any x, while class vnl rpoly roots
provides a root finder.
Optimization. Classes vnl levenberg marquardt, vnl amoeba, vnl conjugate gradient,
vnl lbfgs allow optimization of user-supplied functions either with or without user-supplied
derivatives.
注意：以前接触过bfgs
Standardized functions and constants. Class vnl math defines constants (pi, e, eps...) and simple functions (sqr, abs, rnd...). Class numeric limits is from the ISO standard document, and provides a way to access basic limits of a type. For example numeric limits::max() returns the maximum value of a short.
Most VNL routines are implemented as wrappers around the high-quality Fortran routines that have
been developed by the numerical analysis community over the last forty years and placed in the public domain. The central repository for these programs is the “netlib” server.4 The National Institute
of Standards and Technology (NIST) provides an excellent search interface to this repository in its
Guide to Available Mathematical Software (GAMS),5 both as a decision tree and a text search.
注意：
ITK also provides additional numerics functionality. A suite of optimizers, that use VNL under
the hood and integrate with the registration framework are available. A large collection of statistics
functions—not available from VNL—are also provided in the Insight/Numerics/Statistics
directory. In addition, a complete finite element (FEM) package is available, primarily to support
the deformable registration in ITK.
4 http://www.netlib.org/
5 http://gams.nist.gov
36
Chapter 3. System Overview
3.4 Data Representation
There are two principle types of data represented in ITK: images and meshes. This functionality is implemented in the classes itk::Image and itk::Mesh, both of which are subclasses of
itk::DataObject. In ITK, data objects are classes that are meant to be passed around the system
and may participate in data flow pipelines (see Section 3.5 on page 37 for more information).
itk::Image represents an n-dimensional, regular sampling of data. The sampling direction is parallel to direction matrix axes, and the origin of the sampling, inter-pixel spacing, and the number
of samples in each direction (i.e., image dimension) can be specified. The sample, or pixel, type in
ITK is arbitrary—a template parameter TPixel specifies the type upon template instantiation. (The
dimensionality of the image must also be specified when the image class is instantiated.) The key is
that the pixel type must support certain operations (for example, addition or difference) if the code is
to compile in all cases (for example, to be processed by a particular filter that uses these operations).
In practice, most applications will use a C++ primitive type (e.g., int, float) or a pre-defined pixel
type and will rarely create a new type of pixel class.
注意：像素类型应当支
持一定的操作
One of the important ITK concepts regarding images is that rectangular, continuous pieces of the
image are known as regions. Regions are used to specify which part of an image to process, for
example in multi-threading, or which part to hold in memory. In ITK there are three common types
of regions:
注意：在ITK中的ROI
1. LargestPossibleRegion—the image in its entirety.
2. BufferedRegion—the portion of the image retained in memory.
3. RequestedRegion—the portion of the region requested by a filter or other class when operating on the image.
注意：由一个类型以及连接列表所定
义的cell集来表示mesh的拓扑结构
The itk::Mesh class represents an n-dimensional, unstructured grid. The topology of the mesh is
represented by a set of cells defined by a type and connectivity list; the connectivity list in turn refers
to points. The geometry of the mesh is defined by the n-dimensional points in combination with
associated cell interpolation functions. Mesh is designed as an adaptive representational structure
that changes depending on the operations performed on it. At a minimum, points and cells are
注意：隐含的连
接关系
required in order to represent a mesh; but it is possible to add additional topological information.
For example, links from the points to the cells that use each point can be added; this provides implicit
neighborhood information assuming the implied topology is the desired one. It is also possible to
specify boundary cells explicitly, to indicate different connectivity from the implied neighborhood
relationships, or to store information on the boundaries of cells.
注意：points？？cells？？？
The mesh is defined in terms of three template parameters: 1) a pixel type associated with the
points, cells, and cell boundaries; 2) the dimension of the points (which in turn limits the maximum
dimension of the cells); and 3) a “mesh traits” template parameter that specifies the types of the
containers and identifiers used to access the points, cells, and/or boundaries. By using the mesh
traits carefully, it is possible to create meshes better suited for editing, or those better suited for
“read-only” operations, allowing a trade-off between representation flexibility, memory, and speed.
3.5. Data Processing Pipeline
37
Mesh is a subclass of itk::PointSet. The PointSet class can be used to represent point clouds or
randomly distributed landmarks, etc. The PointSet class has no associated topology.
3.5 Data Processing Pipeline
While data objects (e.g., images and meshes) are used to represent data, process objects are classes
that operate on data objects and may produce new data objects. Process objects are classed as
sources, filter objects, or mappers. Sources (such as readers) produce data, filter objects take in data
and process it to produce new data, and mappers accept data for output either to a file or some other
system. Sometimes the term filter is used broadly to refer to all three types.
The data processing pipeline ties together data objects (e.g., images and meshes) and process objects.
The pipeline supports an automatic updating mechanism that causes a filter to execute if and only
if its input or its internal state changes. Further, the data pipeline supports streaming, the ability
to automatically break data into smaller pieces, process the pieces one by one, and reassemble the
注意：数据管道支持流处理
processed data into a final result.
Typically data objects and process objects are connected together using the SetInput() and
GetOutput() methods as follows:
using FloatImage2DType = itk::Image;
itk::RandomImageSource::Pointer random;
random = itk::RandomImageSource::New();
random->SetMin(0.0);
random->SetMax(1.0);
itk::ShrinkImageFilter::Pointer shrink;
shrink = itk::ShrinkImageFilter::New();
shrink->SetInput(random->GetOutput());
shrink->SetShrinkFactors(2);
itk::ImageFileWriter::Pointer writer;
writer = itk::ImageFileWriter::New();
writer->SetInput (shrink->GetOutput());
writer->SetFileName( "test.raw" );
writer->Update();
In this example the source object
itk::RandomImageSource is connected to
the
itk::ShrinkImageFilter, and the shrink filter is connected to the mapper
itk::ImageFileWriter. When the Update() method is invoked on the writer, the data
processing pipeline causes each of these filters to execute in order, culminating in writing the final
注意：
data to a file on disk.
38
Chapter 3. System Overview
3.6 Spatial Objects
The ITK spatial object framework supports the philosophy that the task of image segmentation and
registration is actually the task of object processing. The image is but one medium for representing
objects of interest, and much processing and data analysis can and should occur at the object level
注意：无论表示数据的形式，可以
and not based on the medium used to represent the object.
获取物理位置以及集合属性
ITK spatial objects provide a common interface for accessing the physical location and geometric
properties of and the relationship between objects in a scene that is independent of the form used
to represent those objects. That is, the internal representation maintained by a spatial object may
be a list of points internal to an object, the surface mesh of the object, a continuous or parametric
representation of the object’s internal points or surfaces, and so forth.
The capabilities provided by the spatial objects framework supports their use in object segmentation,
registration, surface/volume rendering, and other display and analysis functions. The spatial object
framework extends the concept of a “scene graph” that is common to computer rendering packages
注意：
so as to support these new functions. With the spatial objects framework you can:
1. Specify a spatial object’s parent and children objects. In this way, a liver may contain vessels
and those vessels can be organized in a tree structure.
2. Query if a physical point is inside an object or (optionally) any of its children.
3. Request the value and derivatives, at a physical point, of an associated intensity function, as
specified by an object or (optionally) its children.
4. Specify the coordinate transformation that maps a parent object’s coordinate system into a
child object’s coordinate system.
5. Compute the bounding box of a spatial object and (optionally) its children.
6. Query the resolution at which the object was originally computed. For example, you can
query the resolution (i.e., voxel spacing) of the image used to generate a particular instance of
注意：
a itk::BlobSpatialObject.
Currently implemented types of spatial objects include: Blob, Ellipse, Group, Image, Line, Surface,
and Tube. The itk::Scene object is used to hold a list of spatial objects that may in turn have
children. Each spatial object can be assigned a color property. Each spatial object type has its own
capabilities. For example, the itk::TubeSpatialObject indicates the point where it is connected
with its parent tube.
There are a limited number of spatial objects in ITK, but their number is growing and their potential
is huge. Using the nominal spatial object capabilities, methods such as marching cubes or mutual
information registration can be applied to objects regardless of their internal representation. By
having a common API, the same method can be used to register a parametric representation of a
heart with an individual’s CT data or to register two segmentations of a liver.
注意：不是很明白？？？？
3.7. Wrapping
39
3.7 Wrapping
While the core of ITK is implemented in C++, Python bindings can be automatically generated and
ITK programs can be created using Python. The wrapping process in ITK is capable of handling
generic programming (i.e., extensive use of C++ templates). Systems like VTK, which use their
own wrapping facility, are non-templated and customized to the coding methodology found in the
system, like object ownership conventions. Even systems like SWIG that are designed for general
wrapper generation have difficulty with ITK code because general C++ is difficult to parse. As a
result, the ITK wrapper generator uses a combination of tools to produce language bindings.
注意：ITK封装成Python
bindings的过程
1. CastXML is a Clang-based tool that produces an XML description of an input C++ program.
2. The igenerator.py script in the ITK source tree processes XML information produced by
CastXML and generates standard input files (*.i files) to the next tool (SWIG), indicating
what is to be wrapped and how to wrap it.
3. SWIG produces the appropriate Python bindings.
To learn more about the wrapping process, please see the section on module wrapping, Section 9.5.
The wrapping process is orchestrated by a number of CMake macros found in the Wrapping directory. The result of the wrapping process is a set of shared libraries (.so in Linux or .dlls on Windows)
that can be used by interpreted languages.
There is almost a direct translation from C++, with the differences being the particular syntactical
requirements of each language. For example, to dilate an image using a custom structuring element
using the Python wrapping:
注意：不同语言存在差异，但是可以
直接将C++的代码转换为具有特定语
法要求的每一个语言（编程）
inputImage = sys.argv[1]
outputImage = sys.argv[2]
radiusValue = int(sys.argv[3])
PixelType = itk.UC
Dimension = 2
ImageType = itk.Image[PixelType, Dimension]
reader = itk.ImageFileReader[ImageType].New()
reader.SetFileName(inputImage)
StructuringElementType = itk.FlatStructuringElement[Dimension]
structuringElement = StructuringElementType.Ball(radiusValue)
dilateFilter = itk.BinaryDilateImageFilter[
ImageType, ImageType, StructuringElementType].New()
dilateFilter.SetInput(reader.GetOutput())
dilateFilter.SetKernel(structuringElement)
40
Chapter 3. System Overview
The same code in C++ would appear as follows:
const char * inputImage = argv[1];
const char * outputImage = argv[2];
const unsigned int radiusValue = atoi( argv[3] );
using PixelType = unsigned char;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( inputImage );
using StructuringElementType = itk::FlatStructuringElement< Dimension >;
StructuringElementType::RadiusType radius;
radius.Fill( radiusValue );
StructuringElementType structuringElement =
StructuringElementType::Ball( radius );
using BinaryDilateImageFilterType = itk::BinaryDilateImageFilter< ImageType,
ImageType, StructuringElementType >;
BinaryDilateImageFilterType::Pointer dilateFilter =
BinaryDilateImageFilterType::New();
dilateFilter->SetInput( reader->GetOutput() );
dilateFilter->SetKernel( structuringElement );
This example demonstrates an important difference between C++ and a wrapped language such
as Python. Templated classes must be instantiated prior to wrapping. That is, the template
parameters must be specified as part of the wrapping process. In the example above, the
ImageFileReader[ImageType] indicates that this class, implementing an image source, has been
instantiated using an input and output image type of two-dimensional unsigned char values (i.e., UC).
To see the types available for a given filter, use the .GetTypes() method.
print(itk.ImageFileReader.GetTypes())
Typically just a few common types are selected for the wrapping process to avoid an explosion
of types and hence, library size. To add a new type, re-run the wrapping process to produce new
libraries. Some high-level options for these types, such as common pixels types and image dimensions, are specified during CMake configuration. The types of specific classes that should be
instantiated, based on these basic options, are defined by the *.wrap files in the wrapping directory
of a module.
Conversion of common, basic wrapped ITK classes to native Python types is supported. For example, conversion between the itk::Index and Python list or tuple is possible:
Dimesion = 3
index = itk.Index[Dimension]()
注意：内置的类型
由wrapping目录中
的*.wrap文件定
义？？
3.7. Wrapping
41
index_as_tuple = tuple(index)
index_as_list = list(index)
region = itk.ImageRegion[Dimension]()
region.SetIndex((0, 2, 0))
The advantage of interpreted languages is that they do not require the lengthy compile/link cycle of
a compiled language like C++. Moreover, they typically come with a suite of packages that provide
useful functionalities. For example, the Python ecosystem provides a variety of powerful tools for
creating sophisticated user interfaces. In the future it is likely that more applications and tests will
be implemented in the various interpreted languages supported by ITK. Other languages like Java,
Ruby, Tcl could also be wrapped in the future.
3.7.1 Python Setup
Install Stable Python Packages
Binary python packages are available in PyPI and can be installed in Python distributions downloaded from Python.org, from system package managers like apt or homebrew, or from distributions
like Anaconda.
To install the ITK Python package, run:
python -m pip install --upgrade pip
python -m pip install itk
Install Latest Python Packages
Binary python packages are built nightly from the Git master branch, and they can be installed by
running:
python -m pip install --upgrade pip
python -m pip install itk \
-f https://github.com/InsightSoftwareConsortium/ITKPythonPackage/releases/tag/latest
Build Python Packages from Source
In order to access the Python interface of ITK, make sure to compile with the CMake ITK WRAP PYTHON option. In addition, choose which pixel types and dimensions to build into the wrapped interface. Supported pixel types are represented in the CMake configuration as variables named ITK WRAP . Supported image dimensions are enumerated in the semicolon-delimited list
ITK WRAP DIMS, the default value of which is 2;3 indicating support for 2- and 3-dimensional
images. The Release CMake build configuration is recommended.
注意：一些Cmake时可以选择的像素
类型以及维度设置
42
Chapter 3. System Overview
After configuration, check to make sure that the values of the following variables are set correctly:
• PYTHON INCLUDE DIR
• PYTHON LIBRARY
• PYTHON EXECUTABLE
particularly if there are multiple Python installations on the system.
注意：虚拟环境工具？？
Python wrappers can be accessed from the build tree without installing the library. An environment
to access the itk Python module can be configured using the Python virtualenv tool, which provides
an isolated working copy of Python without interfering with Python installed at the system level.
Once the virtualenv package is installed on your system, create the virtual environment within the
directory ITK was built in. Copy the WrapITK.pth file to the lib/python2.7/site-packages on
Unix and Lib/site-packages on Windows, of the virtualenv. For example,
virtualenv --system-site-packages wrapitk-venv
cd wrapitk-venv/lib/python2.7/site-packages
cp /path/to/ITK-Wrapped/Wrapping/Generators/Python/WrapITK.pth .
cd ../../../../wrapitk-venv/bin
./python /usr/bin/ipython
import itk
注意：dll需添加至环境变量
On Windows, it is also necessary to add the ITK build directory containing the .dll files to your
PATH environmental variable if ITK is built with the CMake option BUILD SHARED LIBS enabled.
For example, the directory containing .dll files for an ITK build at C:\ITK-build when built with
Visual Studio in the Release configuration is C:\ITK-build\bin\Release.
CHAPTER
FOUR
DATA REPRESENTATION
This chapter introduces the basic classes responsible for representing data in ITK. The most common
classes are itk::Image, itk::Mesh and itk::PointSet.
4.1 Image
The itk::Image class follows the spirit of Generic Programming, where types are separated from
the algorithmic behavior of the class. ITK supports images with any pixel type and any spatial
dimension.
4.1.1 Creating an Image
The source code for this section can be found in the file
Image1.cxx.
This example illustrates how to manually construct an itk::Image class. The following is the
minimal code needed to instantiate, declare and create the Image class.
First, the header file of the Image class must be included.
#include "itkImage.h"
Then we must decide with what type to represent the pixels and what the dimension of the image
will be. With these two parameters we can instantiate the Image class. Here we create a 3D image
with unsigned short pixel data.
using ImageType = itk::Image< unsigned short, 3 >;
44
Chapter 4. Data Representation
The image can then be created by invoking the New() operator from the corresponding image type
and assigning the result to a itk::SmartPointer.
ImageType::Pointer image = ImageType::New();
In ITK, images exist in combination with one or more regions. A region is a subset of the image and
indicates a portion of the image that may be processed by other classes in the system. One of the
most common regions is the LargestPossibleRegion, which defines the image in its entirety. Other
important regions found in ITK are the BufferedRegion, which is the portion of the image actually
maintained in memory, and the RequestedRegion, which is the region requested by a filter or other
注
class when operating on the image.
意：
In ITK, manually creating an image requires that the image is instantiated as previously shown, and
注意：手动创建image需要1、如前
that regions describing the image are then associated with it.
的实例化2、与其相关联的regions
A region is defined by two classes: the itk::Index and itk::Size classes. The origin of the
region within the image is defined by the Index. The extent, or size, of the region is defined by the
Size. When an image is created manually, the user is responsible for defining the image size and
the index at which the image grid starts. These two parameters make it possible to process selected
regions.
The Index is represented by a n-dimensional array where each component is an integer indicating—
in topological image coordinates—the initial pixel of the image.
ImageType::IndexType start;
start[0] =
start[1] =
start[2] =
0;
0;
0;
// first index on X
// first index on Y
// first index on Z
The region size is represented by an array of the same dimension as the image (using the itk::Size
class). The components of the array are unsigned integers indicating the extent in pixels of the image
along every dimension.
ImageType::SizeType size;
size[0]
size[1]
size[2]
= 200;
= 200;
= 200;
// size along X
// size along Y
// size along Z
Having defined the starting index and the image size, these two parameters are used to create an
itk::ImageRegion object which basically encapsulates both concepts. The region is initialized
with the starting index and size of the image.
4.1. Image
45
ImageType::RegionType region;
region.SetSize( size );
region.SetIndex( start );
Finally, the region is passed to the Image object in order to define its extent and origin. The
SetRegions method sets the LargestPossibleRegion, BufferedRegion, and RequestedRegion simultaneously. Note that none of the operations performed to this point have allocated memory for the
image pixel data. It is necessary to invoke the Allocate() method to do this. Allocate does not
require any arguments since all the information needed for memory allocation has already been
provided by the region.
注意：
image->SetRegions( region );
image->Allocate();
In practice it is rare to allocate and initialize an image directly. Images are typically read from a
source, such a file or data acquisition hardware. The following example illustrates how an image
can be read from a file.
4.1.2 Reading an Image from a File
The source code for this section can be found in the file
Image2.cxx.
The first thing required to read an image from a file is to include the header file of the
itk::ImageFileReader class.
#include "itkImageFileReader.h"
Then, the image type should be defined by specifying the type used to represent pixels and the
dimensions of the image.
using PixelType = unsigned char;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
Using the image type, it is now possible to instantiate the image reader class. The image type is used
as a template parameter to define how the data will be represented once it is loaded into memory.
This type does not have to correspond exactly to the type stored in the file. However, a conversion
based on C-style type casting is used, so the type chosen to represent the data on disk must be
注意：这个类型是加载数据之后在内
存中表示像素的类型，不是必须与存
储于文件中的类型相对应；两者之间
应当可以基于C风格类型转换
46
Chapter 4. Data Representation
sufficient to characterize it accurately. Readers do not apply any transformation to the pixel data
other than casting from the pixel type of the file to the pixel type of the ImageFileReader. The
following illustrates a typical instantiation of the ImageFileReader type.
using ReaderType = itk::ImageFileReader< ImageType >;
The reader type can now be used to create one reader object. A itk::SmartPointer (defined by
the ::Pointer notation) is used to receive the reference to the newly created reader. The New()
method is invoked to create an instance of the image reader.
ReaderType::Pointer reader = ReaderType::New();
The minimal information required by the reader is the filename of the image to be loaded in memory.
This is provided through the SetFileName() method. The file format here is inferred from the filename extension. The user may also explicitly specify the data format using the itk::ImageIOBase
注意：指定需要读取的数据格式
class (a list of possibilities can be found in the inheritance diagram of this class.).
const char * filename = argv[1];
reader->SetFileName( filename );
Reader objects are referred to as pipeline source objects; they respond to pipeline update requests
and initiate the data flow in the pipeline. The pipeline update mechanism ensures that the reader
only executes when a data request is made to the reader and the reader has not read any data. In the
current example we explicitly invoke the Update() method because the output of the reader is not
connected to other filters. In normal application the reader’s output is connected to the input of an
image filter and the update invocation on the filter triggers an update of the reader. The following
line illustrates how an explicit update is invoked on the reader.
reader->Update();
Access to the newly read image can be gained by calling the GetOutput() method on the reader.
This method can also be called before the update request is sent to the reader. The reference to the
image will be valid even though the image will be empty until the reader actually executes.
ImageType::Pointer image = reader->GetOutput();
Any attempt to access image data before the reader executes will yield an image with no pixel data.
It is likely that a program crash will result since the image will not have been properly initialized.
4.1. Image
47
4.1.3 Accessing Pixel Data
The source code for this section can be found in the file
Image3.cxx.
This example illustrates the use of the SetPixel() and GetPixel() methods. These two methods
provide direct access to the pixel data contained in the image. Note that these two methods are
relatively slow and should not be used in situations where high-performance access is required.
Image iterators are the appropriate mechanism to efficiently access image pixel data. (See Chapter 6
on page 149 for information about image iterators.)
注意：这两个访问方
式是比较慢的，在由
性能要求时，应避免
使用；可以使用图像
迭代器以及指针
The individual position of a pixel inside the image is identified by a unique index. An index is
an array of integers that defines the position of the pixel along each dimension of the image. The
IndexType is automatically defined by the image and can be accessed using the scope operator
itk::Index. The length of the array will match the dimensions of the associated image.
The following code illustrates the declaration of an index variable and the assignment of values to
each of its components. Please note that no SmartPointer is used to access the Index. This is
because Index is a lightweight object that is not intended to be shared between objects. It is more
efficient to produce multiple copies of these small objects than to share them using the SmartPointer
mechanism.
The following lines declare an instance of the index type and initialize its content in order to associate
it with a pixel position in the image.
const ImageType::IndexType pixelIndex = {{27,29,37}}; // Position of {X,Y,Z}
Having defined a pixel position with an index, it is then possible to access the content of the pixel in
the image. The GetPixel() method allows us to get the value of the pixels.
ImageType::PixelType
pixelValue = image->GetPixel( pixelIndex );
The SetPixel() method allows us to set the value of the pixel.
image->SetPixel(
pixelIndex,
pixelValue+1
);
注意：这两个方法是低效的，不应该常用
Please note that GetPixel() returns the pixel value using copy and not reference semantics. Hence,
the method cannot be used to modify image data values.
Remember that both SetPixel() and GetPixel() are inefficient and should only be used for debugging or for supporting interactions like querying pixel values by clicking with the mouse.
注意：
48
Chapter 4. Data Representation
Size=7x6
Spacing=( 20.0, 30.0 )
300
Spacing[0]
Physical extent=( 140.0, 180.0 )
20.0
Linear Interpolation Region
Pixel Corner Centered Region
30.0
200
Spacing[1]
250
Pixel Coverage
Pixel Centered Region
150
Pixel Coordinates
100
50
Image Origin
Origin=(60.0,70.0)
0
0
50
100
150
200
Figure 4.1: Geometrical concepts associated with the ITK image.
4.1.4 Defining Origin and Spacing
The source code for this section can be found in the file
Image4.cxx.
Even though ITK can be used to perform general image processing tasks, the primary purpose of
the toolkit is the processing of medical image data. In that respect, additional information about the
images is considered mandatory. In particular the information associated with the physical spacing
between pixels and the position of the image in space with respect to some world coordinate system
are extremely important.
Image origin, voxel directions (i.e. orientation), and spacing are fundamental to many applications.
Registration, for example, is performed in physical coordinates. Improperly defined spacing, direction, and origins will result in inconsistent results in such processes. Medical images with no spatial
information should not be used for medical diagnosis, image analysis, feature extraction, assisted radiation therapy or image guided surgery. In other words, medical images lacking spatial information
are not only useless but also hazardous.
Figure 4.1 illustrates the main geometrical concepts associated with the itk::Image. In this figure,
circles are used to represent the center of pixels. The value of the pixel is assumed to exist as a
Dirac delta function located at the pixel center. Pixel spacing is measured between the pixel centers
and can be different along each dimension. The image origin is associated with the coordinates of
the first pixel in the image. For this simplified example, the voxel lattice is perfectly aligned with
physical space orientation, and the image direction is therefore an identity mapping. If the voxel
lattice samples were rotated with respect to physical space, then the image direction would contain
4.1. Image
49
a rotation matrix.
A pixel is considered to be the rectangular region surrounding the pixel center holding the data value.
Image spacing is represented in a FixedArray whose size matches the dimension of the image. In
order to manually set the spacing of the image, an array of the corresponding type must be created.
The elements of the array should then be initialized with the spacing between the centers of adjacent
pixels. The following code illustrates the methods available in the itk::Image class for dealing
with spacing and origin.
ImageType::SpacingType spacing;
// Units (e.g.,
spacing[0] = 0.33;
spacing[1] = 0.33;
spacing[2] = 1.20;
mm, inches, etc.) are defined by the application.
// spacing along X
// spacing along Y
// spacing along Z
The array can be assigned to the image using the SetSpacing() method.
image->SetSpacing( spacing );
注意：返回的是
一个引用类型
The spacing information can be retrieved from an image by using the GetSpacing() method. This
method returns a reference to a FixedArray. The returned object can then be used to read the
contents of the array. Note the use of the const keyword to indicate that the array will not be
modified.
const ImageType::SpacingType& sp = image->GetSpacing();
std::cout << "Spacing = ";
std::cout << sp[0] << ", " << sp[1] << ", " << sp[2] << std::endl;
The image origin is managed in a similar way to the spacing. A Point of the appropriate dimension
must first be allocated. The coordinates of the origin can then be assigned to every component. These
coordinates correspond to the position of the first pixel of the image with respect to an arbitrary
reference system in physical space. It is the user’s responsibility to make sure that multiple images
used in the same application are using a consistent reference system. This is extremely important in
image registration applications.
The following code illustrates the creation and assignment of a variable suitable for initializing the
image origin.
// coordinates of the center of the first pixel in N-D
ImageType::PointType newOrigin;
newOrigin.Fill(0.0);
image->SetOrigin( newOrigin );
50
Chapter 4. Data Representation
The origin can also be retrieved from an image by using the GetOrigin() method. This will return
a reference to a Point. The reference can be used to read the contents of the array. Note again the
use of the const keyword to indicate that the array contents will not be modified.
const ImageType::PointType & origin = image->GetOrigin();
std::cout << "Origin =
std::cout << origin[0]
<< origin[1]
<< origin[2]
";
<< ", "
<< ", "
<< std::endl;
The image direction matrix represents the orientation relationships between the image samples and
physical space coordinate systems. The image direction matrix is an orthonormal matrix that describes the possible permutation of image index values and the rotational aspects that are needed
to properly reconcile image index organization with physical space axis. The image directions is
a NxN matrix where N is the dimension of the image. An identity image direction indicates that
increasing values of the 1st, 2nd, 3rd index element corresponds to increasing values of the 1st, 2nd
and 3rd physical space axis respectively, and that the voxel samples are perfectly aligned with the
physical space axis.
The following code illustrates the creation and assignment of a variable suitable for initializing the
image direction with an identity.
// coordinates of the center of the first pixel in N-D
ImageType::DirectionType direction;
direction.SetIdentity();
image->SetDirection( direction );
The direction can also be retrieved from an image by using the GetDirection() method. This will
return a reference to a Matrix. The reference can be used to read the contents of the array. Note
again the use of the const keyword to indicate that the matrix contents can not be modified.
const ImageType::DirectionType& direct = image->GetDirection();
std::cout << "Direction = " << std::endl;
std::cout << direct << std::endl;
Once the spacing, origin, and direction of the image samples have been initialized, the image will
correctly map pixel indices to and from physical space coordinates. The following code illustrates
how a point in physical space can be mapped into an image index for the purpose of reading the
注意：
content of the closest pixel.
First, a itk::Point type must be declared. The point type is templated over the type used to
represent coordinates and over the dimension of the space. In this particular case, the dimension of
the point must match the dimension of the image.
4.1. Image
51
using PointType = itk::Point< double, ImageType::ImageDimension >;
The itk::Point class, like an itk::Index, is a relatively small and simple object. This means
that no itk::SmartPointer is used here and the objects are simply declared as instances, like any
other C++ class. Once the point is declared, its components can be accessed using traditional array
notation. In particular, the [] operator is available. For efficiency reasons, no bounds checking is
performed on the index used to access a particular point component. It is the user’s responsibility to
make sure that the index is in the range {0, Dimension − 1}.
PointType point;
point[0] = 1.45;
point[1] = 7.21;
point[2] = 9.28;
// x coordinate
// y coordinate
// z coordinate
The image will map the point to an index using the values of the current spacing and origin. An index
object must be provided to receive the results of the mapping. The index object can be instantiated
by using the IndexType defined in the image type.
ImageType::IndexType pixelIndex;
The TransformPhysicalPointToIndex() method of the image class will compute the pixel index
closest to the point provided. The method checks for this index to be contained inside the current
buffered pixel data. The method returns a boolean indicating whether the resulting index falls inside
the buffered region or not. The output index should not be used when the returned value of the
method is false.
The following lines illustrate the point to index mapping and the subsequent use of the pixel index
for accessing pixel data from the image.
const bool isInside =
image->TransformPhysicalPointToIndex( point, pixelIndex );
if ( isInside )
{
ImageType::PixelType pixelValue = image->GetPixel( pixelIndex );
pixelValue += 5;
image->SetPixel( pixelIndex, pixelValue );
}
Remember that GetPixel() and SetPixel() are very inefficient methods for accessing pixel data.
Image iterators should be used when massive access to pixel data is required.
The following example illustrates the mathematical relationships between image index locations and
its corresponding physical point representation for a given Image.
Let us imagine that a graphical user interface exists where the end user manually selects the voxel
index location of the left eye in a volume with a mouse interface. We need to convert that in注意：index（GUI中）<--->points
（physical）
52
Chapter 4. Data Representation
dex location to a physical location so that laser guided surgery can be accurately performed. The
TransformIndexToPhysicalPoint method can be used for this.
const ImageType::IndexType LeftEyeIndex = GetIndexFromMouseClick();
ImageType::PointType LeftEyePoint;
image->TransformIndexToPhysicalPoint(LeftEyeIndex,LeftEyePoint);
For a given index I3X1 , the physical location P3X1 is calculated as following:
注意：重要！！！
P3X1 = O3X1 + D3X3 ∗ diag(S3X1)3x3 ∗ I3X1
(4.1)
where D is an orthonormal direction cosines matrix and S is the image spacing diagonal matrix.
In matlab syntax the conversions are:
% Non-identity Spacing and Direction
spacing=diag( [0.9375, 0.9375, 1.5] );
direction=[0.998189, 0.0569345, -0.0194113;
0.0194429, -7.38061e-08, 0.999811;
0.0569237, -0.998378, -0.00110704];
point = origin + direction * spacing * LeftEyeIndex
A corresponding mathematical expansion of the C/C++ code is:
using MatrixType = itk::Matrix;
MatrixType SpacingMatrix;
SpacingMatrix.Fill( 0.0F );
const ImageType::SpacingType & ImageSpacing = image->GetSpacing();
SpacingMatrix( 0,0 ) = ImageSpacing[0];
SpacingMatrix( 1,1 ) = ImageSpacing[1];
SpacingMatrix( 2,2 ) = ImageSpacing[2];
const ImageType::DirectionType & ImageDirectionCosines =
image->GetDirection();
const ImageType::PointType &ImageOrigin = image->GetOrigin();
using VectorType = itk::Vector< double, Dimension >;
VectorType LeftEyeIndexVector;
LeftEyeIndexVector[0]= LeftEyeIndex[0];
LeftEyeIndexVector[1]= LeftEyeIndex[1];
LeftEyeIndexVector[2]= LeftEyeIndex[2];
ImageType::PointType LeftEyePointByHand =
ImageOrigin + ImageDirectionCosines * SpacingMatrix * LeftEyeIndexVector;
4.1. Image
53
4.1.5 RGB Images
注意：RGB是人类使
用三个光谱选择传
感器来分析可见光
的生理能力的一种
表示
The term RGB (Red, Green, Blue) stands for a color representation commonly used in digital imaging. RGB is a representation of the human physiological capability to analyze visual light using
three spectral-selective sensors [7, 9]. The human retina possess different types of light sensitive
cells. Three of them, known as cones, are sensitive to color [5] and their regions of sensitivity
loosely match regions of the spectrum that will be perceived as red, green and blue respectively. The
rods on the other hand provide no color discrimination and favor high resolution and high sensitivity.1 A fifth type of receptors, the ganglion cells, also known as circadian2 receptors are sensitive
to the lighting conditions that differentiate day from night. These receptors evolved as a mechanism
for synchronizing the physiology with the time of the day. Cellular controls for circadian rythms are
present in every cell of an organism and are known to be exquisitively precise [6].
The RGB space has been constructed as a representation of a physiological response to light by the
three types of cones in the human eye. RGB is not a Vector space. For example, negative numbers
are not appropriate in a color space because they will be the equivalent of “negative stimulation” on
the human eye. In the context of colorimetry, negative color values are used as an artificial construct
for color comparison in the sense that
ColorA = ColorB − ColorC
(4.2)
is just a way of saying that we can produce ColorB by combining ColorA and ColorC. However, we
must be aware that (at least in emitted light) it is not possible to subtract light. So when we mention
Equation 4.2 we actually mean
ColorB = ColorA + ColorC
(4.3)
On the other hand, when dealing with printed color and with paint, as opposed to emitted light like
in computer screens, the physical behavior of color allows for subtraction. This is because strictly
speaking the objects that we see as red are those that absorb all light frequencies except those in the
red section of the spectrum [9].
The concept of addition and subtraction of colors has to be carefully interpreted. In fact, RGB has a
different definition regarding whether we are talking about the channels associated to the three color
sensors of the human eye, or to the three phosphors found in most computer monitors or to the color
inks that are used for printing reproduction. Color spaces are usually non linear and do not even
from a group. For example, not all visible colors can be represented in RGB space [9].
ITK introduces the itk::RGBPixel type as a support for representing the values of an RGB color
space. As such, the RGBPixel class embodies a different concept from the one of an itk::Vector
in space. For this reason, the RGBPixel lacks many of the operators that may be naively expected
from it. In particular, there are no defined operations for subtraction or addition.
itk::RGBPixel从itk::Vector中实现
1 The
的，但是缺少很多itk::Vector的许
多基本操作
human eye is capable of perceiving a single isolated photon.
2 The term Circadian refers to the cycle of day and night, that is, events that are repeated with 24 hours intervals.
54
Chapter 4. Data Representation
When you intend to find the “Mean” of two RGBType pixels, you are assuming that the color in
the visual “middle” of the two input pixels can be calculated through a linear operation on their
numerical representation. This is unfortunately not the case in color spaces due to the fact that they
are based on a human physiological response [7].
If you decide to interpret RGB images as simply three independent channels then you should rather
use the itk::Vector type as pixel type. In this way, you will have access to the set of operations
that are defined in Vector spaces. The current implementation of the RGBPixel in ITK presumes
that RGB color images are intended to be used in applications where a formal interpretation of color
is desired, therefore only the operations that are valid in a color space are available in the RGBPixel
class.
注意：实际的color
space操作不同
于类型中定义的操作
The following example illustrates how RGB images can be represented in ITK.
The source code for this section can be found in the file
RGBImage.cxx.
Thanks to the flexibility offered by the Generic Programming style on which ITK is based, it is
possible to instantiate images of arbitrary pixel type. The following example illustrates how a color
image with RGB pixels can be defined.
A class intended to support the RGB pixel type is available in ITK. You could also define your own
pixel class and use it to instantiate a custom image type. In order to use the itk::RGBPixel class,
it is necessary to include its header file.
#include "itkRGBPixel.h"
The RGB pixel class is templated over a type used to represent each one of the red, green and blue
pixel components. A typical instantiation of the templated class is as follows.
using PixelType = itk::RGBPixel< unsigned char >;
The type is then used as the pixel template parameter of the image.
using ImageType = itk::Image< PixelType, 3 >;
The image type can be used to instantiate other filter, for example, an itk::ImageFileReader
object that will read the image from a file.
using ReaderType = itk::ImageFileReader< ImageType >;
Access to the color components of the pixels can now be performed using the methods provided by
the RGBPixel class.
4.1. Image
55
PixelType onePixel = image->GetPixel( pixelIndex );
PixelType::ValueType red
= onePixel.GetRed();
PixelType::ValueType green = onePixel.GetGreen();
PixelType::ValueType blue = onePixel.GetBlue();
The subindex notation can also be used since the itk::RGBPixel inherits the [] operator from the
itk::FixedArray class.
red
= onePixel[0]; // extract Red
component
green = onePixel[1]; // extract Green component
blue = onePixel[2]; // extract Blue component
std::cout << "Pixel values:" << std::endl;
std::cout << "Red = "
<< itk::NumericTraits::PrintType(red)
<< std::endl;
std::cout << "Green = "
<< itk::NumericTraits::PrintType(green)
<< std::endl;
std::cout << "Blue = "
<< itk::NumericTraits::PrintType(blue)
<< std::endl;
4.1.6 Vector Images
The source code for this section can be found in the file
VectorImage.cxx.
注意：向量图像
Many image processing tasks require images of non-scalar pixel type. A typical example is an image
of vectors. This is the image type required to represent the gradient of a scalar image. The following
注意：Vector与STL
code illustrates how to instantiate and use an image whose pixels are of vector type.
vector不一样
For convenience we use the itk::Vector class to define the pixel type. The Vector class is intended
to represent a geometrical vector in space. It is not intended to be used as an array container like the
std::vector in STL. If you are interested in containers, the itk::VectorContainer class may
provide the functionality you want.
The first step is to include the header file of the Vector class.
#include "itkVector.h"
The Vector class is templated over the type used to represent the coordinate in space and over the
dimension of the space. In this example, we want the vector dimension to match the image dimen-
中的数组容器
56
Chapter 4. Data Representation
sion, but this is by no means a requirement. We could have defined a four-dimensional image with
three-dimensional vectors as pixels.
using PixelType = itk::Vector< float, 3 >;
using ImageType = itk::Image< PixelType, 3 >;
The Vector class inherits the operator [] from the itk::FixedArray class. This makes it possible
to access the Vector’s components using index notation.
ImageType::PixelType pixelValue;
pixelValue[0] = 1.345;
// x component
pixelValue[1] = 6.841;
// y component
pixelValue[2] = 3.295;
// x component
We can now store this vector in one of the image pixels by defining an index and invoking the
SetPixel() method.
image->SetPixel(
pixelIndex,
pixelValue
);
4.1.7 Importing Image Data from a Buffer
The source code for this section can be found in the file
Image5.cxx.
This example illustrates how to import data into the itk::Image class. This is particularly useful
for interfacing with other software systems. Many systems use a contiguous block of memory as a
buffer for image pixel data. The current example assumes this is the case and feeds the buffer into
an itk::ImportImageFilter, thereby producing an image as output.
Here we create a synthetic image with a centered sphere in a locally allocated buffer and pass this
block of memory to the ImportImageFilter. This example is set up so that on execution, the user
must provide the name of an output file as a command-line argument.
First, the header file of the itk::ImportImageFilter class must be included.
#include "itkImage.h"
#include "itkImportImageFilter.h"
Next, we select the data type used to represent the image pixels. We assume that the external block
of memory uses the same data type to represent the pixels.
4.1. Image
57
using PixelType = unsigned char;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
The type of the ImportImageFilter is instantiated in the following line.
using ImportFilterType = itk::ImportImageFilter< PixelType, Dimension >;
A filter object created using the New() method is then assigned to a SmartPointer.
ImportFilterType::Pointer importFilter = ImportFilterType::New();
This filter requires the user to specify the size of the image to be produced as output. The
SetRegion() method is used to this end. The image size should exactly match the number of
pixels available in the locally allocated buffer.
ImportFilterType::SizeType size;
size[0]
size[1]
size[2]
= 200;
= 200;
= 200;
// size along X
// size along Y
// size along Z
ImportFilterType::IndexType start;
start.Fill( 0 );
ImportFilterType::RegionType region;
region.SetIndex( start );
region.SetSize( size );
importFilter->SetRegion( region );
The origin of the output image is specified with the SetOrigin() method.
const itk::SpacePrecisionType origin[ Dimension ] = { 0.0, 0.0, 0.0 };
importFilter->SetOrigin( origin );
The spacing of the image is passed with the SetSpacing() method.
// spacing isotropic volumes to 1.0
const itk::SpacePrecisionType spacing[ Dimension ] =
importFilter->SetSpacing( spacing );
{ 1.0, 1.0, 1.0 };
Next we allocate the memory block containing the pixel data to be passed to the
ImportImageFilter. Note that we use exactly the same size that was specified with the
58
Chapter 4. Data Representation
SetRegion() method. In a practical application, you may get this buffer from some other library
using a different data structure to represent the images.
const unsigned int numberOfPixels = size[0] * size[1] * size[2];
auto * localBuffer = new PixelType[ numberOfPixels ];
Here we fill up the buffer with a binary sphere. We use simple for() loops here, similar to
those found in the C or FORTRAN programming languages. Note that ITK does not use for()
loops in its internal code to access pixels. All pixel access tasks are instead performed using an
itk::ImageIterator that supports the management of n-dimensional images.
constexpr double radius2 = radius * radius;
PixelType * it = localBuffer;
for(unsigned int z=0; z < size[2]; z++)
{
const double dz = static_cast( z )
- static_cast(size[2])/2.0;
for(unsigned int y=0; y < size[1]; y++)
{
const double dy = static_cast( y )
- static_cast(size[1])/2.0;
for(unsigned int x=0; x < size[0]; x++)
{
const double dx = static_cast( x )
- static_cast(size[0])/2.0;
const double d2 = dx*dx + dy*dy + dz*dz;
*it++ = ( d2 < radius2 ) ? 255 : 0;
}
}
}
The buffer is passed to the ImportImageFilter with the SetImportPointer() method. Note that
the last argument of this method specifies who will be responsible for deleting the memory block
once it is no longer in use. A false value indicates that the ImportImageFilter will not try to
delete the buffer when its destructor is called. A true value, on the other hand, will allow the filter
注意：谁负责释放buffer
to delete the memory block upon destruction of the import filter.
For the ImportImageFilter to appropriately delete the memory block, the memory must be allocated with the C++ new() operator. Memory allocated with other memory allocation mechanisms,
such as C malloc or calloc, will not be deleted properly by the ImportImageFilter. In other
words, it is the application programmer’s responsibility to ensure that ImportImageFilter is only
given permission to delete the C++ new operator-allocated memory.
const bool importImageFilterWillOwnTheBuffer = true;
importFilter->SetImportPointer( localBuffer, numberOfPixels,
importImageFilterWillOwnTheBuffer );
注意：
4.2. PointSet
59
Finally, we can connect the output of this filter to a pipeline. For simplicity we just use a writer here,
but it could be any other filter.
using WriterType = itk::ImageFileWriter< ImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[1] );
writer->SetInput( importFilter->GetOutput() );
Note that we do not call delete on the buffer since we pass true as the last argument of
SetImportPointer(). Now the buffer is owned by the ImportImageFilter.
4.2 PointSet
4.2.1 Creating a PointSet
The source code for this section can be found in the file
PointSet1.cxx.
注意：PointSet-->Mesh
The itk::PointSet is a basic class intended to represent geometry in the form of a set of points
in N-dimensional space. It is the base class for the itk::Mesh providing the methods necessary to
manipulate sets of points. Points can have values associated with them. The type of such values is
defined by a template parameter of the itk::PointSet class (i.e., TPixelType). Two basic interaction styles of PointSets are available in ITK. These styles are referred to as static and dynamic.
The first style is used when the number of points in the set is known in advance and is not expected
to change as a consequence of the manipulations performed on the set. The dynamic style, on the
other hand, is intended to support insertion and removal of points in an efficient manner. Distinguishing between the two styles is meant to facilitate the fine tuning of a PointSet’s behavior while
optimizing performance and memory management.
In order to use the PointSet class, its header file should be included.
#include "itkPointSet.h"
Then we must decide what type of value to associate with the points. This is generally called the
PixelType in order to make the terminology consistent with the itk::Image. The PointSet is
also templated over the dimension of the space in which the points are represented. The following
declaration illustrates a typical instantiation of the PointSet class.
using PointSetType = itk::PointSet< unsigned short, 3 >;
A PointSet object is created by invoking the New() method on its type. The resulting object must be
assigned to a SmartPointer. The PointSet is then reference-counted and can be shared by multiple
60
Chapter 4. Data Representation
objects. The memory allocated for the PointSet will be released when the number of references to
the object is reduced to zero. This simply means that the user does not need to be concerned with
invoking the Delete() method on this class. In fact, the Delete() method should never be called
directly within any of the reference-counted ITK classes.
PointSetType::Pointer pointsSet = PointSetType::New();
Following the principles of Generic Programming, the PointSet class has a set of associated defined types to ensure that interacting objects can be declared with compatible types. This set of
type definitions is commonly known as a set of traits. Among the traits of the PointSet class is
PointType, which is used by the point set to represent points in space. The following declaration
takes the point type as defined in the PointSet traits and renames it to be conveniently used in the
global namespace.
using PointType = PointSetType::PointType;
The PointType can now be used to declare point objects to be inserted in the PointSet. Points are
fairly small objects, so it is inconvenient to manage them with reference counting and smart pointers.
They are simply instantiated as typical C++ classes. The Point class inherits the [] operator from
the itk::Array class. This makes it possible to access its components using index notation. For
efficiency’s sake no bounds checking is performed during index access. It is the user’s responsibility
to ensure that the index used is in the range {0, Dimension − 1}. Each of the components in the point
is associated with space coordinates. The following code illustrates how to instantiate a point and
initialize its components.
PointType p0;
p0[0] = -1.0;
p0[1] = -1.0;
p0[2] = 0.0;
//
//
//
x coordinate
y coordinate
z coordinate
注意：需要提供Points的唯一ID
（即key）
Points are inserted in the PointSet by using the SetPoint() method. This method requires the user
to provide a unique identifier for the point. The identifier is typically an unsigned integer that will
enumerate the points as they are being inserted. The following code shows how three points are
inserted into the PointSet.
pointsSet->SetPoint( 0, p0 );
pointsSet->SetPoint( 1, p1 );
pointsSet->SetPoint( 2, p2 );
It is possible to query the PointSet in order to determine how many points have been inserted into it.
This is done with the GetNumberOfPoints() method as illustrated below.
4.2. PointSet
61
const unsigned int numberOfPoints = pointsSet->GetNumberOfPoints();
std::cout << numberOfPoints << std::endl;
注意：返回的是一个指针
Points can be read from the PointSet by using the GetPoint() method and the integer identifier. The
point is stored in a pointer provided by the user. If the identifier provided does not match an existing
point, the method will return false and the contents of the point will be invalid. The following code
illustrates point access using defensive programming.
PointType pp;
bool pointExists =
pointsSet->GetPoint( 1, & pp );
if( pointExists )
{
std::cout << "Point is = " << pp << std::endl;
}
注意：凡是这种方式都不是最高效
的；迭代器以及直接访问才是最高效
的
GetPoint() and SetPoint() are not the most efficient methods to access points in the PointSet. It
is preferable to get direct access to the internal point container defined by the traits and use iterators
to walk sequentially over the list of points (as shown in the following example).
4.2.2 Getting Access to Points
The source code for this section can be found in the file
PointSet2.cxx.
The itk::PointSet class uses an internal container to manage the storage of itk::Points. It is
more efficient, in general, to manage points by using the access methods provided directly on the
points container. The following example illustrates how to interact with the point container and how
to use point iterators.
The type is defined by the traits of the PointSet class. The following line conveniently takes the
PointsContainer type from the PointSet traits and declares it in the global namespace.
using PointsContainer = PointSetType::PointsContainer;
注意：
The actual type of PointsContainer depends on what style of PointSet is being
used. The dynamic PointSet uses itk::MapContainer while the static PointSet uses
itk::VectorContainer. The vector and map containers are basically ITK wrappers around the
STL classes std::map and std::vector. By default, PointSet uses a static style, and therefore the default type of point container is VectorContainer. Both map and vector containers are templated over the type of element they contain. In this case they are templated over
PointType. Containers are reference counted objects, created with the New() method and assigned
to a itk::SmartPointer. The following line creates a point container compatible with the type of
the PointSet from which the trait has been taken.
注意：PointSet
|
—— dynamic：MapContainer—— 封装了stl：map
PointsContainer|
|
—— static：VectorContainer—— 封装了stl：vector（常用）
Points（PointType objects）
62
Chapter 4. Data Representation
PointsContainer::Pointer points = PointsContainer::New();
Points can now be defined using the PointType trait from the PointSet.
using PointType = PointSetType::PointType;
PointType p0;
PointType p1;
p0[0] = -1.0; p0[1] = 0.0; p0[2] = 0.0; // Point 0 = {-1,0,0 }
p1[0] = 1.0; p1[1] = 0.0; p1[2] = 0.0; // Point 1 = { 1,0,0 }
The created points can be inserted in the PointsContainer using the generic method
InsertElement() which requires an identifier to be provided for each point.
unsigned int pointId = 0;
points->InsertElement( pointId++ , p0 );
points->InsertElement( pointId++ , p1 );
Finally, the PointsContainer can be assigned to the PointSet. This will substitute any previously existing PointsContainer assigned to the PointSet. The assignment is done using the
SetPoints() method.
pointSet->SetPoints( points );
The PointsContainer object can be obtained from the PointSet using the GetPoints() method.
This method returns a pointer to the actual container owned by the PointSet which is then assigned
to a SmartPointer.
PointsContainer::Pointer points2 = pointSet->GetPoints();
注
意：
The most efficient way to sequentially visit the points is to use the iterators provided by PointsContainer. The Iterator type belongs to the traits of the PointsContainer classes. It behaves pretty
much like the STL iterators.3 The Points iterator is not a reference counted class, so it is created
directly from the traits without using SmartPointers.
using PointsIterator = PointsContainer::Iterator;
The subsequent use of the iterator follows what you may expect from a STL iterator. The iterator
to the first point is obtained from the container with the Begin() method and assigned to another
iterator.
3 If you dig deep enough into the code, you will discover that these iterators are actually ITK wrappers around STL
iterators.
4.2. PointSet
63
PointsIterator pointIterator = points->Begin();
The ++ operator on the iterator can be used to advance from one point to the next. The actual value
of the Point to which the iterator is pointing can be obtained with the Value() method. The loop for
walking through all the points can be controlled by comparing the current iterator with the iterator
returned by the End() method of the PointsContainer. The following lines illustrate the typical loop
for walking through the points.
PointsIterator end = points->End();
while( pointIterator != end )
{
PointType p = pointIterator.Value();
std::cout << p << std::endl;
++pointIterator;
}
// access the point
// print the point
// advance to next point
Note that as in STL, the iterator returned by the End() method is not a valid iterator. This is called
a past-end iterator in order to indicate that it is the value resulting from advancing one step after
visiting the last element in the container.
The number of elements stored in a container can be queried with the Size() method. In the case
of the PointSet, the following two lines of code are equivalent, both of them returning the number
of points in the PointSet.
std::cout << pointSet->GetNumberOfPoints() << std::endl;
std::cout << pointSet->GetPoints()->Size() << std::endl;
4.2.3 Getting Access to Data in Points
The source code for this section can be found in the file
PointSet3.cxx.
The itk::PointSet class was designed to interact with the Image class. For this reason it was
found convenient to allow the points in the set to hold values that could be computed from images.
The value associated with the point is referred as PixelType in order to make it consistent with
image terminology. Users can define the type as they please thanks to the flexibility offered by the
Generic Programming approach used in the toolkit. The PixelType is the first template parameter
of the PointSet.
The following code defines a particular type for a pixel type and instantiates a PointSet class with it.
64
Chapter 4. Data Representation
using PixelType = unsigned short;
using PointSetType = itk::PointSet< PixelType, 3 >;
Data can be inserted into the PointSet using the SetPointData() method. This method requires the
user to provide an identifier. The data in question will be associated to the point holding the same
identifier. It is the user’s responsibility to verify the appropriate matching between inserted data and
inserted points. The following line illustrates the use of the SetPointData() method.
unsigned int dataId = 0;
PixelType value
= 79;
pointSet->SetPointData( dataId++, value );
Data associated with points can be read from the PointSet using the GetPointData() method. This
method requires the user to provide the identifier to the point and a valid pointer to a location where
the pixel data can be safely written. In case the identifier does not match any existing identifier on
the PointSet the method will return false and the pixel value returned will be invalid. It is the user’s
responsibility to check the returned boolean value before attempting to use it.
const bool found = pointSet->GetPointData( dataId, & value );
if( found )
{
std::cout << "Pixel value = " << value << std::endl;
}
The SetPointData() and GetPointData() methods are not the most efficient way to get access
to point data. It is far more efficient to use the Iterators provided by the PointDataContainer.
注意：
Data associated with points is internally stored in PointDataContainers. In the same way as
with points, the actual container type used depend on whether the style of the PointSet is static
or dynamic. Static point sets will use an itk::VectorContainer while dynamic point sets will
use an itk::MapContainer. The type of the data container is defined as one of the traits in the
PointSet. The following declaration illustrates how the type can be taken from the traits and used to
conveniently declare a similar type on the global namespace.
using PointDataContainer = PointSetType::PointDataContainer;
Using the type it is now possible to create an instance of the data container. This is a standard
reference counted object, henceforth it uses the New() method for creation and assigns the newly
created object to a SmartPointer.
PointDataContainer::Pointer pointData = PointDataContainer::New();
Pixel data can be inserted in the container with the method InsertElement(). This method requires
an identified to be provided for each point data.
4.2. PointSet
65
unsigned int pointId = 0;
PixelType value0 = 34;
PixelType value1 = 67;
pointData->InsertElement( pointId++ , value0 );
pointData->InsertElement( pointId++ , value1 );
Finally the PointDataContainer can be assigned to the PointSet. This will substitute any previously
existing PointDataContainer on the PointSet. The assignment is done using the SetPointData()
method.
pointSet->SetPointData( pointData );
The PointDataContainer can be obtained from the PointSet using the GetPointData() method.
This method returns a pointer (assigned to a SmartPointer) to the actual container owned by the
PointSet.
PointDataContainer::Pointer pointData2 = pointSet->GetPointData();
The most efficient way to sequentially visit the data associated with points is to use the iterators
provided by PointDataContainer. The Iterator type belongs to the traits of the PointsContainer
classes. The iterator is not a reference counted class, so it is just created directly from the traits
without using SmartPointers.
using PointDataIterator = PointDataContainer::Iterator;
The subsequent use of the iterator follows what you may expect from a STL iterator. The iterator
to the first point is obtained from the container with the Begin() method and assigned to another
iterator.
PointDataIterator pointDataIterator = pointData2->Begin();
The ++ operator on the iterator can be used to advance from one data point to the next. The actual
value of the PixelType to which the iterator is pointing can be obtained with the Value() method.
The loop for walking through all the point data can be controlled by comparing the current iterator
with the iterator returned by the End() method of the PointsContainer. The following lines illustrate
the typical loop for walking through the point data.
PointDataIterator end = pointData2->End();
while( pointDataIterator != end )
{
PixelType p = pointDataIterator.Value(); // access the pixel data
66
Chapter 4. Data Representation
std::cout << p << std::endl;
++pointDataIterator;
}
// print the pixel data
// advance to next pixel/point
Note that as in STL, the iterator returned by the End() method is not a valid iterator. This is called
a past-end iterator in order to indicate that it is the value resulting from advancing one step after
注意：PointSet
visiting the last element in the container.
|
—— dynamic：MapContainer—— 封装了stl：map
PointDataContainer|
|
—— static：VectorContainer—— 封装了stl：vector（常用）
PixelType（value）（前面的是存放点数据，现在这个是直接存放值数据）
4.2.4 RGB as Pixel Type
The source code for this section can be found in the file
RGBPointSet.cxx.
The following example illustrates how a point set can be parameterized to manage a particular pixel
type. In this case, pixels of RGB type are used. The first step is then to include the header files of
the itk::RGBPixel and itk::PointSet classes.
#include "itkRGBPixel.h"
#include "itkPointSet.h"
Then, the pixel type can be defined by selecting the type to be used to represent each one of the RGB
components.
using PixelType = itk::RGBPixel< float >;
The newly defined pixel type is now used to instantiate the PointSet type and subsequently create a
point set object.
using PointSetType = itk::PointSet< PixelType, 3 >;
PointSetType::Pointer pointSet = PointSetType::New();
The following code generates a circle and assigns RGB values to the points. The components of the
RGB values in this example are computed to represent the position of the points.
PointSetType::PixelType pixel;
PointSetType::PointType point;
unsigned int pointId = 0;
constexpr double radius = 3.0;
for(unsigned int i=0; i<360; i++)
{
const double angle = i * itk::Math::pi / 180.0;
point[0] = radius * std::sin( angle );
point[1] = radius * std::cos( angle );
4.2. PointSet
67
point[2] = 1.0;
pixel.SetRed(
point[0] * 2.0 );
pixel.SetGreen( point[1] * 2.0 );
pixel.SetBlue(
point[2] * 2.0 );
pointSet->SetPoint( pointId, point );
pointSet->SetPointData( pointId, pixel );
pointId++;
}
All the points on the PointSet are visited using the following code.
using PointIterator = PointSetType::PointsContainer::ConstIterator;
PointIterator pointIterator = pointSet->GetPoints()->Begin();
PointIterator pointEnd
= pointSet->GetPoints()->End();
while( pointIterator != pointEnd )
{
point = pointIterator.Value();
std::cout << point << std::endl;
++pointIterator;
}
Note that here the ConstIterator was used instead of the Iterator since the pixel values are not
expected to be modified. ITK supports const-correctness at the API level.
All the pixel values on the PointSet are visited using the following code.
using PointDataIterator = PointSetType::PointDataContainer::ConstIterator;
PointDataIterator pixelIterator = pointSet->GetPointData()->Begin();
= pointSet->GetPointData()->End();
PointDataIterator pixelEnd
while( pixelIterator != pixelEnd )
{
pixel = pixelIterator.Value();
std::cout << pixel << std::endl;
++pixelIterator;
}
Again, please note the use of the ConstIterator instead of the Iterator.
4.2.5 Vectors as Pixel Type
The source code for this section can be found in the file
PointSetWithVectors.cxx.
This example illustrates how a point set can be parameterized to manage a particular pixel type.
It is quite common to associate vector values with points for producing geometric representations.
The following code shows how vector values can be used as the pixel type on the PointSet class.
68
Chapter 4. Data Representation
The itk::Vector class is used here as the pixel type. This class is appropriate for representing the
relative position between two points. It could then be used to manage displacements, for example.
In order to use the vector class it is necessary to include its header file along with the header of the
point set.
#include "itkVector.h"
#include "itkPointSet.h"
Figure 4.2: Vectors as PixelType.
The Vector class is templated over the
type used to represent the spatial coordinates
and
over
the
space
dimension.
Since the PixelType is independent of the PointType, we are free to select any dimension for the vectors to
be used as pixel type.
However, for
the sake of producing an interesting example,
we will use vectors that represent displacements of the points in the
PointSet.
Those vectors are then selected to be of the same dimension as the
PointSet.
constexpr unsigned int Dimension = 3;
using PixelType = itk::Vector< float, Dimension >;
Then we use the PixelType (which are actually Vectors) to instantiate the PointSet type and subsequently create a PointSet object.
using PointSetType = itk::PointSet< PixelType, Dimension >;
PointSetType::Pointer pointSet = PointSetType::New();
The following code is generating a sphere and assigning vector values to the points. The components
of the vectors in this example are computed to represent the tangents to the circle as shown in
Figure 4.2.
PointSetType::PixelType
PointSetType::PointType
tangent;
point;
unsigned int pointId = 0;
constexpr double radius = 300.0;
for(unsigned int i=0; i<360; i++)
注意：存储点，上面还可以存储点的
值
注意：
vector
中存储
的是位
移
4.2. PointSet
69
{
const double angle = i * itk::Math::pi / 180.0;
point[0] = radius * std::sin( angle );
point[1] = radius * std::cos( angle );
point[2] = 1.0;
// flat on the Z plane
tangent[0] = std::cos(angle);
tangent[1] = -std::sin(angle);
tangent[2] = 0.0; // flat on the Z plane
pointSet->SetPoint( pointId, point );
pointSet->SetPointData( pointId, tangent );
pointId++;
}
We can now visit all the points and use the vector on the pixel values to apply a displacement on the
points. This is along the spirit of what a deformable model could do at each one of its iterations.
using PointDataIterator = PointSetType::PointDataContainer::ConstIterator;
PointDataIterator pixelIterator = pointSet->GetPointData()->Begin();
PointDataIterator pixelEnd
= pointSet->GetPointData()->End();
using PointIterator = PointSetType::PointsContainer::Iterator;
PointIterator pointIterator = pointSet->GetPoints()->Begin();
PointIterator pointEnd
= pointSet->GetPoints()->End();
while( pixelIterator != pixelEnd && pointIterator != pointEnd )
{
pointIterator.Value() = pointIterator.Value() + pixelIterator.Value();
++pixelIterator;
++pointIterator;
}
Note that the ConstIterator was used here instead of the normal Iterator since the pixel values
are only intended to be read and not modified. ITK supports const-correctness at the API level.
The itk::Vector class has overloaded the + operator with the itk::Point. In other words,
vectors can be added to points in order to produce new points. This property is exploited in the
center of the loop in order to update the points positions with a single statement.
We can finally visit all the points and print out the new values
pointIterator = pointSet->GetPoints()->Begin();
pointEnd
= pointSet->GetPoints()->End();
while( pointIterator != pointEnd )
{
std::cout << pointIterator.Value() << std::endl;
++pointIterator;
}
Note that itk::Vector is not the appropriate class for representing normals to surfaces and gradients of functions. This is due to the way vectors behave under affine transforms. ITK has a specific
70
Chapter 4. Data Representation
class for representing normals and function gradients. This is the itk::CovariantVector class.
注意：用于表示法向量以及函数梯度
的类
4.2.6 Normals as Pixel Type
The source code for this section can be found in the file
PointSetWithCovariantVectors.cxx.
注意：用于表示梯
度以及法向量
It is common to represent geometric objects by using points on their surfaces and normals associated
with those points. This structure can be easily instantiated with the itk::PointSet class.
The natural class for representing normals to surfaces and gradients of functions is the
itk::CovariantVector. A covariant vector differs from a vector in the way it behaves under
affine transforms, in particular under anisotropic scaling. If a covariant vector represents the gradient of a function, the transformed covariant vector will still be the valid gradient of the transformed
function, a property which would not hold with a regular vector.
The following example demonstrates how a CovariantVector can be used as the PixelType for the
PointSet class. The example illustrates how a deformable model could move under the influence
of the gradient of a potential function.
In order to use the CovariantVector class it is necessary to include its header file along with the
header of the point set.
#include "itkCovariantVector.h"
#include "itkPointSet.h"
The CovariantVector class is templated over the type used to represent the spatial coordinates and
over the space dimension. Since the PixelType is independent of the PointType, we are free to select
any dimension for the covariant vectors to be used as pixel type. However, we want to illustrate here
the spirit of a deformable model. It is then required for the vectors representing gradients to be of
the same dimension as the points in space.
constexpr unsigned int Dimension = 3;
using PixelType = itk::CovariantVector< float, Dimension >;
Then we use the PixelType (which are actually CovariantVectors) to instantiate the PointSet type
and subsequently create a PointSet object.
using PointSetType = itk::PointSet< PixelType, Dimension >;
PointSetType::Pointer pointSet = PointSetType::New();
The following code generates a circle and assigns gradient values to the points. The components of
the CovariantVectors in this example are computed to represent the normals to the circle.
4.2. PointSet
PointSetType::PixelType
PointSetType::PointType
71
gradient;
point;
unsigned int pointId = 0;
constexpr double radius = 300.0;
for(unsigned int i=0; i<360; i++)
{
const double angle = i * std::atan(1.0) / 45.0;
point[0] = radius * std::sin( angle );
point[1] = radius * std::cos( angle );
point[2] = 1.0;
// flat on the Z plane
gradient[0] = std::sin(angle);
gradient[1] = std::cos(angle);
gradient[2] = 0.0; // flat on the Z plane
pointSet->SetPoint( pointId, point );
pointSet->SetPointData( pointId, gradient );
pointId++;
}
We can now visit all the points and use the vector on the pixel values to apply a deformation on the
points by following the gradient of the function. This is along the spirit of what a deformable model
could do at each one of its iterations. To be more formal we should use the function gradients as
forces and multiply them by local stress tensors in order to obtain local deformations. The resulting
deformations would finally be used to apply displacements on the points. However, to shorten the
example, we will ignore this complexity for the moment.
using PointDataIterator = PointSetType::PointDataContainer::ConstIterator;
PointDataIterator pixelIterator = pointSet->GetPointData()->Begin();
PointDataIterator pixelEnd
= pointSet->GetPointData()->End();
using PointIterator = PointSetType::PointsContainer::Iterator;
PointIterator pointIterator = pointSet->GetPoints()->Begin();
PointIterator pointEnd
= pointSet->GetPoints()->End();
while( pixelIterator != pixelEnd && pointIterator != pointEnd )
{
point
= pointIterator.Value();
gradient = pixelIterator.Value();
for(unsigned int i=0; i;
Meshes typically require large amounts of memory. For this reason, they are reference counted
注意：
4.3. Mesh
73
objects, managed using itk::SmartPointers. The following line illustrates how a mesh is created
by invoking the New() method on MeshType and assigning the result to a SmartPointer.
MeshType::Pointer mesh = MeshType::New();
Management of points in a Mesh is identical to that in a PointSet. The type of point associated with
the mesh can be obtained through the PointType trait. The following code shows the creation of
points compatible with the mesh type defined above and the assignment of values to its coordinates.
MeshType::PointType
MeshType::PointType
MeshType::PointType
MeshType::PointType
p0;
p1;
p2;
p3;
p0[0]= -1.0; p0[1]= -1.0; p0[2]= 0.0; //
p1[0]= 1.0; p1[1]= -1.0; p1[2]= 0.0; //
p2[0]= 1.0; p2[1]= 1.0; p2[2]= 0.0; //
p3[0]= -1.0; p3[1]= 1.0; p3[2]= 0.0; //
first
second
third
fourth
point
point
point
point
( -1, -1, 0 )
( 1, -1, 0 )
( 1, 1, 0 )
( -1, 1, 0 )
注意：
The points can now be inserted into the Mesh using the SetPoint() method. Note that points are
copied into the mesh structure, meaning that the local instances of the points can now be modified
without affecting the Mesh content.
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
0,
1,
2,
3,
p0
p1
p2
p3
);
);
);
);
The current number of points in a mesh can be queried with the GetNumberOfPoints() method.
std::cout << "Points = " << mesh->GetNumberOfPoints() << std::endl;
The points can now be efficiently accessed using the Iterator to the PointsContainer as was done
in the previous section for the PointSet.
using PointsIterator = MeshType::PointsContainer::Iterator;
A point iterator is initialized to the first point with the Begin() method of the PointsContainer.
PointsIterator pointIterator = mesh->GetPoints()->Begin();
The ++ operator is used to advance the iterator from one point to the next. The value associated
with the Point to which the iterator is pointing is obtained with the Value() method. The loop
74
Chapter 4. Data Representation
for walking through all the points is controlled by comparing the current iterator with the iterator
returned by the End() method of the PointsContainer. The following illustrates the typical loop
for walking through the points of a mesh.
PointsIterator end = mesh->GetPoints()->End();
while( pointIterator != end )
{
MeshType::PointType p = pointIterator.Value(); // access the point
std::cout << p << std::endl;
// print the point
++pointIterator;
// advance to next point
}
4.3.2 Inserting Cells
The source code for this section can be found in the file
Mesh2.cxx.
注意：可以插入cells，在mesh中
A itk::Mesh can contain a variety of cell types. Typical cells are the itk::LineCell,
itk::TriangleCell,
itk::QuadrilateralCell,
itk::TetrahedronCell, and
itk::PolygonCell. Additional flexibility is provided for managing cells at the price of a
bit more of complexity than in the case of point management.
The following code creates a polygonal line in order to illustrate the simplest case of cell management in a mesh. The only cell type used here is the LineCell. The header file of this class must be
included.
#include "itkLineCell.h"
For consistency with Mesh, cell types have to be configured with a number of custom types taken
from the mesh traits. The set of traits relevant to cells are packaged by the Mesh class into the
CellType trait. This trait needs to be passed to the actual cell types at the moment of their instantiation. The following line shows how to extract the Cell traits from the Mesh type.
using CellType = MeshType::CellType;
The LineCell type can now be instantiated using the traits taken from the Mesh.
using LineType = itk::LineCell< CellType >;
注意：points与cells两者的存储差
异
The main difference in the way cells and points are managed by the Mesh is that points are stored
by copy on the PointsContainer while cells are stored as pointers in the CellsContainer. The
reason for using pointers is that cells use C++ polymorphism on the mesh. This means that the mesh
4.3. Mesh
75
is only aware of having pointers to a generic cell which is the base class of all the specific cell types.
This architecture makes it possible to combine different cell types in the same mesh. Points, on the
other hand, are of a single type and have a small memory footprint, which makes it efficient to copy
them directly into the container.
Managing cells by pointers adds another level of complexity to the Mesh since it is now necessary to
establish a protocol to make clear who is responsible for allocating and releasing the cells’ memory.
This protocol is implemented in the form of a specific type of pointer called the CellAutoPointer.
This pointer, based on the itk::AutoPointer, differs in many respects from the SmartPointer.
The CellAutoPointer has an internal pointer to the actual object and a boolean flag that indicates
whether the CellAutoPointer is responsible for releasing the cell memory when the time comes
for its own destruction. It is said that a CellAutoPointer owns the cell when it is responsible for
its destruction. At any given time many CellAutoPointers can point to the same cell, but only one
CellAutoPointer can own the cell.
注意：
The CellAutoPointer trait is defined in the MeshType and can be extracted as follows.
using CellAutoPointer = CellType::CellAutoPointer;
Note that the CellAutoPointer points to a generic cell type. It is not aware of the actual type of
the cell, which could be (for example) a LineCell, TriangleCell or TetrahedronCell. This fact
will influence the way in which we access cells later on.
At this point we can actually create a mesh and insert some points on it.
MeshType::Pointer mesh = MeshType::New();
MeshType::PointType p0;
MeshType::PointType p1;
MeshType::PointType p2;
p0[0] = -1.0; p0[1] = 0.0; p0[2] = 0.0;
p1[0] = 1.0; p1[1] = 0.0; p1[2] = 0.0;
p2[0] = 1.0; p2[1] = 1.0; p2[2] = 0.0;
mesh->SetPoint( 0, p0 );
mesh->SetPoint( 1, p1 );
mesh->SetPoint( 2, p2 );
The following code creates two CellAutoPointers and initializes them with newly created cell
objects. The actual cell type created in this case is LineType. Note that cells are created with the
normal new C++ operator. The CellAutoPointer takes ownership of the received pointer by using the
method TakeOwnership(). Even though this may seem verbose, it is necessary in order to make it
explicit that the responsibility of memory release is assumed by the AutoPointer.
76
Chapter 4. Data Representation
CellAutoPointer line0;
CellAutoPointer line1;
line0.TakeOwnership( new LineType );
line1.TakeOwnership( new LineType );
The LineCells should now be associated with points in the mesh. This is done using the identifiers assigned to points when they were inserted in the mesh. Every cell type has a specific number of points
that must be associated with it.4 For example, a LineCell requires two points, a TriangleCell
requires three, and a TetrahedronCell requires four. Cells use an internal numbering system for
points. It is simply an index in the range {0, NumberO f Points − 1}. The association of points and
cells is done by the SetPointId() method, which requires the user to provide the internal index of
the point in the cell and the corresponding PointIdentifier in the Mesh. The internal cell index
is the first parameter of SetPointId() while the mesh point-identifier is the second.
line0->SetPointId( 0, 0 ); // line between points 0 and 1
line0->SetPointId( 1, 1 );
line1->SetPointId( 0, 1 ); // line between points 1 and 2
line1->SetPointId( 1, 2 );
Cells are inserted in the mesh using the SetCell() method. It requires an identifier and the AutoPointer to the cell. The Mesh will take ownership of the cell to which the CellAutoPointer
is pointing. This is done internally by the SetCell() method. In this way, the destruction of the
CellAutoPointer will not induce the destruction of the associated cell.
mesh->SetCell( 0, line0 );
mesh->SetCell( 1, line1 );
注意：调用完SetCell（）方法之
后，CellAutoPointer不在拥有cell
的所有权；即不负责释放cell内存
After serving as an argument of the SetCell() method, a CellAutoPointer no longer holds ownership of the cell. It is important not to use this same CellAutoPointer again as argument to
SetCell() without first securing ownership of another cell.
The number of Cells currently inserted in the mesh can be queried with the GetNumberOfCells()
method.
std::cout << "Cells
= " << mesh->GetNumberOfCells() << std::endl;
In a way analogous to points, cells can be accessed using Iterators to the CellsContainer in the
mesh. The trait for the cell iterator can be extracted from the mesh and used to define a local type.
4
Some cell types like polygons have a variable number of points associated with them.
4.3. Mesh
77
using CellIterator = MeshType::CellsContainer::Iterator;
Then the iterators to the first and past-end cell in the mesh can be obtained respectively with the
Begin() and End() methods of the CellsContainer. The CellsContainer of the mesh is returned by the GetCells() method.
CellIterator cellIterator = mesh->GetCells()->Begin();
CellIterator end
= mesh->GetCells()->End();
Finally, a standard loop is used to iterate over all the cells. Note the use of the Value() method used
to get the actual pointer to the cell from the CellIterator. Note also that the value returned is a pointer
to the generic CellType. This pointer must be downcast in order to be used as actual LineCell types.
Safe down-casting is performed with the dynamic cast operator, which will throw an exception if
the conversion cannot be safely performed.
while( cellIterator != end )
{
MeshType::CellType * cellptr = cellIterator.Value();
auto * line = dynamic_cast( cellptr );
if(line == nullptr)
{
continue;
}
std::cout << line->GetNumberOfPoints() << std::endl;
++cellIterator;
}
4.3.3 Managing Data in Cells
The source code for this section can be found in the file
Mesh3.cxx.
Just as custom data can be associated with points in the mesh, it is also possible to associate custom
data with cells. The type of the data associated with the cells can be different from the data type
associated with points. By default, however, these two types are the same. The following example
illustrates how to access data associated with cells. The approach is analogous to the one used to
access point data.
Consider the example of a mesh containing lines on which values are associated with each line. The
mesh and cell header files should be included first.
78
Chapter 4. Data Representation
#include "itkMesh.h"
#include "itkLineCell.h"
Then the PixelType is defined and the mesh type is instantiated with it.
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 2 >;
The itk::LineCell type can now be instantiated using the traits taken from the Mesh.
using CellType = MeshType::CellType;
using LineType = itk::LineCell< CellType >;
Let’s now create a Mesh and insert some points into it. Note that the dimension of the points matches
the dimension of the Mesh. Here we insert a sequence of points that look like a plot of the log()
function. We add the vnl math::eps value in order to avoid numerical errors when the point id is
zero. The value of vnl math::eps is the difference between 1.0 and the least value greater than
1.0 that is representable in this computer.
MeshType::Pointer mesh = MeshType::New();
using PointType = MeshType::PointType;
PointType point;
constexpr unsigned int numberOfPoints = 10;
for(unsigned int id=0; id( id ); // x
point[1] = std::log( static_cast( id ) + itk::Math::eps );
mesh->SetPoint( id, point );
}
// y
A set of line cells is created and associated with the existing points by using point identifiers. In this
simple case, the point identifiers can be deduced from cell identifiers since the line cells are ordered
in the same way.
CellType::CellAutoPointer line;
const unsigned int numberOfCells = numberOfPoints-1;
for(unsigned int cellId=0; cellIdSetPointId( 0, cellId
); // first point
line->SetPointId( 1, cellId+1 ); // second point
mesh->SetCell( cellId, line );
// insert the cell
}
4.3. Mesh
79
Data associated with cells is inserted in the itk::Mesh by using the SetCellData() method. It
requires the user to provide an identifier and the value to be inserted. The identifier should match
one of the inserted cells. In this simple example, the square of the cell identifier is used as cell data.
Note the use of static cast to PixelType in the assignment.
for(unsigned int cellId=0; cellIdSetCellData( cellId, static_cast( cellId * cellId ) );
}
Cell data can be read from the Mesh with the GetCellData() method. It requires the user to provide
the identifier of the cell for which the data is to be retrieved. The user should provide also a valid
pointer to a location where the data can be copied.
for(unsigned int cellId=0; cellId(0.0);
mesh->GetCellData( cellId, &value );
std::cout << "Cell " << cellId << " = " << value << std::endl;
}
Neither SetCellData() or GetCellData() are efficient ways to access cell data. More efficient
access to cell data can be achieved by using the Iterators built into the CellDataContainer.
using CellDataIterator = MeshType::CellDataContainer::ConstIterator;
Note that the ConstIterator is used here because the data is only going to be read. This approach
is exactly the same already illustrated for getting access to point data. The iterator to the first cell
data item can be obtained with the Begin() method of the CellDataContainer. The past-end
iterator is returned by the End() method. The cell data container itself can be obtained from the
mesh with the method GetCellData().
CellDataIterator cellDataIterator = mesh->GetCellData()->Begin();
CellDataIterator end
= mesh->GetCellData()->End();
Finally, a standard loop is used to iterate over all the cell data entries. Note the use of the Value()
method to get the value associated with the data entry. PixelType elements are copied into the local
variable cellValue.
while( cellDataIterator != end )
{
PixelType cellValue = cellDataIterator.Value();
std::cout << cellValue << std::endl;
++cellDataIterator;
}
80
Chapter 4. Data Representation
4.3.4 Customizing the Mesh
The source code for this section can be found in the file
MeshTraits.cxx.
This section illustrates the full power of Generic Programming. This is sometimes perceived as too
much of a good thing!
The toolkit has been designed to offer flexibility while keeping the complexity of the code to a moderate level. This is achieved in the Mesh by hiding most of its parameters and defining reasonable
defaults for them.
The generic concept of a mesh integrates many different elements. It is possible in principle to use
independent types for every one of such elements. The mechanism used in generic programming for
specifying the many different types involved in a concept is called traits. They are basically the list
注意：
of all types that interact with the current class.
The itk::Mesh is templated over three parameters. So far only two of them have been discussed,
namely the PixelType and the Dimension. The third parameter is a class providing the set of traits
required by the mesh. When the third parameter is omitted a default class is used. This default class
is the itk::DefaultStaticMeshTraits. If you want to customize the types used by the mesh, the
way to proceed is to modify the default traits and provide them as the third parameter of the Mesh
注意：自定义mesh所使用的类型
class instantiation.
There are two ways of achieving this.
The first is to use the existing
itk::DefaultStaticMeshTraits class. This class is itself templated over six parameters.
Customizing those parameters could provide enough flexibility to define a very specific kind of
mesh. The second way is to write a traits class from scratch, in which case the easiest way to
proceed is to copy the DefaultStaticMeshTraits into another file and edit its content. Only the
first approach is illustrated here. The second is discouraged unless you are familiar with Generic
Programming, feel comfortable with C++ templates, and have access to an abundant supply of
(Columbian) coffee.
The first step in customizing the mesh is to include the header file of the Mesh and its static traits.
#include "itkMesh.h"
#include "itkDefaultStaticMeshTraits.h"
Then the MeshTraits class is instantiated by selecting the types of each one of its six template
arguments. They are in order
PixelType. The value type associated with every point.
PointDimension. The dimension of the space in which the mesh is embedded.
4.3. Mesh
81
MaxTopologicalDimension. The highest dimension of the mesh cells.
CoordRepType. The type used to represent spacial coordinates.
InterpolationWeightType. The type used to represent interpolation weights.
CellPixelType. The value type associated with every cell.
Let’s define types and values for each one of those elements. For example, the following code
uses points in 3D space as nodes of the Mesh. The maximum dimension of the cells will be two,
meaning that this is a 2D manifold better know as a surface. The data type associated with points is
defined to be a four-dimensional vector. This type could represent values of membership for a fourclass segmentation method. The value selected for the cells are 4 × 3 matrices, which could have
for example the derivative of the membership values with respect to coordinates in space. Finally,
a double type is selected for representing space coordinates on the mesh points and also for the
weight used for interpolating values.
注意：重新定义上述几个mesh特化的
类型，就可以自定义一个新的mesh类
型；维持的是一个typelist吧？？？
constexpr unsigned int PointDimension = 3;
constexpr unsigned int MaxTopologicalDimension = 2;
using PixelType = itk::Vector;
using CellDataType = itk::Matrix;
using CoordinateType = double;
using InterpolationWeightType = double;
using MeshTraits = itk::DefaultStaticMeshTraits<
PixelType, PointDimension, MaxTopologicalDimension,
CoordinateType, InterpolationWeightType, CellDataType >;
using MeshType = itk::Mesh< PixelType, PointDimension, MeshTraits >;
The itk::LineCell type can now be instantiated using the traits taken from the Mesh.
using CellType = MeshType::CellType;
using LineType = itk::LineCell< CellType >;
Let’s now create an Mesh and insert some points on it. Note that the dimension of the points matches
the dimension of the Mesh. Here we insert a sequence of points that look like a plot of the log()
function.
MeshType::Pointer mesh = MeshType::New();
using PointType = MeshType::PointType;
PointType point;
constexpr unsigned int numberOfPoints = 10;
for(unsigned int id=0; idSetPoint( id, point );
}
A set of line cells is created and associated with the existing points by using point identifiers. In this
simple case, the point identifiers can be deduced from cell identifiers since the line cells are ordered
in the same way. Note that in the code above, the values assigned to point components are arbitrary.
In a more realistic example, those values would be computed from another source.
CellType::CellAutoPointer line;
const unsigned int numberOfCells = numberOfPoints-1;
for(unsigned int cellId=0; cellIdSetPointId( 0, cellId
); // first point
line->SetPointId( 1, cellId+1 ); // second point
mesh->SetCell( cellId, line );
// insert the cell
}
Data associated with cells is inserted in the Mesh by using the SetCellData() method. It requires
the user to provide an identifier and the value to be inserted. The identifier should match one of the
inserted cells. In this example, we simply store a CellDataType dummy variable named value.
for(unsigned int cellId=0; cellIdSetCellData( cellId, value );
}
Cell data can be read from the Mesh with the GetCellData() method. It requires the user to provide
the identifier of the cell for which the data is to be retrieved. The user should provide also a valid
pointer to a location where the data can be copied.
for(unsigned int cellId=0; cellIdGetCellData( cellId, &value );
std::cout << "Cell " << cellId << " = " << value << std::endl;
}
注意：这几个Set以及Get函数都不是
最效率的；迭代器才是
Neither SetCellData() or GetCellData() are efficient ways to access cell data. Efficient access
to cell data can be achieved by using the Iterators built into the CellDataContainer.
4.3. Mesh
83
using CellDataIterator = MeshType::CellDataContainer::ConstIterator;
Note that the ConstIterator is used here because the data is only going to be read. This approach
is identical to that already illustrated for accessing point data. The iterator to the first cell data item
can be obtained with the Begin() method of the CellDataContainer. The past-end iterator is
returned by the End() method. The cell data container itself can be obtained from the mesh with the
method GetCellData().
CellDataIterator cellDataIterator = mesh->GetCellData()->Begin();
= mesh->GetCellData()->End();
CellDataIterator end
Finally a standard loop is used to iterate over all the cell data entries. Note the use of the Value()
method used to get the actual value of the data entry. PixelType elements are returned by copy.
while( cellDataIterator != end )
{
CellDataType cellValue = cellDataIterator.Value();
std::cout << cellValue << std::endl;
++cellDataIterator;
}
4.3.5 Topology and the K-Complex
The source code for this section can be found in the file
MeshKComplex.cxx.
The itk::Mesh class supports the representation of formal topologies. In particular the concept
of K-Complex can be correctly represented in the Mesh. An informal definition of K-Complex may
be as follows: a K-Complex is a topological structure in which for every cell of dimension N, its
boundary faces (which are cells of dimension N − 1) also belong to the structure.
This section illustrates how to instantiate a K-Complex structure using the mesh. The example structure is composed of one tetrahedron, its four triangle faces, its six line edges and its four vertices.
The header files of all the cell types involved should be loaded along with the header file of the mesh
class.
#include "itkMesh.h"
#include "itkLineCell.h"
#include "itkTetrahedronCell.h"
Then the PixelType is defined and the mesh type is instantiated with it. Note that the dimension of
the space is three in this case.
注意：K-Complex结构
84
Chapter 4. Data Representation
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 3 >;
The cell type can now be instantiated using the traits taken from the Mesh.
using
using
using
using
using
CellType = MeshType::CellType;
VertexType = itk::VertexCell< CellType >;
LineType = itk::LineCell< CellType >;
TriangleType = itk::TriangleCell< CellType >;
TetrahedronType = itk::TetrahedronCell< CellType >;
The mesh is created and the points associated with the vertices are inserted. Note that there is
an important distinction between the points in the mesh and the itk::VertexCell concept. A
VertexCell is a cell of dimension zero. Its main difference as compared to a point is that the cell
can be aware of neighborhood relationships with other cells. Points are not aware of the existence
of cells. In fact, from the pure topological point of view, the coordinates of points in the mesh are
completely irrelevant. They may as well be absent from the mesh structure altogether. VertexCells
on the other hand are necessary to represent the full set of neighborhood relationships on the KComplex.
The geometrical coordinates of the nodes of a regular tetrahedron can be obtained by taking every
other node from a regular cube.
MeshType::Pointer mesh = MeshType::New();
MeshType::PointType
MeshType::PointType
MeshType::PointType
MeshType::PointType
point0[0]
point1[0]
point2[0]
point3[0]
point0;
point1;
point2;
point3;
= -1; point0[1] = -1; point0[2] = -1;
= 1; point1[1] = 1; point1[2] = -1;
= 1; point2[1] = -1; point2[2] = 1;
= -1; point3[1] = 1; point3[2] = 1;
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
0,
1,
2,
3,
point0
point1
point2
point3
);
);
);
);
We proceed now to create the cells, associate them with the points and insert them on the mesh.
Starting with the tetrahedron we write the following code.
CellType::CellAutoPointer cellpointer;
cellpointer.TakeOwnership( new TetrahedronType );
cellpointer->SetPointId( 0, 0 );
4.3. Mesh
cellpointer->SetPointId( 1, 1
cellpointer->SetPointId( 2, 2
cellpointer->SetPointId( 3, 3
mesh->SetCell( 0, cellpointer
85
);
);
);
);
Four triangular faces are created and associated with the mesh now. The first triangle connects points
0,1,2.
cellpointer.TakeOwnership( new TriangleType );
cellpointer->SetPointId( 0, 0 );
cellpointer->SetPointId( 1, 1 );
cellpointer->SetPointId( 2, 2 );
mesh->SetCell( 1, cellpointer );
The second triangle connects points 0, 2, 3 .
cellpointer.TakeOwnership( new TriangleType );
cellpointer->SetPointId( 0, 0 );
cellpointer->SetPointId( 1, 2 );
cellpointer->SetPointId( 2, 3 );
mesh->SetCell( 2, cellpointer );
The third triangle connects points 0, 3, 1 .
cellpointer.TakeOwnership( new TriangleType );
cellpointer->SetPointId( 0, 0 );
cellpointer->SetPointId( 1, 3 );
cellpointer->SetPointId( 2, 1 );
mesh->SetCell( 3, cellpointer );
The fourth triangle connects points 3, 2, 1 .
cellpointer.TakeOwnership( new TriangleType );
cellpointer->SetPointId( 0, 3 );
cellpointer->SetPointId( 1, 2 );
cellpointer->SetPointId( 2, 1 );
mesh->SetCell( 4, cellpointer );
注意：当AutoPointer传递给SetCell
函数之后，就失去了cell的拥有权；
即可以复用
Note how the CellAutoPointer is reused every time. Reminder: the itk::AutoPointer loses
ownership of the cell when it is passed as an argument of the SetCell() method. The AutoPointer
is attached to a new cell by using the TakeOwnership() method.
The construction of the K-Complex continues now with the creation of the six lines on the tetrahedron edges.
86
Chapter 4. Data Representation
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 0 );
cellpointer->SetPointId( 1, 1 );
mesh->SetCell( 5, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 1 );
cellpointer->SetPointId( 1, 2 );
mesh->SetCell( 6, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 2 );
cellpointer->SetPointId( 1, 0 );
mesh->SetCell( 7, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 1 );
cellpointer->SetPointId( 1, 3 );
mesh->SetCell( 8, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 3 );
cellpointer->SetPointId( 1, 2 );
mesh->SetCell( 9, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 3 );
cellpointer->SetPointId( 1, 0 );
mesh->SetCell( 10, cellpointer );
Finally the zero dimensional cells represented by the itk::VertexCell are created and inserted in
the mesh.
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 0 );
mesh->SetCell( 11, cellpointer );
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 1 );
mesh->SetCell( 12, cellpointer );
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 2 );
mesh->SetCell( 13, cellpointer );
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 3 );
mesh->SetCell( 14, cellpointer );
At this point the Mesh contains four points and fifteen cells enumerated from 0 to 14. The points
can be visited using PointContainer iterators.
4.3. Mesh
87
using PointIterator = MeshType::PointsContainer::ConstIterator;
PointIterator pointIterator = mesh->GetPoints()->Begin();
PointIterator pointEnd
= mesh->GetPoints()->End();
while( pointIterator != pointEnd )
{
std::cout << pointIterator.Value() << std::endl;
++pointIterator;
}
The cells can be visited using CellsContainer iterators.
using CellIterator = MeshType::CellsContainer::ConstIterator;
CellIterator cellIterator = mesh->GetCells()->Begin();
CellIterator cellEnd
= mesh->GetCells()->End();
while( cellIterator != cellEnd )
{
CellType * cell = cellIterator.Value();
std::cout << cell->GetNumberOfPoints() << std::endl;
++cellIterator;
}
Note that cells are stored as pointer to a generic cell type that is the base class of all the specific cell
classes. This means that at this level we can only have access to the virtual methods defined in the
CellType.
The point identifiers to which the cells have been associated can be visited using iterators defined in the CellType trait. The following code illustrates the use of the PointIdIterators.
The PointIdsBegin() method returns the iterator to the first point-identifier in the cell. The
PointIdsEnd() method returns the iterator to the past-end point-identifier in the cell.
using PointIdIterator = CellType::PointIdIterator;
PointIdIterator pointIditer = cell->PointIdsBegin();
PointIdIterator pointIdend = cell->PointIdsEnd();
while( pointIditer != pointIdend )
{
std::cout << *pointIditer << std::endl;
++pointIditer;
}
Note that the point-identifier is obtained from the iterator using the more traditional *iterator
notation instead the Value() notation used by cell-iterators.
Up to here, the topology of the K-Complex is not completely defined since we have only introduced
the cells. ITK allows the user to define explicitly the neighborhood relationships between cells. It
88
Chapter 4. Data Representation
注意：如何表示相互之间的关系
is clear that a clever exploration of the point identifiers could have allowed a user to figure out the
neighborhood relationships. For example, two triangle cells sharing the same two point identifiers
will probably be neighbor cells. Some of the drawbacks on this implicit discovery of neighborhood
relationships is that it takes computing time and that some applications may not accept the same
assumptions. A specific case is surgery simulation. This application typically simulates bistoury
cuts in a mesh representing an organ. A small cut in the surface may be made by specifying that two
triangles are not considered to be neighbors any more.
Neighborhood relationships are represented in the mesh by the notion of BoundaryFeature. Every
cell has an internal list of cell-identifiers pointing to other cells that are considered to be its neighbors. Boundary features are classified by dimension. For example, a line will have two boundary
features of dimension zero corresponding to its two vertices. A tetrahedron will have boundary features of dimension zero, one and two, corresponding to its four vertices, six edges and four triangular
faces. It is up to the user to specify the connections between the cells.
Let’s take in our current example the tetrahedron cell that was associated with the cell-identifier 0
and assign to it the four vertices as boundaries of dimension zero. This is done by invoking the
SetBoundaryAssignment() method on the Mesh class.
MeshType::CellIdentifier cellId = 0;
// the tetrahedron
int dimension = 0;
// vertices
MeshType::CellFeatureIdentifier featureId = 0;
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
dimension,
dimension,
dimension,
dimension,
cellId,
cellId,
cellId,
cellId,
featureId++,
featureId++,
featureId++,
featureId++,
11
12
13
14
);
);
);
);
The featureId is simply a number associated with the sequence of the boundary cells of the same
dimension in a specific cell. For example, the zero-dimensional features of a tetrahedron are its four
vertices. Then the zero-dimensional feature-Ids for this cell will range from zero to three. The onedimensional features of the tetrahedron are its six edges, hence its one-dimensional feature-Ids will
range from zero to five. The two-dimensional features of the tetrahedron are its four triangular faces.
The two-dimensional feature ids will then range from zero to three. The following table summarizes
the use on indices for boundary assignments.
Dimension
0
1
2
CellType
VertexCell
LineCell
TriangleCell
FeatureId range
[0:3]
[0:5]
[0:3]
Cell Ids
{11,12,13,14}
{5,6,7,8,9,10}
{1,2,3,4}
In the code example above, the values of featureId range from zero to three. The cell identifiers of
the triangle cells in this example are the numbers {1,2,3,4}, while the cell identifiers of the vertex
cells are the numbers {11,12,13,14}.
注意：每一个cell
都有一个内置的指
向其他邻居cell的
cell ID链表
4.3. Mesh
89
Let’s now assign one-dimensional boundary features of the tetrahedron. Those are the line cells with
identifiers {5,6,7,8,9,10}. Note that the feature identifier is reinitialized to zero since the count is
independent for each dimension.
cellId
= 0;
dimension = 1;
featureId = 0;
// still the tetrahedron
// one-dimensional features = edges
// reinitialize the count
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
dimension,
dimension,
dimension,
dimension,
dimension,
dimension,
cellId,
cellId,
cellId,
cellId,
cellId,
cellId,
featureId++,
featureId++,
featureId++,
featureId++,
featureId++,
featureId++,
5
6
7
8
9
10
);
);
);
);
);
);
Finally we assign the two-dimensional boundary features of the tetrahedron. These are the four triangular cells with identifiers {1,2,3,4}. The featureId is reset to zero since feature-Ids are independent
on each dimension.
cellId
= 0;
dimension = 2;
featureId = 0;
// still the tetrahedron
// two-dimensional features = triangles
// reinitialize the count
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
mesh->SetBoundaryAssignment(
dimension,
dimension,
dimension,
dimension,
cellId,
cellId,
cellId,
cellId,
featureId++,
featureId++,
featureId++,
featureId++,
1
2
3
4
);
);
);
);
At this point we can query the tetrahedron cell for information about its boundary features. For
example, the number of boundary features of each dimension can be obtained with the method
GetNumberOfBoundaryFeatures().
cellId = 0; // still the tetrahedron
MeshType::CellFeatureCount n0;
MeshType::CellFeatureCount n1;
MeshType::CellFeatureCount n2;
// number of zero-dimensional features
// number of one-dimensional features
// number of two-dimensional features
n0 = mesh->GetNumberOfCellBoundaryFeatures( 0, cellId );
n1 = mesh->GetNumberOfCellBoundaryFeatures( 1, cellId );
n2 = mesh->GetNumberOfCellBoundaryFeatures( 2, cellId );
The boundary assignments can be recovered with the method GetBoundaryAssigment(). For example, the zero-dimensional features of the tetrahedron can be obtained with the following code.
dimension = 0;
for(unsigned int b0=0; b0 < n0; b0++)
{
90
Chapter 4. Data Representation
MeshType::CellIdentifier id;
bool found = mesh->GetBoundaryAssignment( dimension, cellId, b0, &id );
if( found ) std::cout << id << std::endl;
}
The following code illustrates how to set the edge boundaries for one of the triangular faces.
cellId
=
dimension =
featureId =
2;
1;
0;
// one of the triangles
// boundary edges
// start the count of features
mesh->SetBoundaryAssignment( dimension, cellId, featureId++, 7 );
mesh->SetBoundaryAssignment( dimension, cellId, featureId++, 9 );
mesh->SetBoundaryAssignment( dimension, cellId, featureId++, 10 );
注意：熟悉上述API
4.3.6 Representing a PolyLine
The source code for this section can be found in the file
MeshPolyLine.cxx.
This section illustrates how to represent a classical PolyLine structure using the itk::Mesh
A PolyLine only involves zero and one dimensional cells, which are represented by the
itk::VertexCell and the itk::LineCell.
#include "itkMesh.h"
#include "itkLineCell.h"
Then the PixelType is defined and the mesh type is instantiated with it. Note that the dimension of
the space is two in this case.
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 2 >;
The cell type can now be instantiated using the traits taken from the Mesh.
using CellType = MeshType::CellType;
using VertexType = itk::VertexCell< CellType >;
using LineType = itk::LineCell< CellType >;
The mesh is created and the points associated with the vertices are inserted. Note that there is an
important distinction between the points in the mesh and the itk::VertexCell concept. A VertexCell is a cell of dimension zero. Its main difference as compared to a point is that the cell can be
4.3. Mesh
91
aware of neighborhood relationships with other cells. Points are not aware of the existence of cells.
In fact, from the pure topological point of view, the coordinates of points in the mesh are completely
irrelevant. They may as well be absent from the mesh structure altogether. VertexCells on the other
hand are necessary to represent the full set of neighborhood relationships on the Polyline.
In this example we create a polyline connecting the four vertices of a square by using three of the
square sides.
MeshType::Pointer mesh = MeshType::New();
MeshType::PointType
MeshType::PointType
MeshType::PointType
MeshType::PointType
point0[0]
point1[0]
point2[0]
point3[0]
point0;
point1;
point2;
point3;
= -1; point0[1] = -1;
= 1; point1[1] = -1;
= 1; point2[1] = 1;
= -1; point3[1] = 1;
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
mesh->SetPoint(
0,
1,
2,
3,
point0
point1
point2
point3
);
);
);
);
We proceed now to create the cells, associate them with the points and insert them on the mesh.
CellType::CellAutoPointer cellpointer;
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 0 );
cellpointer->SetPointId( 1, 1 );
mesh->SetCell( 0, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 1 );
cellpointer->SetPointId( 1, 2 );
mesh->SetCell( 1, cellpointer );
cellpointer.TakeOwnership( new LineType );
cellpointer->SetPointId( 0, 2 );
cellpointer->SetPointId( 1, 0 );
mesh->SetCell( 2, cellpointer );
Finally the zero dimensional cells represented by the itk::VertexCell are created and inserted in
the mesh.
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 0 );
mesh->SetCell( 3, cellpointer );
92
Chapter 4. Data Representation
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 1 );
mesh->SetCell( 4, cellpointer );
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 2 );
mesh->SetCell( 5, cellpointer );
cellpointer.TakeOwnership( new VertexType );
cellpointer->SetPointId( 0, 3 );
mesh->SetCell( 6, cellpointer );
At this point the Mesh contains four points and three cells. The points can be visited using PointContainer iterators.
using PointIterator = MeshType::PointsContainer::ConstIterator;
PointIterator pointIterator = mesh->GetPoints()->Begin();
PointIterator pointEnd
= mesh->GetPoints()->End();
while( pointIterator != pointEnd )
{
std::cout << pointIterator.Value() << std::endl;
++pointIterator;
}
The cells can be visited using CellsContainer iterators.
using CellIterator = MeshType::CellsContainer::ConstIterator;
CellIterator cellIterator = mesh->GetCells()->Begin();
= mesh->GetCells()->End();
CellIterator cellEnd
while( cellIterator != cellEnd )
{
CellType * cell = cellIterator.Value();
std::cout << cell->GetNumberOfPoints() << std::endl;
++cellIterator;
}
Note that cells are stored as pointer to a generic cell type that is the base class of all the specific cell
classes. This means that at this level we can only have access to the virtual methods defined in the
CellType.
The point identifiers to which the cells have been associated can be visited using iterators
defined in the CellType trait. The following code illustrates the use of the PointIdIterator.
The PointIdsBegin() method returns the iterator to the first point-identifier in the cell. The
PointIdsEnd() method returns the iterator to the past-end point-identifier in the cell.
4.3. Mesh
93
using PointIdIterator = CellType::PointIdIterator;
PointIdIterator pointIditer = cell->PointIdsBegin();
PointIdIterator pointIdend = cell->PointIdsEnd();
while( pointIditer != pointIdend )
{
std::cout << *pointIditer << std::endl;
++pointIditer;
}
Note that the point-identifier is obtained from the iterator using the more traditional *iterator
notation instead the Value() notation used by cell-iterators.
4.3.7 Simplifying Mesh Creation
The source code for this section can be found in the file
AutomaticMesh.cxx.
The itk::Mesh class is extremely general and flexible, but there is some cost to convenience. If
convenience is exactly what you need, then it is possible to get it, in exchange for some of that
flexibility, by means of the itk::AutomaticTopologyMeshSource class. This class automatically
generates an explicit K-Complex, based on the cells you add. It explicitly includes all boundary
information, so that the resulting mesh can be easily traversed. It merges all shared edges, vertices,
and faces, so no geometric feature appears more than once.
注意：
This section shows how you can use the AutomaticTopologyMeshSource to instantiate a mesh representing a K-Complex. We will first generate the same tetrahedron from Section 4.3.5, after which
we will add a hollow one to illustrate some additional features of the mesh source.
The header files of all the cell types involved should be loaded along with the header file of the mesh
class.
#include "itkTriangleCell.h"
#include "itkAutomaticTopologyMeshSource.h"
We then define the necessary types and instantiate the mesh source. Two new types are
IdentifierType and IdentifierArrayType. Every cell in a mesh has an identifier, whose type
is determined by the mesh traits. AutomaticTopologyMeshSource requires that the identifier type of
all vertices and cells be unsigned long, which is already the default. However, if you created a new
mesh traits class to use string tags as identifiers, the resulting mesh would not be compatible with
itk::AutomaticTopologyMeshSource. An IdentifierArrayType is simply an itk::Array of
IdentifierType objects.
94
Chapter 4. Data Representation
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 3 >;
using PointType = MeshType::PointType;
using MeshSourceType = itk::AutomaticTopologyMeshSource< MeshType >;
using IdentifierArrayType = MeshSourceType::IdentifierArrayType;
MeshSourceType::Pointer meshSource;
meshSource = MeshSourceType::New();
Now let us generate the tetrahedron. The following line of code generates all the vertices, edges,
and faces, along with the tetrahedral solid, and adds them to the mesh along with the connectivity
information.
meshSource->AddTetrahedron(
meshSource->AddPoint( -1,
meshSource->AddPoint( 1,
meshSource->AddPoint( 1,
meshSource->AddPoint( -1,
);
-1, -1 ),
1, -1 ),
-1, 1 ),
1, 1 )
The function AutomaticTopologyMeshSource::AddTetrahedron() takes point identifiers
as parameters; the identifiers must correspond to points that have already been added.
AutomaticTopologyMeshSource::AddPoint() returns the appropriate identifier type for the point
being added. It first checks to see if the point is already in the mesh. If so, it returns the ID of the
point in the mesh, and if not, it generates a new unique ID, adds the point with that ID, and returns
the ID.
Actually, AddTetrahedron() behaves in the same way. If the tetrahedron has already been added,
it leaves the mesh unchanged and returns the ID that the tetrahedron already has. If not, it adds the
tetrahedron (and all its faces, edges, and vertices), and generates a new ID, which it returns.
It is also possible to add all the points first, and then add a number of cells using the point IDs
directly. This approach corresponds with the way the data is stored in many file formats for 3D
polygonal models.
First we add the points (in this case the vertices of a larger tetrahedron). This example also illustrates
that AddPoint() can take a single PointType as a parameter if desired, rather than a sequence of
floats. Another possibility (not illustrated) is to pass in a C-style array.
PointType p;
IdentifierArrayType idArray( 4 );
p[ 0 ] = -2;
p[ 1 ] = -2;
p[ 2 ] = -2;
4.3. Mesh
95
idArray[ 0 ] = meshSource->AddPoint( p );
p[ 0 ] =
p[ 1 ] =
p[ 2 ] =
idArray[
2;
2;
-2;
1 ] = meshSource->AddPoint( p );
p[ 0 ] =
p[ 1 ] =
p[ 2 ] =
idArray[
2;
-2;
2;
2 ] = meshSource->AddPoint( p );
p[ 0 ] =
p[ 1 ] =
p[ 2 ] =
idArray[
-2;
2;
2;
3 ] = meshSource->AddPoint( p );
Now we add the cells. This time we are just going to create the boundary of a tetrahedron, so we
must add each face separately.
meshSource->AddTriangle(
meshSource->AddTriangle(
meshSource->AddTriangle(
meshSource->AddTriangle(
idArray[0],
idArray[1],
idArray[2],
idArray[3],
idArray[1],
idArray[2],
idArray[3],
idArray[0],
idArray[2]
idArray[3]
idArray[0]
idArray[1]
);
);
);
);
Actually, we could have called, e.g., AddTriangle( 4, 5, 6 ), since IDs are assigned sequentially starting at zero, and idArray[0] contains the ID for the fifth point added. But you should
only do this if you are confident that you know what the IDs are. If you add the same point twice
and don’t realize it, your count will differ from that of the mesh source.
You may be wondering what happens if you call, say, AddEdge(0, 1) followed by AddEdge(1,
0). The answer is that they do count as the same edge, and so only one edge is added. The order of
the vertices determines an orientation, and the first orientation specified is the one that is kept.
Once you have built the mesh you want, you can access it by calling GetOutput(). Here we send it
to cout, which prints some summary data for the mesh.
In contrast to the case with typical filters, GetOutput() does not trigger an update process. The
mesh is always maintained in a valid state as cells are added, and can be accessed at any time. It
would, however, be a mistake to modify the mesh by some other means until AutomaticTopologyMeshSource is done with it, since the mesh source would then have an inaccurate record of which
注意：
points and cells are currently in the mesh.
AutomaticTopologyMeshSource 完成
之前；访问mesh是会获取错误数据
4.3.8 Iterating Through Cells
The source code for this section can be found in the file
MeshCellsIteration.cxx.
96
Chapter 4. Data Representation
Cells are stored in the itk::Mesh as pointers to a generic cell itk::CellInterface. This implies
that only the virtual methods defined on this base cell class can be invoked. In order to use methods
that are specific to each cell type it is necessary to down-cast the pointer to the actual type of the
cell. This can be done safely by taking advantage of the GetType() method that allows to identify
the actual type of a cell.
Let’s start by assuming a mesh defined with one tetrahedron and all its boundary faces. That is, four
triangles, six edges and four vertices.
The cells can be visited using CellsContainer iterators . The iterator Value() corresponds to a raw
pointer to the CellType base class.
using CellIterator = MeshType::CellsContainer::ConstIterator;
CellIterator cellIterator = mesh->GetCells()->Begin();
CellIterator cellEnd
= mesh->GetCells()->End();
while( cellIterator != cellEnd )
{
CellType * cell = cellIterator.Value();
std::cout << cell->GetNumberOfPoints() << std::endl;
++cellIterator;
}
In order to perform down-casting in a safe manner, the cell type can be queried first using
the GetType() method. Codes for the cell types have been defined with an enum type on the
itkCellInterface.h header file. These codes are :
• VERTEX CELL
• LINE CELL
• TRIANGLE CELL
• QUADRILATERAL CELL
• POLYGON CELL
• TETRAHEDRON CELL
• HEXAHEDRON CELL
• QUADRATIC EDGE CELL
• QUADRATIC TRIANGLE CELL
The method GetType() returns one of these codes. It is then possible to test the type of the cell
before down-casting its pointer to the actual type. For example, the following code visits all the
注意：不同的cell可能存在不同的
type；所以要进行GetType（）操作
4.3. Mesh
97
cells in the mesh and tests which ones are actually of type LINE CELL. Only those cells are downcasted to LineType cells and a method specific for the LineType is invoked.
cellIterator = mesh->GetCells()->Begin();
cellEnd
= mesh->GetCells()->End();
while( cellIterator != cellEnd )
{
CellType * cell = cellIterator.Value();
if( cell->GetType() == CellType::LINE_CELL )
{
auto * line = static_cast( cell );
std::cout << "dimension = " << line->GetDimension();
std::cout << " # points = " << line->GetNumberOfPoints();
std::cout << std::endl;
}
++cellIterator;
}
In order to perform different actions on different cell types a switch statement can be used with
cases for every cell type. The following code illustrates an iteration over the cells and the invocation
of different methods on each cell type.
cellIterator = mesh->GetCells()->Begin();
cellEnd
= mesh->GetCells()->End();
while( cellIterator != cellEnd )
{
CellType * cell = cellIterator.Value();
switch( cell->GetType() )
{
case CellType::VERTEX_CELL:
{
std::cout << "VertexCell : " << std::endl;
auto * line = dynamic_cast( cell );
std::cout << "dimension = " << line->GetDimension()
std::cout << "# points = " << line->GetNumberOfPoints()
break;
}
case CellType::LINE_CELL:
{
std::cout << "LineCell : " << std::endl;
auto * line = dynamic_cast( cell );
std::cout << "dimension = " << line->GetDimension()
std::cout << "# points = " << line->GetNumberOfPoints()
break;
}
case CellType::TRIANGLE_CELL:
{
std::cout << "TriangleCell : " << std::endl;
auto * line = dynamic_cast( cell );
std::cout << "dimension = " << line->GetDimension()
<< std::endl;
<< std::endl;
<< std::endl;
<< std::endl;
<< std::endl;
98
Chapter 4. Data Representation
std::cout <<
break;
}
default:
{
std::cout <<
std::cout <<
std::cout <<
break;
}
}
++cellIterator;
}
"# points
= " << line->GetNumberOfPoints() << std::endl;
"Cell with more than three points" << std::endl;
"dimension = " << cell->GetDimension()
<< std::endl;
"# points = " << cell->GetNumberOfPoints() << std::endl;
4.3.9 Visiting Cells
The source code for this section can be found in the file
MeshCellVisitor.cxx.
In order to facilitate access to particular cell types, a convenience mechanism has been built-in on
the itk::Mesh. This mechanism is based on the Visitor Pattern presented in [3]. The visitor pattern
is designed to facilitate the process of walking through an heterogeneous list of objects sharing a
common base class.
The first requirement for using
CellInterfaceVisitor header file.
the
CellVisitor
mechanism
it
to
include
the
#include "itkCellInterfaceVisitor.h"
The typical mesh types are now declared.
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 3 >;
using CellType = MeshType::CellType;
using
using
using
using
VertexType = itk::VertexCell< CellType >;
LineType = itk::LineCell< CellType >;
TriangleType = itk::TriangleCell< CellType >;
TetrahedronType = itk::TetrahedronCell< CellType >;
Then, a custom CellVisitor class should be declared. In this particular example, the visitor class is
intended to act only on TriangleType cells. The only requirement on the declaration of the visitor
class is that it must provide a method named Visit(). This method expects as arguments a cell
identifier and a pointer to the specific cell type for which this visitor is intended. Nothing prevents a
4.3. Mesh
99
visitor class from providing Visit() methods for several different cell types. The multiple methods
will be differentiated by the natural C++ mechanism of function overload. The following code
illustrates a minimal cell visitor class.
class CustomTriangleVisitor
{
public:
using TriangleType = itk::TriangleCell;
void Visit(unsigned long cellId, TriangleType * t )
{
std::cout << "Cell # " << cellId << " is a TriangleType ";
std::cout << t->GetNumberOfPoints() << std::endl;
}
CustomTriangleVisitor() = default;
virtual ˜CustomTriangleVisitor() = default;
};
This newly defined class will now be used to instantiate a cell visitor. In this particular example we
create a class CustomTriangleVisitor which will be invoked each time a triangle cell is found
while the mesh iterates over the cells.
using TriangleVisitorInterfaceType = itk::CellInterfaceVisitorImplementation<
PixelType,
MeshType::CellTraits,
TriangleType,
CustomTriangleVisitor >;
Note that the actual CellInterfaceVisitorImplementation is templated over the PixelType, the
CellTraits, the CellType to be visited and the Visitor class that defines with will be done with the
cell.
A visitor implementation class can now be created using the normal invocation to its New() method
and assigning the result to a itk::SmartPointer.
TriangleVisitorInterfaceType::Pointer triangleVisitor =
TriangleVisitorInterfaceType::New();
Many different visitors can be configured in this way. The set of all visitors can be registered with
the MultiVisitor class provided for the mesh. An instance of the MultiVisitor class will walk through
the cells and delegate action to every registered visitor when the appropriate cell type is encountered.
using CellMultiVisitorType = CellType::MultiVisitor;
CellMultiVisitorType::Pointer multiVisitor = CellMultiVisitorType::New();
The visitor is registered with the Mesh using the AddVisitor() method.
100
Chapter 4. Data Representation
multiVisitor->AddVisitor( triangleVisitor );
Finally, the iteration over the cells is triggered by calling the method Accept() on the itk::Mesh.
mesh->Accept( multiVisitor );
The Accept() method will iterate over all the cells and for each one will invite the MultiVisitor to
attempt an action on the cell. If no visitor is interested on the current cell type the cell is just ignored
and skipped.
MultiVisitors make it possible to add behavior to the cells without having to create new methods on
the cell types or creating a complex visitor class that knows about every CellType.
注意：MultiVisitors的重要性
4.3.10 More on Visiting Cells
The source code for this section can be found in the file
MeshCellVisitor2.cxx.
The following section illustrates a realistic example of the use of Cell visitors on the itk::Mesh. A
set of different visitors is defined here, each visitor associated with a particular type of cell. All the
visitors are registered with a MultiVisitor class which is passed to the mesh.
The first step is to include the CellInterfaceVisitor header file.
#include "itkCellInterfaceVisitor.h"
The typical mesh types are now declared.
using PixelType = float;
using MeshType = itk::Mesh< PixelType, 3 >;
using CellType = MeshType::CellType;
using
using
using
using
VertexType = itk::VertexCell< CellType >;
LineType = itk::LineCell< CellType >;
TriangleType = itk::TriangleCell< CellType >;
TetrahedronType = itk::TetrahedronCell< CellType >;
Then, custom CellVisitor classes should be declared. The only requirement on the declaration of
each visitor class is to provide a method named Visit(). This method expects as arguments a cell
identifier and a pointer to the specific cell type for which this visitor is intended.
The following Vertex visitor simply prints out the identifier of the point with which the cell is
associated. Note that the cell uses the method GetPointId() without any arguments. This method
is only defined on the VertexCell.
4.3. Mesh
101
class CustomVertexVisitor
{
public:
void Visit(unsigned long cellId, VertexType * t )
{
std::cout << "cell " << cellId << " is a Vertex " << std::endl;
std::cout << "
associated with point id = ";
std::cout << t->GetPointId() << std::endl;
}
virtual ˜CustomVertexVisitor() = default;
};
注意：也就是在类型自定义一个visit
方法而不用去改变类的内部methods
The following Line visitor computes the length of the line. Note that this visitor is slightly more
complicated since it needs to get access to the actual mesh in order to get point coordinates from the
point identifiers returned by the line cell. This is done by holding a pointer to the mesh and querying
the mesh each time point coordinates are required. The mesh pointer is set up in this case with the
SetMesh() method.
class CustomLineVisitor
{
public:
CustomLineVisitor():m_Mesh( nullptr ) {}
virtual ˜CustomLineVisitor() = default;
void SetMesh( MeshType * mesh ) { m_Mesh = mesh; }
void Visit(unsigned long cellId, LineType * t )
{
std::cout << "cell " << cellId << " is a Line " << std::endl;
LineType::PointIdIterator pit = t->PointIdsBegin();
MeshType::PointType p0;
MeshType::PointType p1;
m_Mesh->GetPoint( *pit++, &p0 );
m_Mesh->GetPoint( *pit++, &p1 );
const double length = p0.EuclideanDistanceTo( p1 );
std::cout << " length = " << length << std::endl;
}
private:
MeshType::Pointer m_Mesh;
};
The Triangle visitor below prints out the identifiers of its points. Note the use of the
PointIdIterator and the PointIdsBegin() and PointIdsEnd() methods.
class CustomTriangleVisitor
{
public:
void Visit(unsigned long cellId, TriangleType * t )
{
102
Chapter 4. Data Representation
std::cout << "cell " << cellId << " is a Triangle " << std::endl;
LineType::PointIdIterator pit = t->PointIdsBegin();
LineType::PointIdIterator end = t->PointIdsEnd();
while( pit != end )
{
std::cout << " point id = " << *pit << std::endl;
++pit;
}
}
virtual ˜CustomTriangleVisitor() = default;
};
The TetrahedronVisitor below simply returns the number of faces on this figure.
GetNumberOfFaces() is a method exclusive of 3D cells.
Note that
class CustomTetrahedronVisitor
{
public:
void Visit(unsigned long cellId, TetrahedronType * t )
{
std::cout << "cell " << cellId << " is a Tetrahedron " << std::endl;
std::cout << " number of faces = ";
std::cout << t->GetNumberOfFaces() << std::endl;
}
virtual ˜CustomTetrahedronVisitor() = default;
};
With the cell visitors we proceed now to instantiate CellVisitor implementations. The visitor classes
defined above are used as template arguments of the cell visitor implementation.
using VertexVisitorInterfaceType = itk::CellInterfaceVisitorImplementation<
PixelType, MeshType::CellTraits, VertexType,
CustomVertexVisitor >;
using LineVisitorInterfaceType = itk::CellInterfaceVisitorImplementation<
PixelType, MeshType::CellTraits, LineType,
CustomLineVisitor >;
using TriangleVisitorInterfaceType = itk::CellInterfaceVisitorImplementation<
PixelType, MeshType::CellTraits, TriangleType,
CustomTriangleVisitor >;
using TetrahedronVisitorInterfaceType =
itk::CellInterfaceVisitorImplementation<
PixelType, MeshType::CellTraits, TetrahedronType,
CustomTetrahedronVisitor >;
Note that the actual CellInterfaceVisitorImplementation is templated over the PixelType, the
CellTraits, the CellType to be visited and the Visitor class defining what to do with the cell.
4.3. Mesh
103
A visitor implementation class can now be created using the normal invocation to its New() method
and assigning the result to a itk::SmartPointer.
VertexVisitorInterfaceType::Pointer vertexVisitor =
VertexVisitorInterfaceType::New();
LineVisitorInterfaceType::Pointer lineVisitor =
LineVisitorInterfaceType::New();
TriangleVisitorInterfaceType::Pointer triangleVisitor =
TriangleVisitorInterfaceType::New();
TetrahedronVisitorInterfaceType::Pointer tetrahedronVisitor =
TetrahedronVisitorInterfaceType::New();
Remember that the LineVisitor requires the pointer to the mesh object since it needs to get access to
actual point coordinates. This is done by invoking the SetMesh() method defined above.
lineVisitor->SetMesh( mesh );
Looking carefully you will notice that the SetMesh() method is declared in CustomLineVisitor
but we are invoking it on LineVisitorInterfaceType. This is possible thanks to the way in which
the VisitorInterfaceImplementation is defined. This class derives from the visitor type provided by
the user as the fourth template parameter. LineVisitorInterfaceType is then a derived class of
CustomLineVisitor.
The set of visitors should now be registered with the MultiVisitor class that will walk through the
cells and delegate action to every registered visitor when the appropriate cell type is encountered.
The following lines create a MultiVisitor object.
using CellMultiVisitorType = CellType::MultiVisitor;
CellMultiVisitorType::Pointer multiVisitor = CellMultiVisitorType::New();
Every visitor implementation is registered with the Mesh using the AddVisitor() method.
multiVisitor->AddVisitor(
multiVisitor->AddVisitor(
multiVisitor->AddVisitor(
multiVisitor->AddVisitor(
vertexVisitor
lineVisitor
triangleVisitor
tetrahedronVisitor
);
);
);
);
Finally, the iteration over the cells is triggered by calling the method Accept() on the Mesh class.
mesh->Accept( multiVisitor );
The Accept() method will iterate over all the cells and for each one will invite the MultiVisitor to
注意：
104
Chapter 4. Data Representation
attempt an action on the cell. If no visitor is interested on the current cell type, the cell is just ignored
and skipped.
4.4 Path
4.4.1 Creating a PolyLineParametricPath
The source code for this section can be found in the file
PolyLineParametricPath1.cxx.
This example illustrates how to use the itk::PolyLineParametricPath. This class will typically
be used for representing in a concise way the output of an image segmentation algorithm in 2D. The
PolyLineParametricPath however could also be used for representing any open or close curve in
N-Dimensions as a linear piece-wise approximation.
First, the header file of the PolyLineParametricPath class must be included.
#include "itkPolyLineParametricPath.h"
The path is instantiated over the dimension of the image. In this example the image and path are
two-dimensional.
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< unsigned char, Dimension >;
using PathType = itk::PolyLineParametricPath< Dimension >;
ImageType::ConstPointer image = reader->GetOutput();
PathType::Pointer path = PathType::New();
path->Initialize();
using ContinuousIndexType = PathType::ContinuousIndexType;
ContinuousIndexType cindex;
using ImagePointType = ImageType::PointType;
ImagePointType origin = image->GetOrigin();
ImageType::SpacingType spacing = image->GetSpacing();
ImageType::SizeType
size
= image->GetBufferedRegion().GetSize();
ImagePointType point;
point[0] = origin[0] + spacing[0] * size[0];
point[1] = origin[1] + spacing[1] * size[1];
4.5. Containers
105
image->TransformPhysicalPointToContinuousIndex( origin, cindex );
path->AddVertex( cindex );
image->TransformPhysicalPointToContinuousIndex( point, cindex );
path->AddVertex( cindex );
4.5 Containers
The source code for this section can be found in the file
TreeContainer.cxx.
注意：实现了树结构
This example demonstrates use of the itk::TreeContainer class and associated TreeIterators.
TreeContainer implements the notion of a tree, which is a branching data structure composed of
nodes and edges, where the edges indicate a parent/child relationship between nodes. Each node may
have exactly one parent, except for the root node, which has none. A tree must have exactly one root
node, and a node may not be its own parent. To round out the vocabulary used to discuss this data
structure, two nodes sharing the same parent node are called “siblings,” a childless node is termed a
“leaf,” and a “forest” is a collection of disjoint trees. Note that in the present implementation, it is
the user’s responsibility to enforce these relationships, as no checking is done to ensure a cycle-free
tree. TreeContainer is templated over the type of node, affording the user great flexibility in using
the structure for their particular problem.
Let’s begin by including the appropriate header files.
#include
#include
#include
#include
#include
#include
#include
#include
"itkTreeContainer.h"
"itkChildTreeIterator.h"
"itkLeafTreeIterator.h"
"itkLevelOrderTreeIterator.h"
"itkInOrderTreeIterator.h"
"itkPostOrderTreeIterator.h"
"itkRootTreeIterator.h"
"itkTreeIteratorClone.h"
We first instantiate a tree with int node type.
using NodeType = int;
using TreeType = itk::TreeContainer;
TreeType::Pointer tree = TreeType::New();
Next we set the value of the root node using SetRoot().
106
Chapter 4. Data Representation
tree->SetRoot(0);
Nodes may be added to the tree using the Add() method, where the first argument is the value of the
new node, and the second argument is the value of the parent node.
tree->Add(1,0);
tree->Add(2,0);
tree->Add(3,0);
tree->Add(4,2);
tree->Add(5,2);
tree->Add(6,5);
tree->Add(7,1);
If two nodes have the same value, it is ambiguous which node is intended to be the parent of the new
node; in this case, the first node with that value is selected. As will be demonstrated shortly, this
ambiguity can be avoided by constructing the tree with TreeIterators.
Let’s begin by defining a itk::ChildTreeIterator.
itk::ChildTreeIterator childIt(tree);
Before discussing the particular features of this iterator, however, we will illustrate features common to all TreeIterators, which inherit from itk::TreeIteratorBase. Basic use follows the
convention of other iterators in ITK, relying on the GoToBegin() and IsAtEnd() methods. The
iterator is advanced using the prefix increment ++ operator, whose behavior naturally depends on
the particular iterator being used.
for (childIt.GoToBegin(); !childIt.IsAtEnd(); ++childIt)
{
std::cout << childIt.Get() << std::endl;
}
std::cout << std::endl;
Note that, though not illustrated here, trees may also be traversed using the GoToParent() and
GoToChild() methods.
TreeIterators have a number of useful functions for testing properties of the current node. For
example, GetType() returns an enumerated type corresponding to the type of the particular iterator
being used. These types are as follows:
UNDEFIND, PREORDER, INORDER, POSTORDER, LEVELORDER, CHILD, ROOT, and LEAF.
In the following snippet, we test whether the iterator is of type CHILD, and return from the program
indicating failure if the test returns false.
4.5. Containers
107
if(childIt.GetType() != itk::TreeIteratorBase::CHILD)
{
std::cerr << "Error: The iterator was not of type CHILD." << std::endl;
return EXIT_FAILURE;
}
The value associated with the node can be retrieved and modified using Get() and Set() methods:
int oldValue = childIt.Get();
std::cout << "The node's value is " << oldValue << std::endl;
int newValue = 2;
childIt.Set(newValue);
std::cout << "Now, the node's value is " << childIt.Get() << std::endl;
A number of member functions are defined allowing the user to query information about the current
node’s parent/child relationships:
std::cout << "Is this a leaf node? " << childIt.IsLeaf() << std::endl;
std::cout << "Is this the root node? " << childIt.IsRoot() << std::endl;
std::cout << "Does this node have a parent? " << childIt.HasParent()
<< std::endl;
std::cout << "How many children does this node have? "
<< childIt.CountChildren() << std::endl;
std::cout << "Does this node have a child 1? " << childIt.HasChild(1)
<< std::endl;
In addition to traversing the tree and querying for information, TreeIterators can alter the structure of the tree itself. For example, a node can be added using the Add() methods, child nodes
can be removed using the RemoveChild() method, and the current node can be removed using
the Remove() method. Each of these methods returns a bool indicating whether the alteration was
successful.
To illustrate this, in the following snippet we clear the tree of all nodes, and then repopulate it using
the iterator.
tree->Clear();
itk::PreOrderTreeIterator it(tree);
it.GoToBegin();
it.Add(0);
it.Add(1);
it.Add(2);
it.Add(3);
it.GoToChild(2);
it.Add(4);
it.Add(5);
108
Chapter 4. Data Representation
Every TreeIterator has a Clone() function which returns a copy of the current iterator. Note that
the user should delete the created iterator by hand.
itk::TreeIteratorBase* childItClone = childIt.Clone();
delete childItClone;
Alternatively, itk::TreeIteratorClone can be used to create a generic copy of an iterator.
using IteratorType = itk::TreeIteratorBase;
using IteratorCloneType = itk::TreeIteratorClone;
IteratorCloneType anotherChildItClone = childIt;
We now turn our attention to features of the specific TreeIterator specializations.
ChildTreeIterator, for example, provides a way to iterate through all the children of a node.
for (childIt.GoToBegin(); !childIt.IsAtEnd(); ++childIt)
{
std::cout << childIt.Get();
}
std::cout << std::endl;
The itk::LeafTreeIterator iterates through the leaves of the tree.
itk::LeafTreeIterator leafIt(tree);
for (leafIt.GoToBegin(); !leafIt.IsAtEnd(); ++leafIt)
{
std::cout << leafIt.Get() << std::endl;
}
std::cout << std::endl;
itk::LevelOrderTreeIterator takes three arguments in its constructor: the tree to be traversed,
the maximum depth (or ‘level’), and the starting node. Naturally, this iterator provides a method for
returning the current level.
itk::LevelOrderTreeIterator levelIt(tree,10,tree->GetNode(0));
for (levelIt.GoToBegin(); !levelIt.IsAtEnd(); ++levelIt)
{
std::cout << levelIt.Get()
<< " ("<< levelIt.GetLevel() << ")"
<< std::endl;
}
std::cout << std::endl;
itk::InOrderTreeIterator iterates through the tree from left to right.
4.5. Containers
109
itk::InOrderTreeIterator inOrderIt(tree);
for (inOrderIt.GoToBegin(); !inOrderIt.IsAtEnd(); ++inOrderIt)
{
std::cout << inOrderIt.Get() << std::endl;
}
std::cout << std::endl;
注意：树的前序以及后序遍历
itk::PreOrderTreeIterator iterates through the tree from left to right but do a depth first search.
itk::PreOrderTreeIterator preOrderIt(tree);
for (preOrderIt.GoToBegin(); !preOrderIt.IsAtEnd(); ++preOrderIt)
{
std::cout << preOrderIt.Get() << std::endl;
}
std::cout << std::endl;
The itk::PostOrderTreeIterator iterates through the tree from left to right but goes from the
leaves to the root in the search.
itk::PostOrderTreeIterator postOrderIt(tree);
for (postOrderIt.GoToBegin(); !postOrderIt.IsAtEnd(); ++postOrderIt)
{
std::cout << postOrderIt.Get() << std::endl;
}
std::cout << std::endl;
The itk::RootTreeIterator goes from one node to the root. The second arguments is the starting
node. Here we go from the leaf node (value = 6) up to the root.
itk::RootTreeIterator rootIt(tree,tree->GetNode(4));
for (rootIt.GoToBegin(); !rootIt.IsAtEnd(); ++rootIt)
{
std::cout << rootIt.Get() << std::endl;
}
std::cout << std::endl;
CHAPTER
FIVE
SPATIAL OBJECTS
This chapter introduces the basic classes that describe itk::SpatialObjects.
5.1 Introduction
We promote the philosophy that many of the goals of medical image processing are more effectively
addressed if we consider them in the broader context of object processing. ITK’s Spatial Object
class hierarchy provides a consistent API for querying, manipulating, and interconnecting objects
in physical space. Via this API, methods can be coded to be invariant to the data structure used
to store the objects being processed. By abstracting the representations of objects to support their
representation by data structures other than images, a broad range of medical image analysis research
注意：类型适应性比较强？
is supported; key examples are described in the following.
Model-to-image registration. A mathematical instance of an object can be registered with an image to localize the instance of that object in the image. Using SpatialObjects, mutual information, cross-correlation, and boundary-to-image metrics can be applied without modification to
perform spatial object-to-image registration.
Model-to-model registration. Iterative closest point, landmark, and surface distance minimization
methods can be used with any ITK transform, to rigidly and non-rigidly register image, FEM,
and Fourier descriptor-based representations of objects as SpatialObjects.
Atlas formation. Collections of images or SpatialObjects can be integrated to represent expected
object characteristics and their common modes of variation. Labels can be associated with the
objects of an atlas.
Storing segmentation results from one or multiple scans. Results of segmentations are best
stored in physical/world coordinates so that they can be combined and compared with other
segmentations from other images taken at other resolutions. Segmentation results from hand
drawn contours, pixel labelings, or model-to-image registrations are treated consistently.
112
Chapter 5. Spatial Objects
注意：附属关系
Capturing functional and logical relationships between objects. SpatialObjects can have parent
and children objects. Queries made of an object (such as to determine if a point is inside of
the object) can be made to integrate the responses from the children object. Transformations
applied to a parent can also be propagated to the children. Thus, for example, when a liver
model is moved, its vessels move with it.
Conversion to and from images. Basic functions are provided to render any SpatialObject (or collection of SpatialObjects) into an image.
IO. SpatialObject reading and writing to disk is independent of the SpatialObject class hierarchy.
Meta object IO (through itk::MetaImageIO) methods are provided, and others are easily
defined.
Tubes, blobs, images, surfaces. Are a few of the many SpatialObject data containers and types
provided. New types can be added, generally by only defining one or two member functions
in a derived class.
In the remainder of this chapter several examples are used to demonstrate the many spatial objects
found in ITK and how they can be organized into hierarchies using itk::SceneSpatialObject.
Further the examples illustrate how to use SpatialObject transformations to control and calculate the
position of objects in space.
5.2 Hierarchy
注意：可以将空间对象组合成tree
Spatial objects can be combined to form a hierarchy as a tree. By design, a SpatialObject can
have one parent and only one. Moreover, each transform is stored within each object, therefore the
hierarchy cannot be described as a Directed Acyclic Graph (DAG) but effectively as a tree. The user
is responsible for maintaining the tree structure, no checking is done to ensure a cycle-free tree.
The source code for this section can be found in the file
SpatialObjectHierarchy.cxx.
注意：用户负责维护树形结构
This example describes how itk::SpatialObject can form a hierarchy. This first example also
shows how to create and manipulate spatial objects.
#include "itkSpatialObject.h"
First, we create two spatial objects and give them the names First Object and Second Object,
respectively.
using SpatialObjectType = itk::SpatialObject<3>;
SpatialObjectType::Pointer object1 = SpatialObjectType ::New();
object1->GetProperty()->SetName("First Object");
5.2. Hierarchy
113
SpatialObjectType::Pointer object2 = SpatialObjectType ::New();
object2->GetProperty()->SetName("Second Object");
We then add the second object to the first one by using the AddSpatialObject() method. As a
result object2 becomes a child of object1.
object1->AddSpatialObject(object2);
We can query if an object has a parent by using the HasParent() method. If it has one, the
GetParent() method returns a constant pointer to the parent. In our case, if we ask the parent’s
name of the object2 we should obtain: First Object.
if(object2->HasParent())
{
std::cout << "Name of the parent of the object2: ";
std::cout << object2->GetParent()->GetProperty()->GetName() << std::endl;
}
注意：使用树形结构对空间对象进行
组织
To access the list of children of the object, the GetChildren() method returns a pointer to the (STL)
list of children.
SpatialObjectType::ChildrenListType * childrenList = object1->GetChildren();
std::cout << "object1 has " << childrenList->size() << " child" << std::endl;
SpatialObjectType::ChildrenListType::const_iterator it
= childrenList->begin();
while(it != childrenList->end())
{
std::cout << "Name of the child of the object 1: ";
std::cout << (*it)->GetProperty()->GetName() << std::endl;
++it;
}
Do NOT forget to delete the list of children since the GetChildren() function creates an internal
list.
delete childrenList;
An object can also be removed by using the RemoveSpatialObject() method.
object1->RemoveSpatialObject(object2);
We can query the number of children an object has with the GetNumberOfChildren() method.
114
Chapter 5. Spatial Objects
std::cout << "Number of children for object1: ";
std::cout << object1->GetNumberOfChildren() << std::endl;
The Clear() method erases all the information regarding the object as well as the data. This method
is usually overloaded by derived classes.
object1->Clear();
The output of this first example looks like the following:
Name of the parent of the object2: First Object
object1 has 1 child
Name of the child of the object 1: Second Object
Number of children for object1: 0
5.3 SpatialObject Tree Container
The source code for this section can be found in the file
SpatialObjectTreeContainer.cxx.
This example describes how to use the itk::SpatialObjectTreeContainer to form a hierarchy
of SpatialObjects. First we include the appropriate header file.
#include "itkSpatialObjectTreeContainer.h"
Next we define the type of node and the type of tree we plan to use. Both are templated over the
dimensionality of the space. Let’s create a 2-dimensional tree.
using NodeType = itk::GroupSpatialObject< 2 >;
using TreeType = itk::SpatialObjectTreeContainer< 2 >;
Then, we can create three nodes and set their corresponding identification numbers (using SetId).
NodeType::Pointer object0 = NodeType::New();
object0->SetId(0);
NodeType::Pointer object1 = NodeType::New();
object1->SetId(1);
NodeType::Pointer object2 = NodeType::New();
object2->SetId(2);
The hierarchy is formed using the AddSpatialObject() function.
5.4. Transformations
115
object0->AddSpatialObject(object1);
object1->AddSpatialObject(object2);
After instantiation of the tree we set its root using the SetRoot() function.
TreeType::Pointer tree = TreeType::New();
tree->SetRoot(object0);
The tree iterators described in a previous section of this guide can be used to parse the hierarchy. For
example, via an itk::LevelOrderTreeIterator templated over the type of tree, we can parse the
hierarchy of SpatialObjects. We set the maximum level to 10 which is enough in this case since our
hierarchy is only 2 deep.
注意：深度优先遍历？？
itk::LevelOrderTreeIterator levelIt(tree,10);
levelIt.GoToBegin();
while(!levelIt.IsAtEnd())
{
std::cout << levelIt.Get()->GetId() << " ("<< levelIt.GetLevel()
<< ")" << std::endl;
++levelIt;
}
Tree iterators can also be used to add spatial objects to the hierarchy. Here we show how to use the
itk::PreOrderTreeIterator to add a fourth object to the tree.
NodeType::Pointer object4 = NodeType::New();
itk::PreOrderTreeIterator preIt( tree );
preIt.Add(object4);
5.4 Transformations
The source code for this section can be found in the file
SpatialObjectTransforms.cxx.
This example describes the different transformations associated with a spatial object.
Figure 5.1 shows our set of transformations.
Like the first example, we create two spatial objects and give them the names First Object and
Second Object, respectively.
116
Chapter 5. Spatial Objects
IndexToWorld
Transform
World
ObjectToWorld
Transform
Parent Node
ObjectToParent
Transform
NodeToParentNode
Transform
Node
ObjectToNode
Transform
IndexToObject
Transform
Object
Index
Figure 5.1: Set of transformations associated with a Spatial Object
using SpatialObjectType = itk::SpatialObject<2>;
using TransformType = SpatialObjectType::TransformType;
SpatialObjectType::Pointer object1 = SpatialObjectType ::New();
object1->GetProperty()->SetName("First Object");
SpatialObjectType::Pointer object2 = SpatialObjectType ::New();
object2->GetProperty()->SetName("Second Object");
object1->AddSpatialObject(object2);
注意：空间对象维护的三个内置转换
Instances of itk::SpatialObject maintain three transformations internally that can be used to
compute the position and orientation of data and objects. These transformations are: an IndexToObjectTransform, an ObjectToParentTransform, and an ObjectToWorldTransform. As a convenience
to the user, the global transformation IndexToWorldTransform and its inverse, WorldToIndexTransform, are also maintained by the class. Methods are provided by SpatialObject to access and manipulate these transforms.
The two main transformations, IndexToObjectTransform and ObjectToParentTransform, are applied
successively. ObjectToParentTransform is applied to children.
The IndexToObjectTransform transforms points from the internal data coordinate system of the
object (typically the indices of the image from which the object was defined) to “physical” space
(which accounts for the spacing, orientation, and offset of the indices).
注意：图像索引-->（根据
spacing、orientation、offset
of the indices）物理位置
5.4. Transformations
117
The ObjectToParentTransform transforms points from the object-specific “physical” space to the
“physical” space of its parent object. As one can see from the figure 5.1, the ObjectToParentTransform is composed of two transforms: ObjectToNodeTransform and NodeToParentNodeTransform.
The ObjectToNodeTransform is not applied to the children, but the NodeToParentNodeTransform is.
Therefore, if one sets the ObjectToParentTransform, the NodeToParentNodeTransform is actually
set.
The ObjectToWorldTransform maps points from the reference system of the SpatialObject into the
global coordinate system. This is useful when the position of the object is known only in the global
coordinate frame. Note that by setting this transform, the ObjectToParent transform is recomputed.
These transformations use the itk::FixedCenterOfRotationAffineTransform. They are created in the constructor of the spatial itk::SpatialObject.
First we define an index scaling factor of 2 for the object2. This is done by setting the Scale of the
IndexToObjectTransform.
double scale[2];
scale[0]=2;
scale[1]=2;
object2->GetIndexToObjectTransform()->SetScale(scale);
Next, we apply an offset on the ObjectToParentTransform of the child object. Therefore, object2
is now translated by a vector [4,3] regarding to its parent.
TransformType::OffsetType Object2ToObject1Offset;
Object2ToObject1Offset[0] = 4;
Object2ToObject1Offset[1] = 3;
object2->GetObjectToParentTransform()->SetOffset(Object2ToObject1Offset);
To realize the previous operations on the transformations, we should invoke the
ComputeObjectToWorldTransform() that recomputes all dependent transformations.
object2->ComputeObjectToWorldTransform();
We can now display the ObjectToWorldTransform for both objects. One should notice that the
FixedCenterOfRotationAffineTransform derives from itk::AffineTransform and therefore the
only valid members of the transformation are a Matrix and an Offset. For instance, when we invoke
the Scale() method the internal Matrix is recomputed to reflect this change.
The FixedCenterOfRotationAffineTransform performs the following computation
X ′ = R · (S · X − C) + C + V
(5.1)
Where R is the rotation matrix, S is a scaling factor, C is the center of rotation and V is a translation
118
Chapter 5. Spatial Objects
vector or offset. Therefore the affine matrix M and the affine offset T are defined as:
M = R·S
(5.2)
T = C + V − R ·C
(5.3)
This means that GetScale() and GetOffset() as well as the GetMatrix() might not be set to the
expected value, especially if the transformation results from a composition with another transformation since the composition is done using the Matrix and the Offset of the affine transformation.
Next, we show the two affine transformations corresponding to the two objects.
std::cout
std::cout
std::cout
std::cout
std::cout
std::cout
std::cout
std::cout
<<
<<
<<
<<
<<
<<
<<
<<
"object2 IndexToObject Matrix: " << std::endl;
object2->GetIndexToObjectTransform()->GetMatrix() << std::endl;
"object2 IndexToObject Offset: ";
object2->GetIndexToObjectTransform()->GetOffset() << std::endl;
"object2 IndexToWorld Matrix: " << std::endl;
object2->GetIndexToWorldTransform()->GetMatrix() << std::endl;
"object2 IndexToWorld Offset: ";
object2->GetIndexToWorldTransform()->GetOffset() << std::endl;
Then, we decide to translate the first object which is the parent of the second by a vector [3,3].
This is still done by setting the offset of the ObjectToParentTransform. This can also be done by
setting the ObjectToWorldTransform because the first object does not have any parent and therefore
is attached to the world coordinate frame.
TransformType::OffsetType Object1ToWorldOffset;
Object1ToWorldOffset[0] = 3;
Object1ToWorldOffset[1] = 3;
object1->GetObjectToParentTransform()->SetOffset(Object1ToWorldOffset);
注意：子节点的形变转换依赖于父节
点的形变转换？？
Next we invoke ComputeObjectToWorldTransform() on the modified object. This will propagate
the transformation through all its children.
object1->ComputeObjectToWorldTransform();
Figure 5.2 shows our set of transformations.
Finally, we display the resulting affine transformations.
std::cout
std::cout
std::cout
std::cout
std::cout
std::cout
<<
<<
<<
<<
<<
<<
"object1 IndexToWorld Matrix: " << std::endl;
object1->GetIndexToWorldTransform()->GetMatrix() << std::endl;
"object1 IndexToWorld Offset: ";
object1->GetIndexToWorldTransform()->GetOffset() << std::endl;
"object2 IndexToWorld Matrix: " << std::endl;
object2->GetIndexToWorldTransform()->GetMatrix() << std::endl;
5.5. Types of Spatial Objects
119
7
6
5
Object 2
4
3
2
Object 1
1
1
2
3
4
5
6
7
8
Figure 5.2: Physical positions of the two objects in the world frame (shapes are merely for illustration purposes).
std::cout << "object2 IndexToWorld Offset: ";
std::cout << object2->GetIndexToWorldTransform()->GetOffset() << std::endl;
The output of this second example looks like the following:
object2
2 0
0 2
object2
object2
2 0
0 2
object2
object1
1 0
0 1
object1
object2
2 0
0 2
object2
IndexToObject Matrix:
IndexToObject Offset: 0
IndexToWorld Matrix:
0
IndexToWorld Offset: 4
IndexToWorld Matrix:
3
IndexToWorld Offset: 3
IndexToWorld Matrix:
3
IndexToWorld Offset: 7
6
5.5 Types of Spatial Objects
This section describes in detail the variety of spatial objects implemented in ITK.
120
Chapter 5. Spatial Objects
5.5.1 ArrowSpatialObject
The source code for this section can be found in the file
ArrowSpatialObject.cxx.
This example shows how to create an itk::ArrowSpatialObject. Let’s begin by including the
appropriate header file.
#include "itkArrowSpatialObject.h"
The itk::ArrowSpatialObject, like many SpatialObjects, is templated over the dimensionality
of the object.
using ArrowType = itk::ArrowSpatialObject<3>;
ArrowType::Pointer myArrow = ArrowType::New();
The length of the arrow in the local coordinate frame is done using the SetLength() method. By
default the length is set to 1.
myArrow->SetLength(2);
The direction of the arrow can be set using the SetDirection() method. Calling SetDirection()
modifies the ObjectToParentTransform (not the IndexToObjectTransform). By default the
direction is set along the X axis (first direction).
ArrowType::VectorType direction;
direction.Fill(0);
direction[1] = 1.0;
myArrow->SetDirection(direction);
5.5.2 BlobSpatialObject
The source code for this section can be found in the file
BlobSpatialObject.cxx.
itk::BlobSpatialObject defines an N-dimensional blob. Like other SpatialObjects this class
derives from itk::itkSpatialObject. A blob is defined as a list of points which compose the
object.
Let’s start by including the appropriate header file.
5.5. Types of Spatial Objects
121
#include "itkBlobSpatialObject.h"
BlobSpatialObject is templated over the dimension of the space. A BlobSpatialObject contains a list
of SpatialObjectPoints. Basically, a SpatialObjectPoint has a position and a color.
#include "itkSpatialObjectPoint.h"
First we declare some type definitions.
using BlobType = itk::BlobSpatialObject<3>;
using BlobPointer = BlobType::Pointer;
using BlobPointType = itk::SpatialObjectPoint<3>;
Then, we create a list of points and we set the position of each point in the local coordinate system
using the SetPosition() method. We also set the color of each point to be red.
BlobType::PointListType list;
for( unsigned int i=0; i<4; i++)
{
BlobPointType p;
p.SetPosition(i,i+1,i+2);
p.SetRed(1);
p.SetGreen(0);
p.SetBlue(0);
p.SetAlpha(1.0);
list.push_back(p);
}
Next, we create the blob and set its name using the SetName() function. We also set its Identification
number with SetId() and we add the list of points previously created.
BlobPointer blob = BlobType::New();
blob->GetProperty()->SetName("My Blob");
blob->SetId(1);
blob->SetPoints(list);
The GetPoints() method returns a reference to the internal list of points of the object.
BlobType::PointListType pointList = blob->GetPoints();
std::cout << "The blob contains " << pointList.size();
std::cout << " points" << std::endl;
Then we can access the points using standard STL iterators and GetPosition() and GetColor()
functions return respectively the position and the color of the point.
122
Chapter 5. Spatial Objects
BlobType::PointListType::const_iterator it = blob->GetPoints().begin();
while(it != blob->GetPoints().end())
{
std::cout << "Position = " << (*it).GetPosition() << std::endl;
std::cout << "Color = " << (*it).GetColor() << std::endl;
++it;
}
5.5.3 CylinderSpatialObject
The source code for this section can be found in the file
CylinderSpatialObject.cxx.
This example shows how to create a itk::CylinderSpatialObject. Let’s begin by including the
appropriate header file.
#include "itkCylinderSpatialObject.h"
An itk::CylinderSpatialObject exists only in 3D, therefore, it is not templated.
using CylinderType = itk::CylinderSpatialObject;
We create a cylinder using the standard smart pointers.
CylinderType::Pointer myCylinder = CylinderType::New();
The radius of the cylinder is set using the SetRadius() function. By default the radius is set to 1.
double radius = 3.0;
myCylinder->SetRadius(radius);
The height of the cylinder is set using the SetHeight() function. By default the cylinder is defined
along the X axis (first dimension).
double height = 12.0;
myCylinder->SetHeight(height);
Like any other itk::SpatialObjects, the IsInside() function can be used to query if a point is
inside or outside the cylinder.
5.5. Types of Spatial Objects
123
itk::Point insidePoint;
insidePoint[0]=1;
insidePoint[1]=2;
insidePoint[2]=0;
std::cout << "Is my point "<< insidePoint << " inside the cylinder? : "
<< myCylinder->IsInside(insidePoint) << std::endl;
We can print the cylinder information using the Print() function.
myCylinder->Print(std::cout);
5.5.4 EllipseSpatialObject
The source code for this section can be found in the file
EllipseSpatialObject.cxx.
itk::EllipseSpatialObject defines an n-Dimensional ellipse. Like other spatial objects this
class derives from itk::SpatialObject. Let’s start by including the appropriate header file.
#include "itkEllipseSpatialObject.h"
Like most of the SpatialObjects, the itk::EllipseSpatialObject is templated over the dimension of the space. In this example we create a 3-dimensional ellipse.
using EllipseType = itk::EllipseSpatialObject<3>;
EllipseType::Pointer myEllipse = EllipseType::New();
Then we set a radius for each dimension. By default the radius is set to 1.
EllipseType::ArrayType radius;
for (unsigned int i = 0; i<3; ++i)
{
radius[i] = i;
}
myEllipse->SetRadius(radius);
Or if we have the same radius in each dimension we can do
myEllipse->SetRadius(2.0);
124
Chapter 5. Spatial Objects
We can then display the current radius by using the GetRadius() function:
EllipseType::ArrayType myCurrentRadius = myEllipse->GetRadius();
std::cout << "Current radius is " << myCurrentRadius << std::endl;
注意：查询点是否在对象内部
Like other SpatialObjects, we can query the object if a point is inside the object by using the IsInside(itk::Point) function. This function expects the point to be in world coordinates.
itk::Point insidePoint;
insidePoint.Fill(1.0);
if (myEllipse->IsInside(insidePoint))
{
std::cout << "The point " << insidePoint;
std::cout << " is really inside the ellipse" << std::endl;
}
itk::Point outsidePoint;
outsidePoint.Fill(3.0);
if (!myEllipse->IsInside(outsidePoint))
{
std::cout << "The point " << outsidePoint;
std::cout << " is really outside the ellipse" << std::endl;
}
All spatial objects can be queried for a value at a point. The IsEvaluableAt() function returns a
boolean to know if the object is evaluable at a particular point.
if (myEllipse->IsEvaluableAt(insidePoint))
{
std::cout << "The point " << insidePoint;
std::cout << " is evaluable at the point " << insidePoint << std::endl;
}
If the object is evaluable at that point, the ValueAt() function returns the current value at that
position. Most of the objects returns a boolean value which is set to true when the point is inside
the object and false when it is outside. However, for some objects, it is more interesting to return a
value representing, for instance, the distance from the center of the object or the distance from from
the boundary.
double value;
myEllipse->ValueAt(insidePoint,value);
std::cout << "The value inside the ellipse is: " << value << std::endl;
Like other spatial objects, we can also query the bounding box of the object by using
GetBoundingBox(). The resulting bounding box is expressed in the local frame.
5.5. Types of Spatial Objects
125
myEllipse->ComputeBoundingBox();
EllipseType::BoundingBoxType * boundingBox = myEllipse->GetBoundingBox();
std::cout << "Bounding Box: " << boundingBox->GetBounds() << std::endl;
5.5.5 GaussianSpatialObject
The source code for this section can be found in the file
GaussianSpatialObject.cxx.
This example shows how to create a itk::GaussianSpatialObject which defines a Gaussian in
a N-dimensional space. This object is particularly useful to query the value at a point in physical
space. Let’s begin by including the appropriate header file.
注意：
#include "itkGaussianSpatialObject.h"
The itk::GaussianSpatialObject is templated over the dimensionality of the object.
using GaussianType = itk::GaussianSpatialObject<3>;
GaussianType::Pointer myGaussian = GaussianType::New();
The SetMaximum() function is used to set the maximum value of the Gaussian.
myGaussian->SetMaximum(2);
The radius of the Gaussian is defined by the SetRadius() method. By default the radius is set to
1.0.
myGaussian->SetRadius(3);
The standard ValueAt() function is used to determine the value of the Gaussian at a particular point
in physical space.
itk::Point pt;
pt[0]=1;
pt[1]=2;
pt[2]=1;
double value;
myGaussian->ValueAt(pt, value);
std::cout << "ValueAt(" << pt << ") = " << value << std::endl;
126
Chapter 5. Spatial Objects
5.5.6 GroupSpatialObject
The source code for this section can be found in the file
GroupSpatialObject.cxx.
注意：该类的作用
A itk::GroupSpatialObject does not have any data associated with it. It can be used to group
objects or to add transforms to a current object. In this example we show how to use a GroupSpatialObject.
Let’s begin by including the appropriate header file.
#include "itkGroupSpatialObject.h"
The itk::GroupSpatialObject is templated over the dimensionality of the object.
using GroupType = itk::GroupSpatialObject<3>;
GroupType::Pointer myGroup = GroupType::New();
Next, we create an itk::EllipseSpatialObject and add it to the group.
using EllipseType = itk::EllipseSpatialObject<3>;
EllipseType::Pointer myEllipse = EllipseType::New();
myEllipse->SetRadius(2);
myGroup->AddSpatialObject(myEllipse);
We then translate the group by 10mm in each direction. Therefore the ellipse is translated in physical
space at the same time.
GroupType::VectorType offset;
offset.Fill(10);
myGroup->GetObjectToParentTransform()->SetOffset(offset);
myGroup->ComputeObjectToWorldTransform();
注意：可以统一执行某个操作？？
We can then query if a point is inside the group using the IsInside() function. We need to specify
in this case that we want to consider all the hierarchy, therefore we set the depth to 2.
GroupType::PointType point;
point.Fill(10);
std::cout << "Is my point " << point << " inside?: "
<< myGroup->IsInside(point,2) << std::endl;
Like any other SpatialObjects we can remove the ellipse from the group using the
RemoveSpatialObject() method.
5.5. Types of Spatial Objects
127
myGroup->RemoveSpatialObject(myEllipse);
5.5.7 ImageSpatialObject
The source code for this section can be found in the file
ImageSpatialObject.cxx.
An itk::ImageSpatialObject contains an itk::Image but adds the notion of spatial transformations and parent-child hierarchy. Let’s begin the next example by including the appropriate
header file.
#include "itkImageSpatialObject.h"
We first create a simple 2D image of size 10 by 10 pixels.
using Image = itk::Image;
Image::Pointer image = Image::New();
Image::SizeType size = {{ 10, 10 }};
Image::RegionType region;
region.SetSize(size);
image->SetRegions(region);
image->Allocate();
Next we fill the image with increasing values.
using Iterator = itk::ImageRegionIterator;
Iterator it(image,region);
short pixelValue =0;
for(it.GoToBegin(); !it.IsAtEnd(); ++it, ++pixelValue)
{
it.Set(pixelValue);
}
We can now define the ImageSpatialObject which is templated over the dimension and the pixel type
of the image.
using ImageSpatialObject = itk::ImageSpatialObject<2,short>;
ImageSpatialObject::Pointer imageSO = ImageSpatialObject::New();
Then we set the itkImage to the ImageSpatialObject by using the SetImage() function.
128
Chapter 5. Spatial Objects
imageSO->SetImage(image);
At this point we can use IsInside(), ValueAt() and DerivativeAt() functions inherent in SpatialObjects. The IsInside() value can be useful when dealing with registration.
using Point = itk::Point;
Point insidePoint;
insidePoint.Fill(9);
if( imageSO->IsInside(insidePoint) )
{
std::cout << insidePoint << " is inside the image." << std::endl;
}
The ValueAt() returns the value of the closest pixel, i.e no interpolation, to a given physical point.
double returnedValue;
imageSO->ValueAt(insidePoint,returnedValue);
std::cout << "ValueAt(" << insidePoint << ") = " << returnedValue
<< std::endl;
注意：计算特定点的梯度
The derivative at a specified position in space can be computed using the DerivativeAt() function.
The first argument is the point in physical coordinates where we are evaluating the derivatives. The
second argument is the order of the derivation, and the third argument is the result expressed as a
itk::Vector. Derivatives are computed iteratively using finite differences and, like the ValueAt(),
no interpolator is used.
注意：梯度的计算使用的是有限差分
ImageSpatialObject::OutputVectorType returnedDerivative;
imageSO->DerivativeAt(insidePoint,1,returnedDerivative);
std::cout << "First derivative at " << insidePoint;
std::cout << " = " << returnedDerivative << std::endl;
5.5.8 ImageMaskSpatialObject
The source code for this section can be found in the file
ImageMaskSpatialObject.cxx.
An itk::ImageMaskSpatialObject is similar to the itk::ImageSpatialObject and derived
from it. However, the main difference is that the IsInside() returns true if the pixel intensity in
the image is not zero.
The supported pixel types does not include itk::RGBPixel, itk::RGBAPixel, etc. So far it only
allows to manage images of simple types like unsigned short, unsigned int, or itk::Vector. Let’s
begin by including the appropriate header file.
5.5. Types of Spatial Objects
129
#include "itkImageMaskSpatialObject.h"
The ImageMaskSpatialObject is templated over the dimensionality.
using ImageMaskSpatialObject = itk::ImageMaskSpatialObject<3>;
Next we create an itk::Image of size 50x50x50 filled with zeros except a bright square in the
middle which defines the mask.
using PixelType = ImageMaskSpatialObject::PixelType;
using ImageType = ImageMaskSpatialObject::ImageType;
using Iterator = itk::ImageRegionIterator< ImageType >;
ImageType::Pointer image = ImageType::New();
ImageType::SizeType size = {{ 50, 50, 50 }};
ImageType::IndexType index = {{ 0, 0, 0 }};
ImageType::RegionType region;
region.SetSize(size);
region.SetIndex(index);
image->SetRegions( region );
image->Allocate(true); // initialize buffer to zero
ImageType::RegionType insideRegion;
ImageType::SizeType insideSize
= {{ 30, 30, 30 }};
ImageType::IndexType insideIndex = {{ 10, 10, 10 }};
insideRegion.SetSize( insideSize );
insideRegion.SetIndex( insideIndex );
Iterator it( image, insideRegion );
it.GoToBegin();
while( !it.IsAtEnd() )
{
it.Set( itk::NumericTraits< PixelType >::max() );
++it;
}
Then, we create an ImageMaskSpatialObject.
ImageMaskSpatialObject::Pointer maskSO = ImageMaskSpatialObject::New();
We then pass the corresponding pointer to the image.
maskSO->SetImage(image);
130
Chapter 5. Spatial Objects
We can then test if a physical itk::Point is inside or outside the mask image. This is particularly
useful during the registration process when only a part of the image should be used to compute the
注意：ROI
metric.
ImageMaskSpatialObject::PointType inside;
inside.Fill(20);
std::cout << "Is my point " << inside << " inside my mask? "
<< maskSO->IsInside(inside) << std::endl;
ImageMaskSpatialObject::PointType outside;
outside.Fill(45);
std::cout << "Is my point " << outside << " outside my mask? "
<< !maskSO->IsInside(outside) << std::endl;
5.5.9 LandmarkSpatialObject
The source code for this section can be found in the file
LandmarkSpatialObject.cxx.
itk::LandmarkSpatialObject contains a list of itk::SpatialObjectPoints which have a position and a color. Let’s begin this example by including the appropriate header file.
#include "itkLandmarkSpatialObject.h"
LandmarkSpatialObject is templated over the dimension of the space.
Here we create a 3-dimensional landmark.
using LandmarkType = itk::LandmarkSpatialObject<3>;
using LandmarkPointer = LandmarkType::Pointer;
using LandmarkPointType = itk::SpatialObjectPoint<3>;
LandmarkPointer landmark = LandmarkType::New();
Next, we set some properties of the object like its name and its identification number.
landmark->GetProperty()->SetName("Landmark1");
landmark->SetId(1);
We are now ready to add points into the landmark. We first create a list of SpatialObjectPoint and
for each point we set the position and the color.
LandmarkType::PointListType list;
5.5. Types of Spatial Objects
131
for (unsigned int i=0; i<5; ++i)
{
LandmarkPointType p;
p.SetPosition(i,i+1,i+2);
p.SetColor(1,0,0,1);
list.push_back(p);
}
Then we add the list to the object using the SetPoints() method.
landmark->SetPoints(list);
The current point list can be accessed using the GetPoints() method. The method returns a reference to the (STL) list.
size_t nPoints = landmark->GetPoints().size();
std::cout << "Number of Points in the landmark: " << nPoints << std::endl;
LandmarkType::PointListType::const_iterator it
= landmark->GetPoints().begin();
while(it != landmark->GetPoints().end())
{
std::cout << "Position: " << (*it).GetPosition() << std::endl;
std::cout << "Color: " << (*it).GetColor() << std::endl;
++it;
}
5.5.10 LineSpatialObject
The source code for this section can be found in the file
LineSpatialObject.cxx.
itk::LineSpatialObject defines a line in an n-dimensional space. A line is defined as a list of
points which compose the line, i.e a polyline. We begin the example by including the appropriate
header files.
#include "itkLineSpatialObject.h"
LineSpatialObject is templated over the dimension of the space. A LineSpatialObject contains a list
of LineSpatialObjectPoints. A LineSpatialObjectPoint has a position, n − 1 normals and a color.
Each normal is expressed as a itk::CovariantVector of size N.
First, we define some type definitions and we create our line.
132
Chapter 5. Spatial Objects
using
using
using
using
LineType = itk::LineSpatialObject<3>;
LinePointer = LineType::Pointer;
LinePointType = itk::LineSpatialObjectPoint<3>;
VectorType = itk::CovariantVector;
LinePointer Line = LineType::New();
We create a point list and we set the position of each point in the local coordinate system using the
SetPosition() method. We also set the color of each point to red.
The two normals are set using the SetNormal() function; the first argument is the normal itself and
the second argument is the index of the normal.
LineType::PointListType list;
for (unsigned int i=0; i<3; ++i)
{
LinePointType p;
p.SetPosition(i,i+1,i+2);
p.SetColor(1,0,0,1);
VectorType normal1;
VectorType normal2;
for (unsigned int j=0; j<3; ++j)
{
normal1[j]=j;
normal2[j]=j*2;
}
p.SetNormal(normal1,0);
p.SetNormal(normal2,1);
list.push_back(p);
}
Next, we set the name of the object using SetName(). We also set its identification number with
SetId() and we set the list of points previously created.
Line->GetProperty()->SetName("Line1");
Line->SetId(1);
Line->SetPoints(list);
The GetPoints() method returns a reference to the internal list of points of the object.
LineType::PointListType pointList = Line->GetPoints();
std::cout << "Number of points representing the line: ";
std::cout << pointList.size() << std::endl;
Then we can access the points using standard STL iterators. The GetPosition() and GetColor()
5.5. Types of Spatial Objects
133
functions return respectively the position and the color of the point. Using the GetNormal(unsigned
int) function we can access each normal.
LineType::PointListType::const_iterator it = Line->GetPoints().begin();
while (it != Line->GetPoints().end())
{
std::cout << "Position = " << (*it).GetPosition() << std::endl;
std::cout << "Color = " << (*it).GetColor() << std::endl;
std::cout << "First normal = " << (*it).GetNormal(0) << std::endl;
std::cout << "Second normal = " << (*it).GetNormal(1) << std::endl;
std::cout << std::endl;
++it;
}
5.5.11 MeshSpatialObject
The source code for this section can be found in the file
MeshSpatialObject.cxx.
A itk::MeshSpatialObject contains a pointer to an itk::Mesh but adds the notion of
spatial transformations and parent-child hierarchy. This example shows how to create an
itk::MeshSpatialObject, use it to form a binary image, and write the mesh to disk.
Let’s begin by including the appropriate header file.
#include
#include
#include
#include
"itkSpatialObjectToImageFilter.h"
"itkMeshSpatialObject.h"
"itkSpatialObjectReader.h"
"itkSpatialObjectWriter.h"
The MeshSpatialObject wraps an itk::Mesh, therefore we first create a mesh.
using
using
using
using
using
using
using
using
MeshTrait = itk::DefaultDynamicMeshTraits< float, 3, 3 >;
MeshType = itk::Mesh< float, 3, MeshTrait >;
CellTraits = MeshType::CellTraits;
CellInterfaceType = itk::CellInterface< float, CellTraits >;
TetraCellType = itk::TetrahedronCell< CellInterfaceType >;
PointType = MeshType::PointType;
CellType = MeshType::CellType;
CellAutoPointer = CellType::CellAutoPointer;
MeshType::Pointer myMesh = MeshType::New();
MeshType::CoordRepType testPointCoords[4][3]
注意：？？
134
Chapter 5. Spatial Objects
= { {0,0,0}, {9,0,0}, {9,9,0}, {0,0,9} };
MeshType::PointIdentifier tetraPoints[4] = {0,1,2,4};
int i;
for(i=0; i < 4; ++i)
{
myMesh->SetPoint(i, PointType(testPointCoords[i]));
}
myMesh->SetCellsAllocationMethod(
MeshType::CellsAllocatedDynamicallyCellByCell );
CellAutoPointer testCell1;
testCell1.TakeOwnership( new TetraCellType );
testCell1->SetPointIds(tetraPoints);
myMesh->SetCell(0, testCell1 );
We then create a MeshSpatialObject which is templated over the type of mesh previously defined...
using MeshSpatialObjectType = itk::MeshSpatialObject< MeshType >;
MeshSpatialObjectType::Pointer myMeshSpatialObject =
MeshSpatialObjectType::New();
... and pass the Mesh pointer to the MeshSpatialObject
myMeshSpatialObject->SetMesh(myMesh);
The actual pointer to the passed mesh can be retrieved using the GetMesh() function, just like any
other SpatialObjects.
myMeshSpatialObject->GetMesh();
The GetBoundingBox(), ValueAt(), IsInside() functions can be used to access important information.
std::cout << "Mesh bounds : " <<
myMeshSpatialObject->GetBoundingBox()->GetBounds() << std::endl;
MeshSpatialObjectType::PointType myPhysicalPoint;
myPhysicalPoint.Fill(1);
std::cout << "Is my physical point inside? : " <<
myMeshSpatialObject->IsInside(myPhysicalPoint) << std::endl;
5.5. Types of Spatial Objects
135
Now that we have defined the MeshSpatialObject, we can save the actual mesh using the
itk::SpatialObjectWriter. In order to do so, we need to specify the type of Mesh we are
writing.
using WriterType = itk::SpatialObjectWriter< 3, float, MeshTrait >;
WriterType::Pointer writer = WriterType::New();
Then we set the mesh spatial object and the name of the file and call the the Update() function.
writer->SetInput(myMeshSpatialObject);
writer->SetFileName("myMesh.meta");
writer->Update();
Reading the saved mesh is done using the itk::SpatialObjectReader. Once again we need to
specify the type of mesh we intend to read.
using ReaderType = itk::SpatialObjectReader< 3, float, MeshTrait >;
ReaderType::Pointer reader = ReaderType::New();
We set the name of the file we want to read and call update
reader->SetFileName("myMesh.meta");
reader->Update();
Next, we show how to create a binary image of a MeshSpatialObject using the
itk::SpatialObjectToImageFilter. The resulting image will have ones inside and zeros outside
注意：mesh转换为二值图像，mesh内
the mesh. First we define and instantiate the SpatialObjectToImageFilter.
部全为1，外部全为0 --有用--
using ImageType = itk::Image< unsigned char, 3 >;
using GroupType = itk::GroupSpatialObject< 3 >;
using SpatialObjectToImageFilterType =
itk::SpatialObjectToImageFilter< GroupType, ImageType >;
SpatialObjectToImageFilterType::Pointer imageFilter =
SpatialObjectToImageFilterType::New();
Then we pass the output of the reader, i.e the MeshSpatialObject, to the filter.
imageFilter->SetInput( reader->GetGroup() );
Finally we trigger the execution of the filter by calling the Update() method. Note that depending
on the size of the mesh, the computation time can increase significantly.
注意：时间效率低下
136
Chapter 5. Spatial Objects
imageFilter->Update();
Then we can get the resulting binary image using the GetOutput() function.
ImageType::Pointer myBinaryMeshImage = imageFilter->GetOutput();
5.5.12 SurfaceSpatialObject
The source code for this section can be found in the file
SurfaceSpatialObject.cxx.
itk::SurfaceSpatialObject defines a surface in n-dimensional space. A SurfaceSpatialObject
is defined by a list of points which lie on the surface. Each point has a position and a unique normal.
The example begins by including the appropriate header file.
#include "itkSurfaceSpatialObject.h"
SurfaceSpatialObject is templated over the dimension of the space. A SurfaceSpatialObject contains
a list of SurfaceSpatialObjectPoints. A SurfaceSpatialObjectPoint has a position, a normal and a
color.
First we define some type definitions
using
using
using
using
SurfaceType = itk::SurfaceSpatialObject<3>;
SurfacePointer = SurfaceType::Pointer;
SurfacePointType = itk::SurfaceSpatialObjectPoint<3>;
VectorType = itk::CovariantVector;
SurfacePointer Surface = SurfaceType::New();
We create a point list and we set the position of each point in the local coordinate system using the
SetPosition() method. We also set the color of each point to red.
SurfaceType::PointListType list;
for( unsigned int i=0; i<3; i++)
{
SurfacePointType p;
p.SetPosition(i,i+1,i+2);
p.SetColor(1,0,0,1);
VectorType normal;
for(unsigned int j=0;j<3;j++)
{
5.5. Types of Spatial Objects
137
normal[j]=j;
}
p.SetNormal(normal);
list.push_back(p);
}
Next, we create the surface and set his name using SetName(). We also set its Identification number
with SetId() and we add the list of points previously created.
Surface->GetProperty()->SetName("Surface1");
Surface->SetId(1);
Surface->SetPoints(list);
The GetPoints() method returns a reference to the internal list of points of the object.
SurfaceType::PointListType pointList = Surface->GetPoints();
std::cout << "Number of points representing the surface: ";
std::cout << pointList.size() << std::endl;
Then we can access the points using standard STL iterators. GetPosition() and GetColor()
functions return respectively the position and the color of the point. GetNormal() returns the normal
as a itk::CovariantVector.
SurfaceType::PointListType::const_iterator it
= Surface->GetPoints().begin();
while(it != Surface->GetPoints().end())
{
std::cout << "Position = " << (*it).GetPosition() << std::endl;
std::cout << "Normal = " << (*it).GetNormal() << std::endl;
std::cout << "Color = " << (*it).GetColor() << std::endl;
std::cout << std::endl;
it++;
}
5.5.13 TubeSpatialObject
注意：表示管状结构
itk::TubeSpatialObject represents a base class for the representation of tubular
structures using SpatialObjects.
The classes
itk::VesselTubeSpatialObject and
itk::DTITubeSpatialObject derive from this base class. VesselTubeSpatialObject represents blood vessels extracted for an image and DTITubeSpatialObject is used to represent fiber
tracts from diffusion tensor images.
The source code for this section can be found in the file
TubeSpatialObject.cxx.
138
Chapter 5. Spatial Objects
itk::TubeSpatialObject defines an n-dimensional tube. A tube is defined as a list of centerline
points which have a position, a radius, some normals and other properties. Let’s start by including
the appropriate header file.
#include "itkTubeSpatialObject.h"
TubeSpatialObject is templated over the dimension of the space. A TubeSpatialObject contains a
list of TubeSpatialObjectPoints.
First we define some type definitions and we create the tube.
using
using
using
using
TubeType = itk::TubeSpatialObject<3>;
TubePointer = TubeType::Pointer;
TubePointType = itk::TubeSpatialObjectPoint<3>;
VectorType = TubePointType::CovariantVectorType;
TubePointer tube = TubeType::New();
We create a point list and we set:
1. The position of each point in the local coordinate system using the SetPosition() method.
2. The radius of the tube at this position using SetRadius().
3. The two normals at the tube is set using SetNormal1() and SetNormal2().
4. The color of the point is set to red in our case.
TubeType::PointListType list;
for (i=0; i<5; ++i)
{
TubePointType p;
p.SetPosition(i,i+1,i+2);
p.SetRadius(1);
VectorType normal1;
VectorType normal2;
for (unsigned int j=0; j<3; ++j)
{
normal1[j]=j;
normal2[j]=j*2;
}
p.SetNormal1(normal1);
p.SetNormal2(normal2);
p.SetColor(1,0,0,1);
list.push_back(p);
}
5.5. Types of Spatial Objects
139
Next, we create the tube and set its name using SetName(). We also set its identification number
with SetId() and, at the end, we add the list of points previously created.
tube->GetProperty()->SetName("Tube1");
tube->SetId(1);
tube->SetPoints(list);
The GetPoints() method return a reference to the internal list of points of the object.
TubeType::PointListType pointList = tube->GetPoints();
std::cout << "Number of points representing the tube: ";
std::cout << pointList.size() << std::endl;
The ComputeTangentAndNormals() function computes the normals and the tangent for each point
using finite differences.
tube->ComputeTangentAndNormals();
Then we can access the points using STL iterators. GetPosition() and GetColor() functions return respectively the position and the color of the point. GetRadius() returns the radius at that point.
GetNormal1() and GetNormal1() functions return a itk::CovariantVector and GetTangent()
returns a itk::Vector.
TubeType::PointListType::const_iterator it = tube->GetPoints().begin();
i=0;
while(it != tube->GetPoints().end())
{
std::cout << std::endl;
std::cout << "Point #" << i << std::endl;
std::cout << "Position: " << (*it).GetPosition() << std::endl;
std::cout << "Radius: " << (*it).GetRadius() << std::endl;
std::cout << "Tangent: " << (*it).GetTangent() << std::endl;
std::cout << "First Normal: " << (*it).GetNormal1() << std::endl;
std::cout << "Second Normal: " << (*it).GetNormal2() << std::endl;
std::cout << "Color = " << (*it).GetColor() << std::endl;
it++;
i++;
}
VesselTubeSpatialObject
The source code for this section can be found in the file
VesselTubeSpatialObject.cxx.
140
Chapter 5. Spatial Objects
itk::VesselTubeSpatialObject derives from itk::TubeSpatialObject. It represents a blood
vessel segmented from an image. A VesselTubeSpatialObject is described as a list of centerline
points which have a position, a radius, and normals.
Let’s start by including the appropriate header file.
#include "itkVesselTubeSpatialObject.h"
VesselTubeSpatialObject is templated over the dimension of the space. A VesselTubeSpatialObject
contains a list of VesselTubeSpatialObjectPoints.
First we define some type definitions and we create the tube.
using VesselTubeType = itk::VesselTubeSpatialObject<3>;
using VesselTubePointType = itk::VesselTubeSpatialObjectPoint<3>;
VesselTubeType::Pointer VesselTube = VesselTubeType::New();
We create a point list and we set:
1. The position of each point in the local coordinate system using the SetPosition() method.
2. The radius of the tube at this position using SetRadius().
3. The medialness value describing how the point lies in the middle of the vessel using
SetMedialness().
4. The ridgeness value describing how the point lies on the ridge using SetRidgeness().
5. The branchness value describing if the point is a branch point using SetBranchness().
6. The three alpha values corresponding to the eigenvalues of the Hessian using
注意：与管状结构相关的Hessian矩
SetAlpha1(),SetAlpha2() and SetAlpha3().
阵
7. The mark value using SetMark().
8. The color of the point is set to red in this example with an opacity of 1.
VesselTubeType::PointListType list;
for (i=0; i<5; ++i)
{
VesselTubePointType p;
p.SetPosition(i,i+1,i+2);
p.SetRadius(1);
p.SetAlpha1(i);
p.SetAlpha2(i+1);
p.SetAlpha3(i+2);
p.SetMedialness(i);
5.5. Types of Spatial Objects
141
p.SetRidgeness(i);
p.SetBranchness(i);
p.SetMark(true);
p.SetColor(1,0,0,1);
list.push_back(p);
}
Next, we create the tube and set its name using SetName(). We also set its identification number
with SetId() and, at the end, we add the list of points previously created.
VesselTube->GetProperty()->SetName("VesselTube");
VesselTube->SetId(1);
VesselTube->SetPoints(list);
The GetPoints() method return a reference to the internal list of points of the object.
VesselTubeType::PointListType pointList = VesselTube->GetPoints();
std::cout << "Number of points representing the blood vessel: ";
std::cout << pointList.size() << std::endl;
Then we can access the points using STL iterators. GetPosition() and GetColor() functions
return respectively the position and the color of the point.
VesselTubeType::PointListType::const_iterator
it = VesselTube->GetPoints().begin();
i=0;
while(it != VesselTube->GetPoints().end())
{
std::cout << std::endl;
std::cout << "Point #" << i << std::endl;
std::cout << "Position: " << (*it).GetPosition() << std::endl;
std::cout << "Radius: " << (*it).GetRadius() << std::endl;
std::cout << "Medialness: " << (*it).GetMedialness() << std::endl;
std::cout << "Ridgeness: " << (*it).GetRidgeness() << std::endl;
std::cout << "Branchness: " << (*it).GetBranchness() << std::endl;
std::cout << "Mark: " << (*it).GetMark() << std::endl;
std::cout << "Alpha1: " << (*it).GetAlpha1() << std::endl;
std::cout << "Alpha2: " << (*it).GetAlpha2() << std::endl;
std::cout << "Alpha3: " << (*it).GetAlpha3() << std::endl;
std::cout << "Color = " << (*it).GetColor() << std::endl;
++it;
++i;
}
142
Chapter 5. Spatial Objects
DTITubeSpatialObject
The source code for this section can be found in the file
DTITubeSpatialObject.cxx.
itk::DTITubeSpatialObject derives from itk::TubeSpatialObject. It represents a fiber
tracts from Diffusion Tensor Imaging. A DTITubeSpatialObject is described as a list of centerline points which have a position, a radius, normals, the fractional anisotropy (FA) value, the ADC
value, the geodesic anisotropy (GA) value, the eigenvalues and vectors as well as the full tensor
matrix.
Let’s start by including the appropriate header file.
#include "itkDTITubeSpatialObject.h"
DTITubeSpatialObject is templated over the dimension of the space. A DTITubeSpatialObject contains a list of DTITubeSpatialObjectPoints.
First we define some type definitions and we create the tube.
using DTITubeType = itk::DTITubeSpatialObject<3>;
using DTITubePointType = itk::DTITubeSpatialObjectPoint<3>;
DTITubeType::Pointer dtiTube = DTITubeType::New();
We create a point list and we set:
1. The position of each point in the local coordinate system using the SetPosition() method.
2. The radius of the tube at this position using SetRadius().
3. The FA value using AddField(DTITubePointType::FA).
4. The ADC value using AddField(DTITubePointType::ADC).
5. The GA value using AddField(DTITubePointType::GA).
6. The full tensor matrix supposed to be symmetric definite positive value using
SetTensorMatrix().
7. The color of the point is set to red in our case.
DTITubeType::PointListType list;
for (i=0; i<5; ++i)
{
DTITubePointType p;
p.SetPosition(i,i+1,i+2);
5.5. Types of Spatial Objects
143
p.SetRadius(1);
p.AddField(DTITubePointType::FA,i);
p.AddField(DTITubePointType::ADC,2*i);
p.AddField(DTITubePointType::GA,3*i);
p.AddField("Lambda1",4*i);
p.AddField("Lambda2",5*i);
p.AddField("Lambda3",6*i);
auto * v = new float[6];
for(unsigned int k=0;k<6;k++)
{
v[k] = k;
}
p.SetTensorMatrix(v);
delete[] v;
p.SetColor(1,0,0,1);
list.push_back(p);
}
Next, we create the tube and set its name using SetName(). We also set its identification number
with SetId() and, at the end, we add the list of points previously created.
dtiTube->GetProperty()->SetName("DTITube");
dtiTube->SetId(1);
dtiTube->SetPoints(list);
The GetPoints() method return a reference to the internal list of points of the object.
DTITubeType::PointListType pointList = dtiTube->GetPoints();
std::cout << "Number of points representing the fiber tract: ";
std::cout << pointList.size() << std::endl;
Then we can access the points using STL iterators. GetPosition() and GetColor() functions
return respectively the position and the color of the point.
DTITubeType::PointListType::const_iterator it = dtiTube->GetPoints().begin();
i=0;
while(it != dtiTube->GetPoints().end())
{
std::cout << std::endl;
std::cout << "Point #" << i << std::endl;
std::cout << "Position: " << (*it).GetPosition() << std::endl;
std::cout << "Radius: " << (*it).GetRadius() << std::endl;
std::cout << "FA: " << (*it).GetField(DTITubePointType::FA) << std::endl;
std::cout << "ADC: " << (*it).GetField(DTITubePointType::ADC) << std::endl;
std::cout << "GA: " << (*it).GetField(DTITubePointType::GA) << std::endl;
std::cout << "Lambda1: " << (*it).GetField("Lambda1") << std::endl;
std::cout << "Lambda2: " << (*it).GetField("Lambda2") << std::endl;
std::cout << "Lambda3: " << (*it).GetField("Lambda3") << std::endl;
std::cout << "TensorMatrix: " << (*it).GetTensorMatrix()[0] << " : ";
144
Chapter 5. Spatial Objects
std::cout
std::cout
std::cout
std::cout
std::cout
std::cout
++it;
++i;
}
<<
<<
<<
<<
<<
<<
(*it).GetTensorMatrix()[1] << " : ";
(*it).GetTensorMatrix()[2] << " : ";
(*it).GetTensorMatrix()[3] << " : ";
(*it).GetTensorMatrix()[4] << " : ";
(*it).GetTensorMatrix()[5] << std::endl;
"Color = " << (*it).GetColor() << std::endl;
5.6 SceneSpatialObject
The source code for this section can be found in the file
SceneSpatialObject.cxx.
This example describes how to use the itk::SceneSpatialObject. A SceneSpatialObject contains a collection of SpatialObjects. This example begins by including the appropriate header file.
#include "itkSceneSpatialObject.h"
An SceneSpatialObject is templated over the dimension of the space which requires all the objects
referenced by the SceneSpatialObject to have the same dimension.
First we define some type definitions and we create the SceneSpatialObject.
using SceneSpatialObjectType = itk::SceneSpatialObject<3>;
SceneSpatialObjectType::Pointer scene = SceneSpatialObjectType::New();
Then we create two itk::EllipseSpatialObjects.
using EllipseType = itk::EllipseSpatialObject<3>;
EllipseType::Pointer ellipse1 = EllipseType::New();
ellipse1->SetRadius(1);
ellipse1->SetId(1);
EllipseType::Pointer ellipse2 = EllipseType::New();
ellipse2->SetId(2);
ellipse2->SetRadius(2);
Then we add the two ellipses into the SceneSpatialObject.
scene->AddSpatialObject(ellipse1);
scene->AddSpatialObject(ellipse2);
5.6. SceneSpatialObject
145
We can query the number of object in the SceneSpatialObject with the GetNumberOfObjects()
function. This function takes two optional arguments: the depth at which we should count the
number of objects (default is set to infinity) and the name of the object to count (default is set to
ITK NULLPTR). This allows the user to count, for example, only ellipses.
std::cout << "Number of objects in the SceneSpatialObject = ";
std::cout << scene->GetNumberOfObjects() << std::endl;
The GetObjectById() returns the first object in the SceneSpatialObject that has the specified identification number.
std::cout << "Object in the SceneSpatialObject with an ID == 2: "
<< std::endl;
scene->GetObjectById(2)->Print(std::cout);
Objects can also be removed from the SceneSpatialObject using the RemoveSpatialObject()
function.
scene->RemoveSpatialObject(ellipse1);
The list of current objects in the SceneSpatialObject can be retrieved using the GetObjects()
method. Like the GetNumberOfObjects() method, GetObjects() can take two arguments: a
search depth and a matching name.
SceneSpatialObjectType::ObjectListType * myObjectList = scene->GetObjects();
std::cout << "Number of objects in the SceneSpatialObject = ";
std::cout << myObjectList->size() << std::endl;
In some cases, it is useful to define the hierarchy by using ParentId() and the current identification
number. This results in having a flat list of SpatialObjects in the SceneSpatialObject. Therefore,
the SceneSpatialObject provides the FixHierarchy() method which reorganizes the Parent-Child
hierarchy based on identification numbers.
scene->FixHierarchy();
The scene can also be cleared by using the Clear() function.
scene->Clear();
146
Chapter 5. Spatial Objects
5.7 Read/Write SpatialObjects
The source code for this section can be found in the file
ReadWriteSpatialObject.cxx.
Reading and writing SpatialObjects is a fairly simple task.
The classes
itk::SpatialObjectReader and itk::SpatialObjectWriter are used to read and write
these objects, respectively. (Note these classes make use of the MetaIO auxiliary I/O routines and
therefore have a .meta file suffix.)
We begin this example by including the appropriate header files.
#include "itkSpatialObjectReader.h"
#include "itkSpatialObjectWriter.h"
#include "itkEllipseSpatialObject.h"
Next, we create a SpatialObjectWriter that is templated over the dimension of the object(s) we want
to write.
using WriterType = itk::SpatialObjectWriter<3>;
WriterType::Pointer writer = WriterType::New();
For this example, we create an itk::EllipseSpatialObject.
using EllipseType = itk::EllipseSpatialObject<3>;
EllipseType::Pointer ellipse = EllipseType::New();
ellipse->SetRadius(3);
Finally, we set to the writer the object to write using the SetInput() method and we set the name
of the file with SetFileName() and call the Update() method to actually write the information.
writer->SetInput(ellipse);
writer->SetFileName("ellipse.meta");
writer->Update();
Now we are ready to open the freshly created object. We first create a SpatialObjectReader which
is also templated over the dimension of the object in the file. This means that the file should contain
注意：读取文件中的objects维度与
only objects with the same dimension.
定义的reader中的objects维度应一
致
using ReaderType = itk::SpatialObjectReader<3>;
ReaderType::Pointer reader = ReaderType::New();
Next we set the name of the file to read using SetFileName() and we call the Update() method to
read the file.
5.8. Statistics Computation via SpatialObjects
147
reader->SetFileName("ellipse.meta");
reader->Update();
To get the objects in the file you can call the GetScene() method or the GetGroup() method.
GetScene() returns an pointer to a itk::SceneSpatialObject.
ReaderType::SceneType * scene = reader->GetScene();
std::cout << "Number of objects in the scene: ";
std::cout << scene->GetNumberOfObjects() << std::endl;
ReaderType::GroupType * group = reader->GetGroup();
std::cout << "Number of objects in the group: ";
std::cout << group->GetNumberOfChildren() << std::endl;
5.8 Statistics Computation via SpatialObjects
The source code for this section can be found in the file
SpatialObjectToImageStatisticsCalculator.cxx.
This example describes how to use the itk::SpatialObjectToImageStatisticsCalculator to
compute statistics of an itk::Image only in a region defined inside a given itk::SpatialObject.
#include "itkSpatialObjectToImageStatisticsCalculator.h"
We first create a test image using the itk::RandomImageSource
using ImageType = itk::Image< unsigned char, 2 >;
using RandomImageSourceType = itk::RandomImageSource< ImageType >;
RandomImageSourceType::Pointer randomImageSource
= RandomImageSourceType::New();
ImageType::SizeValueType size[2];
size[0] = 10;
size[1] = 10;
randomImageSource->SetSize(size);
randomImageSource->Update();
ImageType::Pointer image = randomImageSource->GetOutput();
Next we create an itk::EllipseSpatialObject with a radius of 2. We also move the ellipse to
the center of the image by increasing the offset of the IndexToObjectTransform.
using EllipseType = itk::EllipseSpatialObject<2>;
EllipseType::Pointer ellipse = EllipseType::New();
ellipse->SetRadius(2);
148
Chapter 5. Spatial Objects
EllipseType::VectorType offset;
offset.Fill(5);
ellipse->GetIndexToObjectTransform()->SetOffset(offset);
ellipse->ComputeObjectToParentTransform();
Then we can create the itk::SpatialObjectToImageStatisticsCalculator.
using CalculatorType = itk::SpatialObjectToImageStatisticsCalculator<
ImageType, EllipseType >;
CalculatorType::Pointer calculator = CalculatorType::New();
We pass a pointer to the image to the calculator.
calculator->SetImage(image);
注意：统计图像处于空间对象区域内
的图像信息（交集？？）
We also pass the SpatialObject. The statistics will be computed inside the SpatialObject (Internally
the calculator is using the IsInside() function).
calculator->SetSpatialObject(ellipse);
At the end we trigger the computation via the Update() function and we can retrieve the mean and
the covariance matrix using GetMean() and GetCovarianceMatrix() respectively.
calculator->Update();
std::cout << "Sample mean = " << calculator->GetMean() << std::endl;
std::cout << "Sample covariance = " << calculator->GetCovarianceMatrix();
CHAPTER
SIX
ITERATORS
This chapter introduces the image iterator, an important generic programming construct for image
processing in ITK. An iterator is a generalization of the familiar C programming language pointer
used to reference data in memory. ITK has a wide variety of image iterators, some of which are
highly specialized to simplify common image processing tasks.
The next section is a brief introduction that defines iterators in the context of ITK. Section 6.2 describes the programming interface common to most ITK image iterators. Sections 6.3–6.4 document
specific ITK iterator types and provide examples of how they are used.
6.1 Introduction
Generic programming models define functionally independent components called containers and algorithms. Container objects store data and algorithms operate on data. To access data in containers,
algorithms use a third class of objects called iterators. An iterator is an abstraction of a memory
pointer. Every container type must define its own iterator type, but all iterators are written to provide a common interface so that algorithm code can reference data in a generic way and maintain
functional independence from containers.
The iterator is so named because it is used for iterative, sequential access of container values. Iterators appear in for and while loop constructs, visiting each data point in turn. A C pointer, for
example, is a type of iterator. It can be moved forward (incremented) and backward (decremented)
through memory to sequentially reference elements of an array. Many iterator implementations have
an interface similar to a C pointer.
In ITK we use iterators to write generic image processing code for images instantiated with different
combinations of pixel type, pixel container type, and dimensionality. Because ITK image iterators
are specifically designed to work with image containers, their interface and implementation is optimized for image processing tasks. Using the ITK iterators instead of accessing data directly through
the itk::Image interface has many advantages. Code is more compact and often generalizes automatically to higher dimensions, algorithms run much faster, and iterators simplify tasks such as
注意：
150
Chapter 6. Iterators
multithreading and neighborhood-based image processing.
6.2 Programming Interface
This section describes the standard ITK image iterator programming interface. Some specialized
image iterators may deviate from this standard or provide additional methods.
6.2.1 Creating Iterators
All image iterators have at least one template parameter that is the image type over which they
iterate. There is no restriction on the dimensionality of the image or on the pixel type of the image.
注意：限定遍历的范围
An iterator constructor requires at least two arguments, a smart pointer to the image to iterate across,
and an image region. The image region, called the iteration region, is a rectilinear area in which iteration is constrained. The iteration region must be wholly contained within the image. More specifically, a valid iteration region is any subregion of the image within the current BufferedRegion.
See Section 4.1 for more information on image regions.
There is a const and a non-const version of most ITK image iterators. A non-const iterator cannot be
instantiated on a non-const image pointer. Const versions of iterators may read, but may not write
pixel values.
Here is a simple example that defines and constructs a simple image iterator for an itk::Image.
using ImageType = itk::Image;
using ConstIteratorType = itk::ImageRegionConstIterator< ImageType >;
using IteratorType = itk::ImageRegionIterator< ImageType >;
ImageType::Pointer image = SomeFilter->GetOutput();
ConstIteratorType constIterator( image, image->GetRequestedRegion() );
IteratorType iterator( image, image->GetRequestedRegion() );
6.2.2 Moving Iterators
An iterator is described as walking its iteration region. At any time, the iterator will reference, or
“point to”, one pixel location in the N-dimensional (ND) image. Forward iteration goes from the
beginning of the iteration region to the end of the iteration region. Reverse iteration, goes from just
past the end of the region back to the beginning. There are two corresponding starting positions for
iterators, the begin position and the end position. An iterator can be moved directly to either of these
two positions using the following methods.
• GoToBegin() Points the iterator to the first valid data element in the region.
6.2. Programming Interface
151
itk::Image
BEGIN Position
Iteration region
END Position
Figure 6.1: Normal path of an iterator through a 2D image. The iteration region is shown in a darker shade. An
arrow denotes a single iterator step, the result of one ++ operation.
注意：最后一个有效元素位置的后一
个位置
• GoToEnd() Points the iterator to one position past the last valid element in the region.
Note that the end position is not actually located within the iteration region. This is important
to remember because attempting to dereference an iterator at its end position will have undefined
results.
ITK iterators are moved back and forth across their iterations using the decrement and increment
operators.
• operator++() Increments the iterator one position in the positive direction. Only the
prefix increment operator is defined for ITK image iterators.
• operator--() Decrements the iterator one position in the negative direction. Only the
prefix decrement operator is defined for ITK image iterators.
Figure 6.1 illustrates typical iteration over an image region. Most iterators increment and decrement
in the direction of the fastest increasing image dimension, wrapping to the first position in the next
higher dimension at region boundaries. In other words, an iterator first moves across columns, then
注意：迭代器遍历图像区域
down rows, then from slice to slice, and so on.
时的顺序
In addition to sequential iteration through the image, some iterators may define random access operators. Unlike the increment operators, random access operators may not be optimized for speed and
require some knowledge of the dimensionality of the image and the extent of the iteration region to
use properly.
• operator+=( OffsetType ) Moves the iterator to the pixel position at the current index plus specified itk::Offset.
152
Chapter 6. Iterators
• operator-=( OffsetType ) Moves the iterator to the pixel position at the current index minus specified Offset.
• SetPosition( IndexType ) Moves the iterator to the given itk::Index position.
注意：通常慢
The SetPosition() method may be extremely slow for more complicated iterator types. In general,
it should only be used for setting a starting iteration position, like you would use GoToBegin() or
GoToEnd().
Some iterators do not follow a predictable path through their iteration regions and have no fixed beginning or ending pixel locations. A conditional iterator, for example, visits pixels only if they have
certain values or connectivities. Random iterators, increment and decrement to random locations
and may even visit a given pixel location more than once.
An iterator can be queried to determine if it is at the end or the beginning of its iteration region.
• bool IsAtEnd() True if the iterator points to one position past the end of the iteration
region.
• bool IsAtBegin() True if the iterator points to the first position in the iteration region.
The method is typically used to test for the end of reverse iteration.
An iterator can also report its current image index position.
• IndexType GetIndex() Returns the Index of the image pixel that the iterator currently
points to.
For efficiency, most ITK image iterators do not perform bounds checking. It is possible to move an
iterator outside of its valid iteration region. Dereferencing an out-of-bounds iterator will produce
undefined results.
6.2.3 Accessing Data
ITK image iterators define two basic methods for reading and writing pixel values.
• PixelType Get() Returns the value of the pixel at the iterator position.
• void Set( PixelType ) Sets the value of the pixel at the iterator position. Not defined
for const versions of iterators.
The Get() and Set() methods are inlined and optimized for speed so that their use is equivalent
to dereferencing the image buffer directly. There are a few common cases, however, where using
Get() and Set() do incur a penalty. Consider the following code, which fetches, modifies, and then
writes a value back to the same pixel location.
6.2. Programming Interface
153
it.Set( it.Get() + 1 );
As written, this code requires one more memory dereference than is necessary. Some iterators define
注意：一下函数可以避免出现两次解
a third data access method that avoids this penalty.
引用
• PixelType &Value() Returns a reference to the pixel at the iterator position.
The Value() method can be used as either an lval or an rval in an expression. It has all the properties
of operator*. The Value() method makes it possible to rewrite our example code more efficiently.
it.Value()++;
Consider using the Value() method instead of Get() or Set() when a call to operator= on a
pixel is non-trivial, such as when working with vector pixels, and operations are done in-place in the
image. The disadvantage of using Value is that it cannot support image adapters (see Section 7 on
page 191 for more information about image adaptors).
6.2.4 Iteration Loops
Using the methods described in the previous sections, we can now write a simple example to do
pixel-wise operations on an image. The following code calculates the squares of all values in an
input image and writes them to an output image.
ConstIteratorType in( inputImage,
inputImage->GetRequestedRegion() );
IteratorType out( outputImage, inputImage->GetRequestedRegion() );
for ( in.GoToBegin(), out.GoToBegin(); !in.IsAtEnd(); ++in, ++out )
{
out.Set( in.Get() * in.Get() );
}
Notice that both the input and output iterators are initialized over the same region, the
RequestedRegion of inputImage. This is good practice because it ensures that the output iterator walks exactly the same set of pixel indices as the input iterator, but does not require that the
output and input be the same size. The only requirement is that the input image must contain a
region (a starting index and size) that matches the RequestedRegion of the output image.
Equivalent code can be written by iterating through the image in reverse. The syntax is slightly more
awkward because the end of the iteration region is not a valid position and we can only test whether
the iterator is strictly equal to its beginning position. It is often more convenient to write reverse
iteration in a while loop.
154
Chapter 6. Iterators
in.GoToEnd();
out.GoToEnd();
while ( ! in.IsAtBegin() )
{
--in;
--out;
out.Set( in.Get() * in.Get() );
}
6.3 Image Iterators
This section describes iterators that walk rectilinear image regions and reference a single pixel at a
time. The itk::ImageRegionIterator is the most basic ITK image iterator and the first choice for
most applications. The rest of the iterators in this section are specializations of ImageRegionIterator
that are designed make common image processing tasks more efficient or easier to implement.
6.3.1 ImageRegionIterator
The source code for this section can be found in the file
ImageRegionIterator.cxx.
The itk::ImageRegionIterator is optimized for iteration speed and is the first choice for iterative, pixel-wise operations when location in the image is not important. ImageRegionIterator is the
least specialized of the ITK image iterator classes. It implements all of the methods described in the
preceding section.
The following example illustrates the use of itk::ImageRegionConstIterator and ImageRegionIterator. Most of the code constructs introduced apply to other ITK iterators as well. This
simple application crops a subregion from an image by copying its pixel values into to a second,
smaller image.
We begin by including the appropriate header files.
#include "itkImageRegionIterator.h"
Next we define a pixel type and corresponding image type. ITK iterator classes expect the image
type as their template parameter.
constexpr unsigned int Dimension = 2;
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, Dimension >;
6.3. Image Iterators
155
using ConstIteratorType = itk::ImageRegionConstIterator< ImageType >;
using IteratorType = itk::ImageRegionIterator< ImageType>;
Information about the subregion to copy is read from the command line. The subregion is defined
by an itk::ImageRegion object, with a starting grid index and a size (Section 4.1).
ImageType::RegionType inputRegion;
ImageType::RegionType::IndexType inputStart;
ImageType::RegionType::SizeType size;
inputStart[0] = ::std::stoi( argv[3] );
inputStart[1] = ::std::stoi( argv[4] );
size[0]
size[1]
= ::std::stoi( argv[5] );
= ::std::stoi( argv[6] );
inputRegion.SetSize( size );
inputRegion.SetIndex( inputStart );
The destination region in the output image is defined using the input region size, but a different start
index. The starting index for the destination region is the corner of the newly generated image.
ImageType::RegionType outputRegion;
ImageType::RegionType::IndexType outputStart;
outputStart[0] = 0;
outputStart[1] = 0;
outputRegion.SetSize( size );
outputRegion.SetIndex( outputStart );
After reading the input image and checking that the desired subregion is, in fact, contained in the
input, we allocate an output image. It is fundamental to set valid values to some of the basic image
information during the copying process. In particular, the starting index of the output region is now
filled up with zero values and the coordinates of the physical origin are computed as a shift from the
origin of the input image. This is quite important since it will allow us to later register the extracted
region against the original image.
ImageType::Pointer outputImage = ImageType::New();
outputImage->SetRegions( outputRegion );
const ImageType::SpacingType& spacing = reader->GetOutput()->GetSpacing();
const ImageType::PointType& inputOrigin = reader->GetOutput()->GetOrigin();
double
outputOrigin[ Dimension ];
for(unsigned int i=0; i< Dimension; i++)
{
156
Chapter 6. Iterators
outputOrigin[i] = inputOrigin[i] + spacing[i] * inputStart[i];
}
outputImage->SetSpacing( spacing );
outputImage->SetOrigin( outputOrigin );
outputImage->Allocate();
The necessary images and region definitions are now in place. All that is left to do is to create the
iterators and perform the copy. Note that image iterators are not accessed via smart pointers so they
are light-weight objects that are instantiated on the stack. Also notice how the input and output
iterators are defined over the same corresponding region. Though the images are different sizes,
they both contain the same target subregion.
ConstIteratorType inputIt(
IteratorType
outputIt(
reader->GetOutput(), inputRegion );
outputImage,
outputRegion );
inputIt.GoToBegin();
outputIt.GoToBegin();
while( !inputIt.IsAtEnd() )
{
outputIt.Set( inputIt.Get()
++inputIt;
++outputIt;
}
);
The while loop above is a common construct in ITK. The beauty of these four lines of code is that
they are equally valid for one, two, three, or even ten dimensional data, and no knowledge of the
size of the image is necessary. Consider the ugly alternative of ten nested for loops for traversing
an image.
Let’s run this example on the image FatMRISlice.png found in Examples/Data. The command
line arguments specify the input and output file names, then the x, y origin and the x, y size of the
cropped subregion.
ImageRegionIterator FatMRISlice.png ImageRegionIteratorOutput.png 20 70 210 140
The output is the cropped subregion shown in Figure 6.2.
6.3.2 ImageRegionIteratorWithIndex
The source code for this section can be found in the file
ImageRegionIteratorWithIndex.cxx.
The “WithIndex” family of iterators was designed for algorithms that use both the value and the
location of image pixels in calculations. Unlike itk::ImageRegionIterator, which calculates an
6.3. Image Iterators
157
Figure 6.2: Cropping a region from an image. The original image is shown at left. The image on the right is the
result of applying the ImageRegionIterator example code.
index only when asked for, itk::ImageRegionIteratorWithIndex maintains its index location
as a member variable that is updated during the increment or decrement process. Iteration speed is
注意：存在性能影响
penalized, but the index queries are more efficient.
The following example illustrates the use of ImageRegionIteratorWithIndex. The algorithm mirrors
a 2D image across its x-axis (see itk::FlipImageFilter for an ND version). The algorithm
makes extensive use of the GetIndex() method.
We start by including the proper header file.
#include "itkImageRegionIteratorWithIndex.h"
For this example, we will use an RGB pixel type so that we can process color images. Like most
other ITK image iterator, ImageRegionIteratorWithIndex class expects the image type as its single
template parameter.
constexpr unsigned int Dimension = 2;
using RGBPixelType = itk::RGBPixel< unsigned char >;
using ImageType = itk::Image< RGBPixelType, Dimension >;
using IteratorType = itk::ImageRegionIteratorWithIndex< ImageType >;
An ImageType smart pointer called inputImage points to the output of the image reader. After
updating the image reader, we can allocate an output image of the same size, spacing, and origin as
the input image.
158
Chapter 6. Iterators
ImageType::Pointer outputImage = ImageType::New();
outputImage->SetRegions( inputImage->GetRequestedRegion() );
outputImage->CopyInformation( inputImage );
outputImage->Allocate();
Next we create the iterator that walks the output image. This algorithm requires no iterator for the
input image.
IteratorType outputIt( outputImage, outputImage->GetRequestedRegion() );
This axis flipping algorithm works by iterating through the output image, querying the iterator for
its index, and copying the value from the input at an index mirrored across the x-axis.
ImageType::IndexType requestedIndex =
outputImage->GetRequestedRegion().GetIndex();
ImageType::SizeType requestedSize =
outputImage->GetRequestedRegion().GetSize();
for ( outputIt.GoToBegin(); !outputIt.IsAtEnd(); ++outputIt)
{
ImageType::IndexType idx = outputIt.GetIndex();
idx[0] = requestedIndex[0] + requestedSize[0] - 1 - idx[0];
outputIt.Set( inputImage->GetPixel(idx) );
}
注意：
Let’s run this example on the image VisibleWomanEyeSlice.png found in the Examples/Data
directory. Figure 6.3 shows how the original image has been mirrored across its x-axis in the output.
6.3.3 ImageLinearIteratorWithIndex
The source code for this section can be found in the file
ImageLinearIteratorWithIndex.cxx.
注意：用于一行一行处理图像
The itk::ImageLinearIteratorWithIndex is designed for line-by-line processing of an image.
It walks a linear path along a selected image direction parallel to one of the coordinate axes of the
image. This iterator conceptually breaks an image into a set of parallel lines that span the selected
image dimension.
Like all image iterators, movement of the ImageLinearIteratorWithIndex is constrained within an
image region R. The line ℓ through which the iterator moves is defined by selecting a direction and
an origin. The line ℓ extends from the origin to the upper boundary of R. The origin can be moved
to any position along the lower boundary of R.
Several additional methods are defined for this iterator to control movement of the iterator along the
line ℓ and movement of the origin of ℓ.
6.3. Image Iterators
159
Figure 6.3: Results of using ImageRegionIteratorWithIndex to mirror an image across an axis. The original
image is shown at left. The mirrored output is shown at right.
• NextLine() Moves the iterator to the beginning pixel location of the next line in the image.
The origin of the next line is determined by incrementing the current origin along the fastest
increasing dimension of the subspace of the image that excludes the selected dimension.
• PreviousLine() Moves the iterator to the last valid pixel location in the previous line.
The origin of the previous line is determined by decrementing the current origin along the
fastest increasing dimension of the subspace of the image that excludes the selected dimension.
• GoToBeginOfLine() Moves the iterator to the beginning pixel of the current line.
• GoToEndOfLine() Moves the iterator to one past the last valid pixel of the current line.
• GoToReverseBeginOfLine() Moves the iterator to the last valid pixel of the current
line.
• IsAtReverseEndOfLine() Returns true if the iterator points to one position before the
beginning pixel of the current line.
• IsAtEndOfLine() Returns true if the iterator points to one position past the last valid
pixel of the current line.
160
Chapter 6. Iterators
The following code example shows how to use the ImageLinearIteratorWithIndex. It implements
the same algorithm as in the previous example, flipping an image across its x-axis. Two line iterators
are iterated in opposite directions across the x-axis. After each line is traversed, the iterator origins
are stepped along the y-axis to the next line.
Headers for both the const and non-const versions are needed.
#include "itkImageLinearIteratorWithIndex.h"
The RGB image and pixel types are defined as in the previous example. The ImageLinearIteratorWithIndex class and its const version each have single template parameters, the image type.
using IteratorType = itk::ImageLinearIteratorWithIndex< ImageType >;
using ConstIteratorType = itk::ImageLinearConstIteratorWithIndex;
After reading the input image, we allocate an output image that of the same size, spacing, and origin.
ImageType::Pointer outputImage = ImageType::New();
outputImage->SetRegions( inputImage->GetRequestedRegion() );
outputImage->CopyInformation( inputImage );
outputImage->Allocate();
Next we create the two iterators. The const iterator walks the input image, and the non-const iterator
walks the output image. The iterators are initialized over the same region. The direction of iteration
is set to 0, the x dimension.
ConstIteratorType inputIt( inputImage, inputImage->GetRequestedRegion() );
IteratorType outputIt( outputImage, inputImage->GetRequestedRegion() );
inputIt.SetDirection(0);
outputIt.SetDirection(0);
Each line in the input is copied to the output. The input iterator moves forward across columns while
the output iterator moves backwards.
for ( inputIt.GoToBegin(), outputIt.GoToBegin(); ! inputIt.IsAtEnd();
outputIt.NextLine(), inputIt.NextLine())
{
inputIt.GoToBeginOfLine();
outputIt.GoToEndOfLine();
while ( ! inputIt.IsAtEndOfLine() )
{
--outputIt;
outputIt.Set( inputIt.Get() );
++inputIt;
}
}
6.3. Image Iterators
161
Running this example on VisibleWomanEyeSlice.png produces the same output image shown in
Figure 6.3.
The source code for this section can be found in the file
ImageLinearIteratorWithIndex2.cxx.
This example shows how to use the itk::ImageLinearIteratorWithIndex for computing the
mean across time of a 4D image where the first three dimensions correspond to spatial coordinates
and the fourth dimension corresponds to time. The result of the mean across time is to be stored in
a 3D image.
#include "itkImageLinearConstIteratorWithIndex.h"
First we declare the types of the images, the 3D and 4D readers.
using PixelType = unsigned char;
using Image3DType = itk::Image< PixelType, 3 >;
using Image4DType = itk::Image< PixelType, 4 >;
using Reader4DType = itk::ImageFileReader< Image4DType >;
using Writer3DType = itk::ImageFileWriter< Image3DType >;
Next, define the necessary types for indices, points, spacings, and size.
Image3DType::Pointer image3D = Image3DType::New();
using Index3DType = Image3DType::IndexType;
using Size3DType = Image3DType::SizeType;
using Region3DType = Image3DType::RegionType;
using Spacing3DType = Image3DType::SpacingType;
using Origin3DType = Image3DType::PointType;
using
using
using
using
Index4DType = Image4DType::IndexType;
Size4DType = Image4DType::SizeType;
Spacing4DType = Image4DType::SpacingType;
Origin4DType = Image4DType::PointType;
Here we make sure that the values for our resultant 3D mean image match up with the input 4D
image.
for( unsigned int i=0; i < 3; i++)
{
= size4D[i];
size3D[i]
index3D[i]
= index4D[i];
spacing3D[i] = spacing4D[i];
origin3D[i] = origin4D[i];
}
image3D->SetSpacing( spacing3D );
162
Chapter 6. Iterators
image3D->SetOrigin( origin3D
);
Region3DType region3D;
region3D.SetIndex( index3D );
region3D.SetSize( size3D );
image3D->SetRegions( region3D
image3D->Allocate();
);
Next we iterate over time in the input image series, compute the average, and store that value in the
corresponding pixel of the output 3D image.
IteratorType it( image4D, region4D );
it.SetDirection( 3 ); // Walk along time dimension
it.GoToBegin();
while( !it.IsAtEnd() )
{
SumType sum = itk::NumericTraits< SumType >::ZeroValue();
it.GoToBeginOfLine();
index4D = it.GetIndex();
while( !it.IsAtEndOfLine() )
{
sum += it.Get();
++it;
}
MeanType mean = static_cast< MeanType >( sum ) /
static_cast< MeanType >( timeLength );
index3D[0] = index4D[0];
index3D[1] = index4D[1];
index3D[2] = index4D[2];
image3D->SetPixel( index3D, static_cast< PixelType >( mean ) );
it.NextLine();
}
As you can see, we avoid to use a 3D iterator to walk over the mean image. The reason is that there
is no guarantee that the 3D iterator will walk in the same order as the 4D. Iterators just adhere to
their contract of visiting every pixel, but do not enforce any particular order for the visits. The linear
iterator guarantees it will visit the pixels along a line of the image in the order in which they are
placed in the line, but does not state in what order one line will be visited with respect to other lines.
Here we simply take advantage of knowing the first three components of the 4D iterator index, and
use them to place the resulting mean value in the output 3D image.
6.3.4 ImageSliceIteratorWithIndex
The source code for this section can be found in the file
ImageSliceIteratorWithIndex.cxx.
6.3. Image Iterators
163
The
itk::ImageSliceIteratorWithIndex
class
is
an
extension
of
itk::ImageLinearIteratorWithIndex from iteration along lines to iteration along both
lines and planes in an image. A slice is a 2D plane spanned by two vectors pointing along
orthogonal coordinate axes. The slice orientation of the slice iterator is defined by specifying its
two spanning axes.
• SetFirstDirection() Specifies the first coordinate axis direction of the slice plane.
• SetSecondDirection() Specifies the second coordinate axis direction of the slice plane.
Several new methods control movement from slice to slice.
• NextSlice() Moves the iterator to the beginning pixel location of the next slice in the
image. The origin of the next slice is calculated by incrementing the current origin index
along the fastest increasing dimension of the image subspace which excludes the first and
second dimensions of the iterator.
• PreviousSlice() Moves the iterator to the last valid pixel location in the previous slice.
The origin of the previous slice is calculated by decrementing the current origin index along
the fastest increasing dimension of the image subspace which excludes the first and second
dimensions of the iterator.
• IsAtReverseEndOfSlice() Returns true if the iterator points to one position before the
beginning pixel of the current slice.
• IsAtEndOfSlice() Returns true if the iterator points to one position past the last valid
pixel of the current slice.
The slice iterator moves line by line using NextLine() and PreviousLine(). The line direction is
parallel to the second coordinate axis direction of the slice plane (see also Section 6.3.3).
注意：
The next code example calculates the maximum intensity projection along one of the coordinate axes
of an image volume. The algorithm is straightforward using ImageSliceIteratorWithIndex because
we can coordinate movement through a slice of the 3D input image with movement through the 2D
planar output.
Here is how the algorithm works. For each 2D slice of the input, iterate through all the pixels line by
line. Copy a pixel value to the corresponding position in the 2D output image if it is larger than the
value already contained there. When all slices have been processed, the output image is the desired
maximum intensity projection.
We include a header for the const version of the slice iterator. For writing values to the 2D projection
image, we use the linear iterator from the previous section. The linear iterator is chosen because it
can be set to follow the same path in its underlying 2D image that the slice iterator follows over each
slice of the 3D image.
164
Chapter 6. Iterators
#include "itkImageSliceConstIteratorWithIndex.h"
#include "itkImageLinearIteratorWithIndex.h"
The pixel type is defined as unsigned short. For this application, we need two image types, a 3D
image for the input, and a 2D image for the intensity projection.
using PixelType = unsigned short;
using ImageType2D = itk::Image< PixelType, 2 >;
using ImageType3D = itk::Image< PixelType, 3 >;
A slice iterator type is defined to walk the input image.
using LinearIteratorType = itk::ImageLinearIteratorWithIndex< ImageType2D >;
using SliceIteratorType = itk::ImageSliceConstIteratorWithIndex;
The projection direction is read from the command line. The projection image will be the size of
the 2D plane orthogonal to the projection direction. Its spanning vectors are the two remaining
coordinate axes in the volume. These axes are recorded in the direction array.
auto projectionDirection =
static_cast( ::std::stoi( argv[3] ) );
unsigned int i, j;
unsigned int direction[2];
for (i = 0, j = 0; i < 3; ++i )
{
if (i != projectionDirection)
{
direction[j] = i;
j++;
}
}
The direction array is now used to define the projection image size based on the input image size.
The output image is created so that its common dimension(s) with the input image are the same
size. For example, if we project along the x axis of the input, the size and origin of the y axes of
the input and output will match. This makes the code slightly more complicated, but prevents a
counter-intuitive rotation of the output.
ImageType2D::RegionType region;
ImageType2D::RegionType::SizeType size;
ImageType2D::RegionType::IndexType index;
ImageType3D::RegionType requestedRegion = inputImage->GetRequestedRegion();
index[ direction[0] ]
= requestedRegion.GetIndex()[ direction[0] ];
注意：？？？
6.3. Image Iterators
165
index[ 1- direction[0] ] = requestedRegion.GetIndex()[ direction[1] ];
size[ direction[0] ]
= requestedRegion.GetSize()[ direction[0] ];
size[ 1- direction[0] ] = requestedRegion.GetSize()[ direction[1] ];
region.SetSize( size );
region.SetIndex( index );
ImageType2D::Pointer outputImage = ImageType2D::New();
outputImage->SetRegions( region );
outputImage->Allocate();
Next we create the necessary iterators. The const slice iterator walks the 3D input image, and the
non-const linear iterator walks the 2D output image. The iterators are initialized to walk the same
linear path through a slice. Remember that the second direction of the slice iterator defines the
direction that linear iteration walks within a slice.
SliceIteratorType inputIt( inputImage, inputImage->GetRequestedRegion() );
LinearIteratorType outputIt( outputImage,
outputImage->GetRequestedRegion() );
inputIt.SetFirstDirection( direction[1] );
inputIt.SetSecondDirection( direction[0] );
outputIt.SetDirection( 1 - direction[0] );
注意：？？？
Now we are ready to compute the projection. The first step is to initialize all of the projection values
to their nonpositive minimum value. The projection values are then updated row by row from the
first slice of the input. At the end of the first slice, the input iterator steps to the first row in the next
slice, while the output iterator, whose underlying image consists of only one slice, rewinds to its first
row. The process repeats until the last slice of the input is processed.
outputIt.GoToBegin();
while ( ! outputIt.IsAtEnd() )
{
while ( ! outputIt.IsAtEndOfLine() )
{
outputIt.Set( itk::NumericTraits::NonpositiveMin() );
++outputIt;
}
outputIt.NextLine();
}
inputIt.GoToBegin();
outputIt.GoToBegin();
while( !inputIt.IsAtEnd() )
{
while ( !inputIt.IsAtEndOfSlice() )
166
Chapter 6. Iterators
Figure 6.4: The maximum intensity projection through three slices of a volume.
{
while ( !inputIt.IsAtEndOfLine() )
{
outputIt.Set( std::max( outputIt.Get(), inputIt.Get() ));
++inputIt;
++outputIt;
}
outputIt.NextLine();
inputIt.NextLine();
}
outputIt.GoToBegin();
inputIt.NextSlice();
}
注意：存在line以及slice的迭代
Running this example code on the 3D image Examples/Data/BrainProtonDensity3Slices.mha
using the z-axis as the axis of projection gives the image shown in Figure 6.4.
6.3.5 ImageRandomConstIteratorWithIndex
The source code for this section can be found in the file
ImageRandomConstIteratorWithIndex.cxx.
注意：
itk::ImageRandomConstIteratorWithIndex was developed to randomly sample pixel values.
When incremented or decremented, it jumps to a random location in its image region.
6.4. Neighborhood Iterators
167
The user must specify a sample size when creating this iterator. The sample size, rather than a specific image index, defines the end position for the iterator. IsAtEnd() returns true when the current
sample number equals the sample size. IsAtBegin() returns true when the current sample number
equals zero. An important difference from other image iterators is that ImageRandomConstIteratorWithIndex may visit the same pixel more than once.
Let’s use the random iterator to estimate some simple image statistics. The next example calculates
an estimate of the arithmetic mean of pixel values.
First, include the appropriate header and declare pixel and image types.
#include "itkImageRandomConstIteratorWithIndex.h"
constexpr unsigned int Dimension = 2;
using PixelType = unsigned short;
using ImageType = itk::Image< PixelType, Dimension >;
using ConstIteratorType = itk::ImageRandomConstIteratorWithIndex;
The input image has been read as inputImage. We now create an iterator with a number of samples
set by command line argument. The call to ReinitializeSeed seeds the random number generator.
The iterator is initialized over the entire valid image region.
ConstIteratorType inputIt( inputImage, inputImage->GetRequestedRegion() );
inputIt.SetNumberOfSamples( ::std::stoi( argv[2]) );
inputIt.ReinitializeSeed();
Now take the specified number of samples and calculate their average value.
float mean = 0.0f;
for ( inputIt.GoToBegin(); ! inputIt.IsAtEnd(); ++inputIt)
{
mean += static_cast( inputIt.Get() );
}
mean = mean / ::std::stod( argv[2] );
The following table shows the results of running this example on several of the data files from
Examples/Data with a range of sample sizes.
6.4 Neighborhood Iterators
In ITK, a pixel neighborhood is loosely defined as a small set of pixels that are locally adjacent to
one another in an image. The size and shape of a neighborhood, as well the connectivity among
168
Chapter 6. Iterators
RatLungSlice1.mha
RatLungSlice2.mha
BrainT1Slice.png
10
50.5
46.7
47.2
Sample Size
100 1000 10000
52.4
53.0
52.4
47.5
47.4
47.6
64.1
68.0
67.8
Table 6.1: Estimates of mean image pixel value using the ImageRandomConstIteratorWithIndex at different
sample sizes.
pixels in a neighborhood, may vary with the application.
Many image processing algorithms are neighborhood-based, that is, the result at a pixel i is computed
from the values of pixels in the ND neighborhood of i. Consider finite difference operations in 2D.
A derivative at pixel index i = ( j, k), for example, is taken as a weighted difference of the values at
( j + 1, k) and ( j − 1, k). Other common examples of neighborhood operations include convolution
filtering and image morphology.
This section describes a class of ITK image iterators that are designed for working with pixel neighborhoods. An ITK neighborhood iterator walks an image region just like a normal image iterator,
but instead of only referencing a single pixel at each step, it simultaneously points to the entire ND
neighborhood of pixels. Extensions to the standard iterator interface provide read and write access to
all neighborhood pixels and information such as the size, extent, and location of the neighborhood.
Neighborhood iterators use the same operators defined in Section 6.2 and the same code constructs
as normal iterators for looping through an image. Figure 6.5 shows a neighborhood iterator moving
through an iteration region. This iterator defines a 3x3 neighborhood around each pixel that it visits.
The center of the neighborhood iterator is always positioned over its current index and all other
neighborhood pixel indices are referenced as offsets from the center index. The pixel under the
center of the neighborhood iterator and all pixels under the shaded area, or extent, of the iterator can
be dereferenced.
In addition to the standard image pointer and iteration region (Section 6.2), neighborhood iterator
constructors require an argument that specifies the extent of the neighborhood to cover. Neighborhood extent is symmetric across its center in each axis and is given as an array of N distances that
are collectively called the radius. Each element d of the radius, where 0 < d < N and N is the
dimensionality of the neighborhood, gives the extent of the neighborhood in pixels for dimension
N. The length of each face of the resulting ND hypercube is 2d + 1 pixels, a distance of d on either
side of the single pixel at the neighbor center. Figure 6.6 shows the relationship between the radius
of the iterator and the size of the neighborhood for a variety of 2D iterator shapes.
The radius of the neighborhood iterator is queried after construction by calling the GetRadius()
method. Some other methods provide some useful information about the iterator and its underlying
image.
• SizeType GetRadius() Returns the ND radius of the neighborhood as an itk::Size.
注意：
6.4. Neighborhood Iterators
169
itk::Image
BEGIN Position
Iteration Region
Neighborhood
Iterator
END Position
Figure 6.5: Path of a 3x3 neighborhood iterator through a 2D image region. The extent of the neighborhood is
indicated by the hashing around the iterator position. Pixels that lie within this extent are accessible through the
iterator. An arrow denotes a single iterator step, the result of one ++ operation.
170
Chapter 6. Iterators
0
1
2
1
0
3
4
5
(−1,0)
(0,0)
(1,0)
2
(−1,−2) (0,−2) (1,−2)
(−1, −1) (0, −1) (1,−1)
3
4
5
(−1,−1) (0,−1) (1,−1)
6
7
8
6
7
8
(−1,1)
(0,1)
(1,1)
(−1,0)
(0,0)
(1,0)
radius = [1,1]
size = [3,3]
1
2
7
8
14
15
16
(−3,1) (−2,1) (−1,1)
radius = [1,0]
size = [3,1]
10
11
0
(0,1)
(1,1)
(0,−2)
12
13
14
1
(−1,2)
(0,2)
(1,2)
(0,−1)
3
9
2
(1,0)
9
4
5
(−3,−1) (−2,−1) (−1,−1) (0,−1) (1,−1) (2,−1)
(−3,0) (−2,0) (−1,0)
1
(0,0)
(−1,1)
radius = [1,2]
size = [3,5]
0
0
(−1,0)
2
(0,0)
6
3
(3,−1)
(0,1)
10
11
12
13
4
(0,0)
(1,0)
(2,0)
(3,0)
(0,2)
17
18
19
20
(0,1)
(1,1)
(2,1)
(3,1)
radius = [0,2]
size = [1,5]
radius = [3,1]
size = [7,3]
Figure 6.6: Several possible 2D neighborhood iterator shapes are shown along with their radii and sizes. A
neighborhood pixel can be dereferenced by its integer index (top) or its offset from the center (bottom). The
center pixel of each iterator is shaded.
6.4. Neighborhood Iterators
171
• const ImageType *GetImagePointer() Returns the pointer to the image referenced by the iterator.
• unsigned long Size() Returns the size in number of pixels of the neighborhood.
The neighborhood iterator interface extends the normal ITK iterator interface for setting and getting
pixel values. One way to dereference pixels is to think of the neighborhood as a linear array where
each pixel has a unique integer index. The index of a pixel in the array is determined by incrementing from the upper-left-forward corner of the neighborhood along the fastest increasing image
dimension: first column, then row, then slice, and so on. In Figure 6.6, the unique integer index is
shown at the top of each pixel. The center pixel is always at position n/2, where n is the size of the
array.
• PixelType GetPixel(const unsigned int i) Returns the value of the pixel at
neighborhood position i.
• void SetPixel(const unsigned int i, PixelType p) Sets the value of the
pixel at position i to p.
Another way to think about a pixel location in a neighborhood is as an ND offset from the neighborhood center. The upper-left-forward corner of a 3x3x3 neighborhood, for example, can be described
by offset (−1, −1, −1). The bottom-right-back corner of the same neighborhood is at offset (1, 1, 1).
In Figure 6.6, the offset from center is shown at the bottom of each neighborhood pixel.
• PixelType GetPixel(const OffsetType &o) Get the value of the pixel at the
position offset o from the neighborhood center.
• void SetPixel(const OffsetType &o, PixelType p) Set the value at the
position offset o from the neighborhood center to the value p.
The neighborhood iterators also provide a shorthand for setting and getting the value at the center of
the neighborhood.
• PixelType GetCenterPixel() Gets the value at the center of the neighborhood.
• void SetCenterPixel(PixelType p) Sets the value at the center of the neighborhood to the value p
There is another shorthand for setting and getting values for pixels that lie some integer distance
from the neighborhood center along one of the image axes.
• PixelType GetNext(unsigned int d) Get the value immediately adjacent to the
neighborhood center in the positive direction along the d axis.
注意：获取邻域的像素
值的方法--要么表示成
一维数组（按行存储）
依据index进行像素值
访问；要么以像素与中
心的的偏移量进行访问
172
Chapter 6. Iterators
注意：
• void SetNext(unsigned int d, PixelType p) Set the value immediately adjacent to the neighborhood center in the positive direction along the d axis to the value p.
• PixelType GetPrevious(unsigned int d) Get the value immediately adjacent
to the neighborhood center in the negative direction along the d axis.
• void SetPrevious(unsigned int d, PixelType p) Set the value immediately adjacent to the neighborhood center in the negative direction along the d axis to the
value p.
• PixelType GetNext(unsigned int d, unsigned int s) Get the value of the
pixel located s pixels from the neighborhood center in the positive direction along the d axis.
• void SetNext(unsigned int d, unsigned int s, PixelType p) Set the
value of the pixel located s pixels from the neighborhood center in the positive direction along
the d axis to value p.
• PixelType GetPrevious(unsigned int d, unsigned int s) Get the value
of the pixel located s pixels from the neighborhood center in the positive direction along the
d axis.
• void SetPrevious(unsigned int d, unsigned int s, PixelType p)
Set the value of the pixel located s pixels from the neighborhood center in the positive
direction along the d axis to value p.
It is also possible to extract or set all of the neighborhood values from an iterator at once using a
regular ITK neighborhood object. This may be useful in algorithms that perform a particularly large
number of calculations in the neighborhood and would otherwise require multiple dereferences of
the same pixels.
• NeighborhoodType GetNeighborhood() Return a itk::Neighborhood of the
same size and shape as the neighborhood iterator and contains all of the values at the iterator position.
• void SetNeighborhood(NeighborhoodType &N) Set all of the values in the
neighborhood at the iterator position to those contained in Neighborhood N, which must be
the same size and shape as the iterator.
Several methods are defined to provide information about the neighborhood.
• IndexType GetIndex() Return the image index of the center pixel of the neighborhood
iterator.
• IndexType GetIndex(OffsetType o) Return the image index of the pixel at offset
o from the neighborhood center.
6.4. Neighborhood Iterators
173
• IndexType GetIndex(unsigned int i) Return the image index of the pixel at array position i.
• OffsetType GetOffset(unsigned int i) Return the offset from the neighborhood center of the pixel at array position i.
• unsigned long GetNeighborhoodIndex(OffsetType o) Return the array position of the pixel at offset o from the neighborhood center.
• std::slice GetSlice(unsigned int n) Return a std::slice through the iterator neighborhood along axis n.
注意：边界处理
A neighborhood-based calculation in a neighborhood close to an image boundary may require data
that falls outside the boundary. The iterator in Figure 6.5, for example, is centered on a boundary
pixel such that three of its neighbors actually do not exist in the image. When the extent of a
neighborhood falls outside the image, pixel values for missing neighbors are supplied according to
a rule, usually chosen to satisfy the numerical requirements of the algorithm. A rule for supplying
out-of-bounds values is called a boundary condition.
ITK neighborhood iterators automatically detect out-of-bounds dereferences and will return values
according to boundary conditions. The boundary condition type is specified by the second, optional
template parameter of the iterator. By default, neighborhood iterators use a Neumann condition
where the first derivative across the boundary is zero. The Neumann rule simply returns the closest
in-bounds pixel value to the requested out-of-bounds location. Several other common boundary
conditions can be found in the ITK toolkit. They include a periodic condition that returns the pixel
value from the opposite side of the data set, and is useful when working with periodic data such as
Fourier transforms, and a constant value condition that returns a set value v for all out-of-bounds
pixel dereferences. The constant value condition is equivalent to padding the image with value v.
注意：该迭代器
所使用的边界处
理规则
Bounds checking is a computationally expensive operation because it occurs each time the iterator is
incremented. To increase efficiency, a neighborhood iterator automatically disables bounds checking
when it detects that it is not necessary. A user may also explicitly disable or enable bounds checking.
Most neighborhood based algorithms can minimize the need for bounds checking through clever
注意：通过技术减少边界检测的需求
definition of iteration regions. These techniques are explored in Section 6.4.1.
• void NeedToUseBoundaryConditionOn() Explicitly turn bounds checking on.
This method should be used with caution because unnecessarily enabling bounds checking
may result in a significant performance decrease. In general you should allow the iterator to
automatically determine this setting.
• void NeedToUseBoundaryConditionOff() Explicitly disable bounds checking.
This method should be used with caution because disabling bounds checking when it is needed
will result in out-of-bounds reads and undefined results.
• void OverrideBoundaryCondition(BoundaryConditionType *b) Overrides the templated boundary condition, using boundary condition object b instead. Object b
174
Chapter 6. Iterators
should not be deleted until it has been released by the iterator. This method can be used to
change iterator behavior at run-time.
• void ResetBoundaryCondition() Discontinues the use of any run-time specified
boundary condition and returns to using the condition specified in the template argument.
• void SetPixel(unsigned int i, PixelType p, bool status) Sets the
value at neighborhood array position i to value p. If the position i is out-of-bounds, status
is set to false, otherwise status is set to true.
The following sections describe the two ITK neighborhood iterator classes,
itk::NeighborhoodIterator and itk::ShapedNeighborhoodIterator. Each has a const and
a non-const version. The shaped iterator is a refinement of the standard NeighborhoodIterator that
supports an arbitrarily-shaped (non-rectilinear) neighborhood.
6.4.1 NeighborhoodIterator
The standard neighborhood iterator class in ITK is the itk::NeighborhoodIterator. Together
with its const version, itk::ConstNeighborhoodIterator, it implements the complete API described above. This section provides several examples to illustrate the use of NeighborhoodIterator.
Basic neighborhood techniques: edge detection
The source code for this section can be found in the file
NeighborhoodIterators1.cxx.
注意：实现简单的Sobel边缘检测算
法
This example uses the itk::NeighborhoodIterator to implement a simple Sobel edge detection
algorithm [4]. The algorithm uses the neighborhood iterator to iterate through an input image and
calculate a series of finite difference derivatives. Since the derivative results cannot be written back
to the input image without affecting later calculations, they are written instead to a second, output
image. Most neighborhood processing algorithms follow this read-only model on their inputs.
We begin by including the proper header files. The itk::ImageRegionIterator will be used to
write the results of computations to the output image. A const version of the neighborhood iterator
is used because the input image is read-only.
#include "itkConstNeighborhoodIterator.h"
#include "itkImageRegionIterator.h"
The finite difference calculations in this algorithm require floating point values. Hence, we define
the image pixel type to be float and the file reader will automatically cast fixed-point data to float.
We declare the iterator types using the image type as the template parameter. The second template
parameter of the neighborhood iterator, which specifies the boundary condition, has been omitted
because the default condition is appropriate for this algorithm.
6.4. Neighborhood Iterators
175
using PixelType = float;
using ImageType = itk::Image< PixelType, 2 >;
using ReaderType = itk::ImageFileReader< ImageType >;
using NeighborhoodIteratorType = itk::ConstNeighborhoodIterator< ImageType >;
using IteratorType = itk::ImageRegionIterator< ImageType>;
The following code creates and executes the ITK image reader. The Update call on the reader object
is surrounded by the standard try/catch blocks to handle any exceptions that may be thrown by
the reader.
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
try
{
reader->Update();
}
catch ( itk::ExceptionObject &err)
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
We can now create a neighborhood iterator to range over the output of the reader. For Sobel edgedetection in 2D, we need a square iterator that extends one pixel away from the neighborhood center
in every dimension.
NeighborhoodIteratorType::RadiusType radius;
radius.Fill(1);
NeighborhoodIteratorType it( radius, reader->GetOutput(),
reader->GetOutput()->GetRequestedRegion() );
The following code creates an output image and iterator.
ImageType::Pointer output = ImageType::New();
output->SetRegions(reader->GetOutput()->GetRequestedRegion());
output->Allocate();
IteratorType out(output, reader->GetOutput()->GetRequestedRegion());
Sobel edge detection uses weighted finite difference calculations to construct an edge magnitude
image. Normally the edge magnitude is the root sum of squares of partial derivatives in all directions,
but for simplicity this example only calculates the x component. The result is a derivative image
biased toward maximally vertical edges.
176
Chapter 6. Iterators
The finite differences are computed from pixels at six locations in the neighborhood. In this example,
we use the iterator GetPixel() method to query the values from their offsets in the neighborhood.
The example in Section 6.4.1 uses convolution with a Sobel kernel instead.
Six positions in the neighborhood are necessary for the finite difference calculations. These positions
are recorded in offset1 through offset6.
NeighborhoodIteratorType::OffsetType
NeighborhoodIteratorType::OffsetType
NeighborhoodIteratorType::OffsetType
NeighborhoodIteratorType::OffsetType
NeighborhoodIteratorType::OffsetType
NeighborhoodIteratorType::OffsetType
offset1
offset2
offset3
offset4
offset5
offset6
=
=
=
=
=
=
{{-1,-1}};
{{1,-1}};
{{-1,0 }};
{{1,0}};
{{-1,1}};
{{1,1}};
It is equivalent to use the six corresponding integer array indices instead. For example, the offsets
(-1,-1) and (1, -1) are equivalent to the integer indices 0 and 2, respectively.
The calculations are done in a for loop that moves the input and output iterators synchronously
across their respective images. The sum variable is used to sum the results of the finite differences.
for (it.GoToBegin(), out.GoToBegin(); !it.IsAtEnd(); ++it, ++out)
{
float sum;
sum = it.GetPixel(offset2) - it.GetPixel(offset1);
sum += 2.0 * it.GetPixel(offset4) - 2.0 * it.GetPixel(offset3);
sum += it.GetPixel(offset6) - it.GetPixel(offset5);
out.Set(sum);
}
注意：
The last step is to write the output buffer to an image file. Writing is done inside a try/catch block
to handle any exceptions. The output is rescaled to intensity range [0, 255] and cast to unsigned char
so that it can be saved and visualized as a PNG image.
注意：这两个操作是必要的
using WritePixelType = unsigned char;
using WriteImageType = itk::Image< WritePixelType, 2 >;
using WriterType = itk::ImageFileWriter< WriteImageType >;
using RescaleFilterType = itk::RescaleIntensityImageFilter<
ImageType, WriteImageType >;
RescaleFilterType::Pointer rescaler = RescaleFilterType::New();
rescaler->SetOutputMinimum( 0 );
rescaler->SetOutputMaximum( 255 );
rescaler->SetInput(output);
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[2] );
writer->SetInput(rescaler->GetOutput());
try
6.4. Neighborhood Iterators
177
Figure 6.7: Applying the Sobel operator in different orientations to an MRI image (left) produces x (center) and
y (right) derivative images.
{
writer->Update();
}
catch ( itk::ExceptionObject &err)
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
The center image of Figure 6.7 shows the output of the Sobel algorithm applied to
Examples/Data/BrainT1Slice.png.
Convolution filtering: Sobel operator
The source code for this section can be found in the file
NeighborhoodIterators2.cxx.
In this example, the Sobel edge-detection routine is rewritten using convolution filtering. Convolution filtering is a standard image processing technique that can be implemented numerically as the
inner product of all image neighborhoods with a convolution kernel [4] [2]. In ITK, we use a class
of objects called neighborhood operators as convolution kernels and a special function object called
itk::NeighborhoodInnerProduct to calculate inner products.
The basic ITK convolution filtering routine is to step through the image with a neighborhood iterator
and use NeighborhoodInnerProduct to find the inner product of each neighborhood with the desired
kernel. The resulting values are written to an output image. This example uses a neighborhood operator called the itk::SobelOperator, but all neighborhood operators can be convolved with im注意：
178
Chapter 6. Iterators
ages using this basic routine. Other examples of neighborhood operators include derivative kernels,
Gaussian kernels, and morphological operators. itk::NeighborhoodOperatorImageFilter is a
generalization of the code in this section to ND images and arbitrary convolution kernels.
We start writing this example by including the header files for the Sobel kernel and the inner product
function.
#include "itkSobelOperator.h"
#include "itkNeighborhoodInnerProduct.h"
Refer to the previous example for a description of reading the input image and setting up the output
image and iterator.
The following code creates a Sobel operator. The Sobel operator requires a direction for its partial
derivatives. This direction is read from the command line. Changing the direction of the derivatives
changes the bias of the edge detection, i.e. maximally vertical or maximally horizontal.
itk::SobelOperator sobelOperator;
sobelOperator.SetDirection( ::std::stoi(argv[3]) );
sobelOperator.CreateDirectional();
The neighborhood iterator is initialized as before, except that now it takes its radius directly from
the radius of the Sobel operator. The inner product function object is templated over image type and
requires no initialization.
NeighborhoodIteratorType::RadiusType radius = sobelOperator.GetRadius();
NeighborhoodIteratorType it( radius, reader->GetOutput(),
reader->GetOutput()->GetRequestedRegion() );
itk::NeighborhoodInnerProduct innerProduct;
Using the Sobel operator, inner product, and neighborhood iterator objects, we can now write a very
simple for loop for performing convolution filtering. As before, out-of-bounds pixel values are
注意：写滤波器的时候一定要考虑边
supplied automatically by the iterator.
界情况
for (it.GoToBegin(), out.GoToBegin(); !it.IsAtEnd(); ++it, ++out)
{
out.Set( innerProduct( it, sobelOperator ) );
}
The output is rescaled and written as in the previous example. Applying this example in the x and y
directions produces the images at the center and right of Figure 6.7. Note that x-direction operator
produces the same output image as in the previous example.
6.4. Neighborhood Iterators
179
Optimizing iteration speed
The source code for this section can be found in the file
NeighborhoodIterators3.cxx.
This example illustrates a technique for improving the efficiency of neighborhood calculations by
eliminating unnecessary bounds checking. As described in Section 6.4, the neighborhood iterator
automatically enables or disables bounds checking based on the iteration region in which it is initialized. By splitting our image into boundary and non-boundary regions, and then processing each
region using a different neighborhood iterator, the algorithm will only perform bounds-checking on
those pixels for which it is actually required. This trick can provide a significant speedup for simple
algorithms such as our Sobel edge detection, where iteration speed is a critical.
Splitting the image into the necessary regions is an easy task when you use the
itk::NeighborhoodAlgorithm::ImageBoundaryFacesCalculator. The face calculator is so
named because it returns a list of the “faces” of the ND dataset. Faces are those regions whose pixels all lie within a distance d from the boundary, where d is the radius of the neighborhood stencil
used for the numerical calculations. In other words, faces are those regions where a neighborhood
iterator of radius d will always overlap the boundary of the image. The face calculator also returns
the single inner region, in which out-of-bounds values are never required and bounds checking is
not necessary.
The face calculator object is defined in itkNeighborhoodAlgorithm.h. We include this file in
addition to those from the previous two examples.
#include "itkNeighborhoodAlgorithm.h"
First we load the input image and create the output image and inner product function as in the
previous examples. The image iterators will be created in a later step. Next we create a face calculator object. An empty list is created to hold the regions that will later on be returned by the face
calculator.
using FaceCalculatorType =
itk::NeighborhoodAlgorithm::ImageBoundaryFacesCalculator;
FaceCalculatorType faceCalculator;
FaceCalculatorType::FaceListType faceList;
The face calculator function is invoked by passing it an image pointer, an image region, and a
neighborhood radius. The image pointer is the same image used to initialize the neighborhood
iterator, and the image region is the region that the algorithm is going to process. The radius is the
radius of the iterator.
Notice that in this case the image region is given as the region of the output image and the image
pointer is given as that of the input image. This is important if the input and output images differ in
size, i.e. the input image is larger than the output image. ITK image filters, for example, operate on
注意：为了提高效
率，将图像划分为
边界区域以及非边
界区域；区别对待
处理
注意：取出哪些覆盖了
边界的区域，由r决定
180
Chapter 6. Iterators
data from the input image but only generate results in the RequestedRegion of the output image,
which may be smaller than the full extent of the input.
faceList = faceCalculator(reader->GetOutput(), output->GetRequestedRegion(),
sobelOperator.GetRadius());
The face calculator has returned a list of 2N + 1 regions. The first element in the list is always the
inner region, which may or may not be important depending on the application. For our purposes it
does not matter because all regions are processed the same way. We use an iterator to traverse the
list of faces.
FaceCalculatorType::FaceListType::iterator fit;
We now rewrite the main loop of the previous example so that each region in the list is processed
by a separate iterator. The iterators it and out are reinitialized over each region in turn. Bounds
checking is automatically enabled for those regions that require it, and disabled for the region that
does not.
IteratorType out;
NeighborhoodIteratorType it;
for ( fit=faceList.begin(); fit != faceList.end(); ++fit)
{
it = NeighborhoodIteratorType( sobelOperator.GetRadius(),
reader->GetOutput(), *fit );
out = IteratorType( output, *fit );
for (it.GoToBegin(), out.GoToBegin(); ! it.IsAtEnd(); ++it, ++out)
{
out.Set( innerProduct(it, sobelOperator) );
}
}
The output is written as before. Results for this example are the same as the previous example. You
may not notice the speedup except on larger images. When moving to 3D and higher dimensions,
the effects are greater because the volume to surface area ratio is usually larger. In other words, as
the number of interior pixels increases relative to the number of face pixels, there is a corresponding
注意：数据越大，该方法速度提升越
increase in efficiency from disabling bounds checking on interior pixels.
明显；因为iner pixels数量大，减
少了大部分boundary checking
Separable convolution: Gaussian filtering
The source code for this section can be found in the file
NeighborhoodIterators4.cxx.
We now introduce a variation on convolution filtering that is useful when a convolution kernel is
separable. In this example, we create a different neighborhood iterator for each axial direction
6.4. Neighborhood Iterators
181
注意：各个轴向进行卷积运算
of the image and then take separate inner products with a 1D discrete Gaussian kernel. The idea
of using several neighborhood iterators at once has applications beyond convolution filtering and
may improve efficiency when the size of the whole neighborhood relative to the portion of the
neighborhood used in calculations becomes large.
The only new class necessary for this example is the Gaussian operator.
#include "itkGaussianOperator.h"
The Gaussian operator, like the Sobel operator, is instantiated with a pixel type and a dimensionality.
Additionally, we set the variance of the Gaussian, which has been read from the command line as
standard deviation.
itk::GaussianOperator< PixelType, 2 > gaussianOperator;
gaussianOperator.SetVariance( ::std::stod(argv[3]) * ::std::stod(argv[3]) );
The only further changes from the previous example are in the main loop. Once again we use the
results from face calculator to construct a loop that processes boundary and non-boundary image
regions separately. Separable convolution, however, requires an additional, outer loop over all the
image dimensions. The direction of the Gaussian operator is reset at each iteration of the outer loop
using the new dimension. The iterators change direction to match because they are initialized with
the radius of the Gaussian operator.
Input and output buffers are swapped at each iteration so that the output of the previous iteration
becomes the input for the current iteration. The swap is not performed on the last iteration.
注意：
ImageType::Pointer input = reader->GetOutput();
for (unsigned int i = 0; i < ImageType::ImageDimension; ++i)
{
注意：每个方向计算一遍，转换成
gaussianOperator.SetDirection(i);
一维的滤波器
gaussianOperator.CreateDirectional();
faceList = faceCalculator(input, output->GetRequestedRegion(),
gaussianOperator.GetRadius());
for ( fit=faceList.begin(); fit != faceList.end(); ++fit )
{
it = NeighborhoodIteratorType( gaussianOperator.GetRadius(),
input, *fit );
out = IteratorType( output, *fit );
for (it.GoToBegin(), out.GoToBegin(); ! it.IsAtEnd(); ++it, ++out)
{
out.Set( innerProduct(it, gaussianOperator) );
}
}
// Swap the input and output buffers
182
Chapter 6. Iterators
Figure 6.8: Results of convolution filtering with a Gaussian kernel of increasing standard deviation σ (from left
to right, σ = 0, σ = 1, σ = 2, σ = 5). Increased blurring reduces contrast and changes the average intensity
value of the image, which causes the image to appear brighter when rescaled.
if (i != ImageType::ImageDimension - 1)
{
ImageType::Pointer tmp = input;
注意：第一维的结果作为第二维的输
input = output;
入，以此类推
output = tmp;
}
}
注意：这步是将数据归于[0， 255]
范围
The output is rescaled and written as in the previous examples. Figure 6.8 shows the results of
Gaussian blurring the image Examples/Data/BrainT1Slice.png using increasing kernel widths.
Slicing the neighborhood
The source code for this section can be found in the file
NeighborhoodIterators5.cxx.
This example introduces slice-based neighborhood processing. A slice, in this context, is a 1D path
through an ND neighborhood. Slices are defined for generic arrays by the std::slice class as a
start index, a step size, and an end index. Slices simplify the implementation of certain neighborhood calculations. They also provide a mechanism for taking inner products with subregions of
neighborhoods.
Suppose, for example, that we want to take partial derivatives in the y direction of a neighborhood,
but offset those derivatives by one pixel position along the positive x direction. For a 3 × 3, 2D
neighborhood iterator, we can construct an std::slice, (start = 2, stride = 3, end = 8),
that represents the neighborhood offsets (1, −1), (1, 0), (1, 1) (see Figure 6.6). If we pass this slice
as an extra argument to the itk::NeighborhoodInnerProduct function, then the inner product
is taken only along that slice. This “sliced” inner product with a 1D itk::DerivativeOperator
gives the desired derivative.
6.4. Neighborhood Iterators
183
The previous separable Gaussian filtering example can be rewritten using slices and slicebased inner products. In general, slice-based processing is most useful when doing many
different calculations on the same neighborhood, where defining multiple iterators as in Section 6.4.1 becomes impractical or inefficient. Good examples of slice-based neighborhood
processing can be found in any of the ND anisotropic diffusion function objects, such as
itk::CurvatureNDAnisotropicDiffusionFunction.
The first difference between this example and the previous example is that the Gaussian operator is
only initialized once. Its direction is not important because it is only a 1D array of coefficients.
itk::GaussianOperator< PixelType, 2 > gaussianOperator;
gaussianOperator.SetDirection(0);
gaussianOperator.SetVariance( ::std::stod(argv[3]) * ::std::stod(argv[3]) );
gaussianOperator.CreateDirectional();
Next we need to define a radius for the iterator. The radius in all directions matches that of the single
extent of the Gaussian operator, defining a square neighborhood.
NeighborhoodIteratorType::RadiusType radius;
radius.Fill( gaussianOperator.GetRadius()[0] );
The inner product and face calculator are defined for the main processing loop as before, but now
the iterator is reinitialized each iteration with the square radius instead of the radius of the operator.
The inner product is taken using a slice along the axial direction corresponding to the current iteration. Note the use of GetSlice() to return the proper slice from the iterator itself. GetSlice() can
only be used to return the slice along the complete extent of the axial direction of a neighborhood.
ImageType::Pointer input = reader->GetOutput();
faceList = faceCalculator(input, output->GetRequestedRegion(), radius);
for (unsigned int i = 0; i < ImageType::ImageDimension; ++i)
{
for ( fit=faceList.begin(); fit != faceList.end(); ++fit )
{
it = NeighborhoodIteratorType( radius, input, *fit );
out = IteratorType( output, *fit );
for (it.GoToBegin(), out.GoToBegin(); ! it.IsAtEnd(); ++it, ++out)
{
out.Set( innerProduct(it.GetSlice(i), it, gaussianOperator) );
}
}
// Swap the input and output buffers
if (i != ImageType::ImageDimension - 1)
{
ImageType::Pointer tmp = input;
input = output;
output = tmp;
注意：
184
Chapter 6. Iterators
}
}
This technique produces exactly the same results as the previous example. A little experimentation,
however, will reveal that it is less efficient since the neighborhood iterator is keeping track of extra,
unused pixel locations for each iteration, while the previous example only references those pixels
that it needs. In cases, however, where an algorithm takes multiple derivatives or convolution products over the same neighborhood, slice-based processing can increase efficiency and simplify the
注意：出现这种情况时，slice技术
implementation.
效率要高
Random access iteration
The source code for this section can be found in the file
NeighborhoodIterators6.cxx.
注意：连通区域标记
算法
Some image processing routines do not need to visit every pixel in an image. Flood-fill and
connected-component algorithms, for example, only visit pixels that are locally connected to one
another. Algorithms such as these can be efficiently written using the random access capabilities of
the neighborhood iterator.
The following example finds local minima. Given a seed point, we can search the neighborhood of
that point and pick the smallest value m. While m is not at the center of our current neighborhood,
we move in the direction of m and repeat the analysis. Eventually we discover a local minimum and
stop. This algorithm is made trivially simple in ND using an ITK neighborhood iterator.
To illustrate the process, we create an image that descends everywhere to a single minimum: a
positive distance transform to a point. The details of creating the distance transform are not relevant
to the discussion of neighborhood iterators, but can be found in the source code of this example.
Some noise has been added to the distance transform image for additional interest.
The variable input is the pointer to the distance transform image. The local minimum algorithm is
initialized with a seed point read from the command line.
ImageType::IndexType index;
index[0] = ::std::stoi(argv[2]);
index[1] = ::std::stoi(argv[3]);
Next we create the neighborhood iterator and position it at the seed point.
NeighborhoodIteratorType::RadiusType radius;
radius.Fill(1);
NeighborhoodIteratorType it(radius, input, input->GetRequestedRegion());
it.SetLocation(index);
6.4. Neighborhood Iterators
185
Figure 6.9: Paths traversed by the neighborhood iterator from different seed points to the local minimum. The
true minimum is at the center of the image. The path of the iterator is shown in white. The effect of noise in the
image is seen as small perturbations in each path.
Searching for the local minimum involves finding the minimum in the current neighborhood, then
shifting the neighborhood in the direction of that minimum. The for loop below records the
itk::Offset of the minimum neighborhood pixel. The neighborhood iterator is then moved using
that offset. When a local minimum is detected, flag will remain false and the while loop will exit.
Note that this code is valid for an image of any dimensionality.
bool flag = true;
while ( flag == true )
{
NeighborhoodIteratorType::OffsetType nextMove;
nextMove.Fill(0);
flag = false;
PixelType min = it.GetCenterPixel();
for (unsigned i = 0; i < it.Size(); i++)
{
if ( it.GetPixel(i) < min )
{
min = it.GetPixel(i);
nextMove = it.GetOffset(i);
flag = true;
}
}
it.SetCenterPixel( 255.0 );
it += nextMove;
}
注意： 这个例子重要
Figure 6.9 shows the results of the algorithm for several seed points. The white line is the path of
the iterator from the seed point to the minimum in the center of the image. The effect of the additive
noise is visible as the small perturbations in the paths.
186
Chapter 6. Iterators
6.4.2 ShapedNeighborhoodIterator
This section describes a variation on the neighborhood iterator called a shaped neighborhood
iterator. A shaped neighborhood is defined like a bit mask, or stencil, with different offsets
in the rectilinear neighborhood of the normal neighborhood iterator turned off or on to create a pattern. Inactive positions (those not in the stencil) are not updated during iteration
and their values cannot be read or written. The shaped iterator is implemented in the class
itk::ShapedNeighborhoodIterator, which is a subclass of itk::NeighborhoodIterator. A
const version, itk::ConstShapedNeighborhoodIterator, is also available.
Like a regular neighborhood iterator, a shaped neighborhood iterator must be initialized with an ND
radius object, but the radius of the neighborhood of a shaped iterator only defines the set of possible
neighbors. Any number of possible neighbors can then be activated or deactivated. The shaped
neighborhood iterator defines an API for activating neighbors. When a neighbor location, defined
relative to the center of the neighborhood, is activated, it is placed on the active list and is then part
of the stencil. An iterator can be “reshaped” at any time by adding or removing offsets from the
active list.
• void ActivateOffset(OffsetType &o) Include the offset o in the stencil of active
neighborhood positions. Offsets are relative to the neighborhood center.
• void DeactivateOffset(OffsetType &o) Remove the offset o from the stencil of
active neighborhood positions. Offsets are relative to the neighborhood center.
• void ClearActiveList() Deactivate all positions in the iterator stencil by clearing the
active list.
• unsigned int GetActiveIndexListSize() Return the number of pixel locations
that are currently active in the shaped iterator stencil.
Because the neighborhood is less rigidly defined in the shaped iterator, the set of pixel access methods is restricted. Only the GetPixel() and SetPixel() methods are available, and calling these
methods on an inactive neighborhood offset will return undefined results.
For the common case of traversing all pixel offsets in a neighborhood, the shaped iterator class
provides an iterator through the active offsets in its stencil. This stencil iterator can be incremented
or decremented and defines Get() and Set() for reading and writing the values in the neighborhood.
注意：shape内的迭代器
• ShapedNeighborhoodIterator::Iterator Begin() Return a const or nonconst iterator through the shaped iterator stencil that points to the first valid location in the
stencil.
• ShapedNeighborhoodIterator::Iterator End() Return a const or non-const
iterator through the shaped iterator stencil that points one position past the last valid location
in the stencil.
注意：就是mask？？
通过offset来创建
mask
6.4. Neighborhood Iterators
187
The functionality and interface of the shaped neighborhood iterator is best described by example. We
will use the ShapedNeighborhoodIterator to implement some binary image morphology algorithms
注意：
(see [4], [2], et al.). The examples that follow implement erosion and dilation.
Shaped neighborhoods: morphological operations
The source code for this section can be found in the file
ShapedNeighborhoodIterators1.cxx.
This example uses itk::ShapedNeighborhoodIterator to implement a binary erosion algorithm.
If we think of an image I as a set of pixel indices, then erosion of I by a smaller set E, called the
structuring element, is the set of all indices at locations x in I such that when E is positioned at x,
every element in E is also contained in I. 注意：满足这个条件的保留下来就是
腐蚀的结果
This type of algorithm is easy to implement with shaped neighborhood iterators because we can use
the iterator itself as the structuring element E and move it sequentially through all positions x. The
result at x is obtained by checking values in a simple iteration loop through the neighborhood stencil.
We need two iterators, a shaped iterator for the input image and a regular image iterator for writing
results to the output image.
#include "itkConstShapedNeighborhoodIterator.h"
#include "itkImageRegionIterator.h"
Since we are working with binary images in this example, an unsigned char pixel type will do.
The image and iterator types are defined using the pixel type.
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, 2 >;
using ShapedNeighborhoodIteratorType =
itk::ConstShapedNeighborhoodIterator;
using IteratorType = itk::ImageRegionIterator< ImageType>;
Refer to the examples in Section 6.4.1 or the source code of this example for a description of how to
read the input image and allocate a matching output image.
The size of the structuring element is read from the command line and used to define a radius for
the shaped neighborhood iterator. Using the method developed in section 6.4.1 to minimize bounds
checking, the iterator itself is not initialized until entering the main processing loop.
unsigned int element_radius = ::std::stoi( argv[3] );
ShapedNeighborhoodIteratorType::RadiusType radius;
radius.Fill(element_radius);
注意：？？？
188
Chapter 6. Iterators
The face calculator object introduced in Section 6.4.1 is created and used as before.
using FaceCalculatorType =
itk::NeighborhoodAlgorithm::ImageBoundaryFacesCalculator;
FaceCalculatorType faceCalculator;
FaceCalculatorType::FaceListType faceList;
FaceCalculatorType::FaceListType::iterator fit;
faceList = faceCalculator( reader->GetOutput(),
output->GetRequestedRegion(),
radius );
Now we initialize some variables and constants.
IteratorType out;
constexpr PixelType background_value = 0;
constexpr PixelType foreground_value = 255;
const auto rad = static_cast(element_radius);
The outer loop of the algorithm is structured as in previous neighborhood iterator examples. Each
region in the face list is processed in turn. As each new region is processed, the input and output
iterators are initialized on that region.
The shaped iterator that ranges over the input is our structuring element and its active stencil must
be created accordingly. For this example, the structuring element is shaped like a circle of radius
element radius. Each of the appropriate neighborhood offsets is activated in the double for loop.
for ( fit=faceList.begin(); fit != faceList.end(); ++fit)
{
ShapedNeighborhoodIteratorType it( radius, reader->GetOutput(), *fit );
out = IteratorType( output, *fit );
// Creates a circular structuring element by activating all the pixels less
// than radius distance from the center of the neighborhood.
for (float y = -rad; y <= rad; y++)
{
for (float x = -rad; x <= rad; x++)
{
ShapedNeighborhoodIteratorType::OffsetType off;
float dis = std::sqrt( x*x + y*y );
if (dis <= rad)
{
off[0] = static_cast(x);
off[1] = static_cast(y);
it.ActivateOffset(off);
}
6.4. Neighborhood Iterators
189
}
}
The inner loop, which implements the erosion algorithm, is fairly simple. The for loop steps the
input and output iterators through their respective images. At each step, the active stencil of the
shaped iterator is traversed to determine whether all pixels underneath the stencil contain the foreground value, i.e. are contained within the set I. Note the use of the stencil iterator, ci, in performing
this check.
// Implements erosion
for (it.GoToBegin(), out.GoToBegin(); !it.IsAtEnd(); ++it, ++out)
{
注意：stencil内部的iterator
ShapedNeighborhoodIteratorType::ConstIterator ci;
bool flag = true;
for (ci = it.Begin(); ci != it.End(); ci++)
{
if (ci.Get() == background_value)
{
flag = false;
break;
}
}
if (flag == true)
{
out.Set(foreground_value);
}
else
{
out.Set(background_value);
}
}
}
The source code for this section can be found in the file
ShapedNeighborhoodIterators2.cxx.
The logic of the inner loop can be rewritten to perform dilation. Dilation of the set I by E is the set
of all x such that E positioned at x contains at least one element in I.
// Implements dilation
for (it.GoToBegin(), out.GoToBegin(); !it.IsAtEnd(); ++it, ++out)
{
ShapedNeighborhoodIteratorType::ConstIterator ci;
bool flag = false;
for (ci = it.Begin(); ci != it.End(); ci++)
{
if (ci.Get() != background_value)
190
Chapter 6. Iterators
Figure 6.10: The effects of morphological operations on a binary image using a circular structuring element of
size 4. From left to right are the original image, erosion, dilation, opening, and closing. The opening operation is
erosion of the image followed by dilation. Closing is dilation of the image followed by erosion.
注意：开闭操作
{
flag = true;
break;
}
}
if (flag == true)
{
out.Set(foreground_value);
}
else
{
out.Set(background_value);
}
}
}
The output image is written and visualized directly as a binary image of unsigned
chars.
Figure 6.10 illustrates some results of erosion and dilation on the image
Examples/Data/BinaryImage.png. Applying erosion and dilation in sequence effects the morphological operations of opening and closing.
CHAPTER
SEVEN
IMAGE ADAPTORS
The purpose of an image adaptor is to make one image appear like another image, possibly of
a different pixel type. A typical example is to take an image of pixel type unsigned char and
present it as an image of pixel type float. The motivation for using image adaptors in this
case is to avoid the extra memory resources required by using a casting filter. When we use the
itk::CastImageFilter for the conversion, the filter creates a memory buffer large enough to store
the float image. The float image requires four times the memory of the original image and contains no useful additional information. Image adaptors, on the other hand, do not require the extra
memory as pixels are converted only when they are read using image iterators (see Chapter 6).
Image adaptors are particularly useful when there is infrequent pixel access, since the actual conversion occurs on the fly during the access operation. In such cases the use of image adaptors
may reduce overall computation time as well as reduce memory usage. The use of image adaptors,
however, can be disadvantageous in some situations. For example, when the downstream filter is
executed multiple times, a CastImageFilter will cache its output after the first execution and will not
re-execute when the filter downstream is updated. Conversely, an image adaptor will compute the
注意：看情况使用
cast every time.
Another application for image adaptors is to perform lightweight pixel-wise operations replacing
the need for a filter. In the toolkit, adaptors are defined for many single valued and single parameter
functions such as trigonometric, exponential and logarithmic functions. For example,
• itk::ExpImageAdaptor
• itk::SinImageAdaptor
• itk::CosImageAdaptor
The following examples illustrate common applications of image adaptors.
192
Chapter 7. Image Adaptors
Filter
A
Image
X
Filter
A
Image
X
Casting Image
Filter
Y
Adaptor
Y
Filter
B
Image
Z
Filter
B
Image
Z
Figure 7.1: The difference between using a CastImageFilter and an ImageAdaptor. ImageAdaptors convert
pixel values when they are accessed by iterators. Thus, they do not produces an intermediate image. In
the example illustrated by this figure, the Image Y is not created by the ImageAdaptor; instead, the image is
simulated on the fly each time an iterator from the filter downstream attempts to access the image data.
7.1 Image Casting
The source code for this section can be found in the file
ImageAdaptor1.cxx.
This example illustrates how the itk::ImageAdaptor can be used to cast an image from one pixel
type to another. In particular, we will adapt an unsigned char image to make it appear as an image
of pixel type float.
We begin by including the relevant headers.
#include "itkImageAdaptor.h"
First, we need to define a pixel accessor class that does the actual conversion. Note that in general,
the only valid operations for pixel accessors are those that only require the value of the input pixel.
As such, neighborhood type operations are not possible. A pixel accessor must provide methods
Set() and Get(), and define the types of InternalPixelType and ExternalPixelType. The
InternalPixelType corresponds to the pixel type of the image to be adapted (unsigned char in
this example). The ExternalPixelType corresponds to the pixel type we wish to emulate with the
ImageAdaptor (float in this case).
class CastPixelAccessor
{
public:
using InternalType = unsigned char;
using ExternalType = float;
注意：图像类型转换
static void Set(InternalType & output, const ExternalType & input)
{
7.1. Image Casting
193
output = static_cast( input );
}
static ExternalType Get( const InternalType & input )
{
return static_cast( input );
}
};
The CastPixelAccessor class simply applies a static cast to the pixel values. We now use this
pixel accessor to define the image adaptor type and create an instance using the standard New()
method.
using InputPixelType = unsigned char;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< InputPixelType, Dimension >;
using ImageAdaptorType = itk::ImageAdaptor< ImageType, CastPixelAccessor >;
ImageAdaptorType::Pointer adaptor = ImageAdaptorType::New();
We also create an image reader templated over the input image type and read the input image from
file.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
The output of the reader is then connected as the input to the image adaptor.
adaptor->SetImage( reader->GetOutput() );
In the following code, we visit the image using an iterator instantiated using the adapted image type
and compute the sum of the pixel values.
using IteratorType = itk::ImageRegionIteratorWithIndex< ImageAdaptorType >;
IteratorType it( adaptor, adaptor->GetBufferedRegion() );
double sum = 0.0;
it.GoToBegin();
while( !it.IsAtEnd() )
{
float value = it.Get();
sum += value;
++it;
}
Although in this example, we are just performing a simple summation, the key concept is that access
to pixels is performed as if the pixel is of type float. Additionally, it should be noted that the adaptor
194
Chapter 7. Image Adaptors
is used as if it was an actual image and not as a filter. ImageAdaptors conform to the same API as
the itk::Image class.
7.2 Adapting RGB Images
The source code for this section can be found in the file
ImageAdaptor2.cxx.
This example illustrates how to use the itk::ImageAdaptor to access the individual components
of an RGB image. In this case, we create an ImageAdaptor that will accept a RGB image as input
and presents it as a scalar image. The pixel data will be taken directly from the red channel of the
original image.
As with the previous example, the bulk of the effort in creating the image adaptor is associated with
the definition of the pixel accessor class. In this case, the accessor converts a RGB vector to a scalar
containing the red channel component. Note that in the following, we do not need to define the
Set() method since we only expect the adaptor to be used for reading data from the image.
class RedChannelPixelAccessor
{
public:
using InternalType = itk::RGBPixel;
using ExternalType = float;
static ExternalType Get( const InternalType & input )
{
return static_cast( input.GetRed() );
}
};
The Get() method simply calls the GetRed() method defined in the itk::RGBPixel class.
Now we use the internal pixel type of the pixel accessor to define the input image type, and then
proceed to instantiate the ImageAdaptor type.
using InputPixelType = RedChannelPixelAccessor::InternalType;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< InputPixelType, Dimension >;
using ImageAdaptorType = itk::ImageAdaptor< ImageType,
RedChannelPixelAccessor >;
ImageAdaptorType::Pointer adaptor = ImageAdaptorType::New();
We create an image reader and connect the output to the adaptor as before.
7.2. Adapting RGB Images
195
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
adaptor->SetImage( reader->GetOutput() );
We create an itk::RescaleIntensityImageFilter and an itk::ImageFileWriter to rescale
the dynamic range of the pixel values and send the extracted channel to an image file. Note that the
image type used for the rescaling filter is the ImageAdaptorType itself. That is, the adaptor type is
used in the same context as an image type.
using OutputImageType = itk::Image< unsigned char, Dimension >;
using RescalerType = itk::RescaleIntensityImageFilter<
ImageAdaptorType,
OutputImageType >;
RescalerType::Pointer rescaler = RescalerType::New();
using WriterType = itk::ImageFileWriter< OutputImageType >;
WriterType::Pointer writer = WriterType::New();
Now we connect the adaptor as the input to the rescaler and set the parameters for the intensity
rescaling.
rescaler->SetOutputMinimum( 0 );
rescaler->SetOutputMaximum( 255 );
rescaler->SetInput( adaptor );
writer->SetInput( rescaler->GetOutput() );
Finally, we invoke the Update() method on the writer and take precautions to catch any exception
that may be thrown during the execution of the pipeline.
try
{
writer->Update();
}
注意：这个必须要
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception caught " << excp << std::endl;
return EXIT_FAILURE;
}
ImageAdaptors for the green and blue channels can easily be implemented by modifying the pixel
accessor of the red channel and then using the new pixel accessor for instantiating the type of an
image adaptor. The following define a green channel pixel accessor.
196
Chapter 7. Image Adaptors
Figure 7.2: Using ImageAdaptor to extract the components of an RGB image. The image on the left is a
subregion of the Visible Woman cryogenic data set. The red, green and blue components are shown from left to
right as scalar images extracted with an ImageAdaptor.
class GreenChannelPixelAccessor
{
public:
using InternalType = itk::RGBPixel;
using ExternalType = float;
static ExternalType Get( const InternalType & input )
{
return static_cast( input.GetGreen() );
}
};
A blue channel pixel accessor is similarly defined.
class BlueChannelPixelAccessor
{
public:
using InternalType = itk::RGBPixel;
using ExternalType = float;
static ExternalType Get( const InternalType & input )
{
return static_cast( input.GetBlue() );
}
};
Figure 7.2 shows the result of extracting the red, green and blue components from a region of the
Visible Woman cryogenic data set.
7.3. Adapting Vector Images
197
7.3 Adapting Vector Images
The source code for this section can be found in the file
ImageAdaptor3.cxx.
This example illustrates the use of itk::ImageAdaptor to obtain access to the components of a
vector image. Specifically, it shows how to manage pixel accessors containing internal parameters.
In this example we create an image of vectors by using a gradient filter. Then, we use an image
adaptor to extract one of the components of the vector image. The vector type used by the gradient
filter is the itk::CovariantVector class.
We start by including the relevant headers.
#include "itkGradientRecursiveGaussianImageFilter.h"
A pixel accessors class may have internal parameters that affect the operations performed on input pixel data. Image adaptors support parameters in their internal pixel accessor by using the
assignment operator. Any pixel accessor which has internal parameters must therefore implement
the assignment operator. The following defines a pixel accessor for extracting components from a
vector pixel. The m Index member variable is used to select the vector component to be returned.
class VectorPixelAccessor
{
public:
using InternalType = itk::CovariantVector;
using ExternalType = float;
VectorPixelAccessor() {}
VectorPixelAccessor & operator=( const VectorPixelAccessor & vpa ) = default;
ExternalType Get( const InternalType & input ) const
{
return static_cast( input[ m_Index ] );
}
void SetIndex( unsigned int index )
{
m_Index = index;
}
private:
unsigned int m_Index{0};
};
The Get() method simply returns the i-th component of the vector as indicated by the index. The
assignment operator transfers the value of the index member variable from one instance of the pixel
accessor to another.
In order to test the pixel accessor,
we generate an image of vectors using the
198
Chapter 7. Image Adaptors
itk::GradientRecursiveGaussianImageFilter. This filter produces an output image of
itk::CovariantVector pixel type. Covariant vectors are the natural representation for gradients
since they are the equivalent of normals to iso-values manifolds.
using InputPixelType = unsigned char;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using VectorPixelType = itk::CovariantVector< float, Dimension >;
using VectorImageType = itk::Image< VectorPixelType, Dimension >;
using GradientFilterType =
itk::GradientRecursiveGaussianImageFilter< InputImageType,
VectorImageType>;
GradientFilterType::Pointer gradient = GradientFilterType::New();
We instantiate the ImageAdaptor using the vector image type as the first template parameter and the
pixel accessor as the second template parameter.
using ImageAdaptorType = itk::ImageAdaptor< VectorImageType,
itk::VectorPixelAccessor >;
注意：搞清楚适配器的使用范围--
ImageAdaptorType::Pointer adaptor = ImageAdaptorType::New();
The index of the component to be extracted is specified from the command line. In the following, we create the accessor, set the index and connect the accessor to the image adaptor using the
SetPixelAccessor() method.
itk::VectorPixelAccessor accessor;
accessor.SetIndex( std::stoi( argv[3] ) );
adaptor->SetPixelAccessor( accessor );
We create a reader to load the image specified from the command line and pass its output as the
input to the gradient filter.
using ReaderType = itk::ImageFileReader< InputImageType >;
ReaderType::Pointer reader = ReaderType::New();
gradient->SetInput( reader->GetOutput() );
reader->SetFileName( argv[1] );
gradient->Update();
We now connect the output of the gradient filter as input to the image adaptor. The adaptor emulates
a scalar image whose pixel values are taken from the selected component of the vector image.
7.4. Adaptors for Simple Computation
199
Figure 7.3: Using ImageAdaptor to access components of a vector image. The input image on the left was
passed through a gradient image filter and the two components of the resulting vector image were extracted
using an image adaptor.
adaptor->SetImage( gradient->GetOutput() );
As in the previous example, we rescale the scalar image before writing the image out to file. Figure 7.3 shows the result of applying the example code for extracting both components of a two
dimensional gradient.
7.4 Adaptors for Simple Computation
The source code for this section can be found in the file
ImageAdaptor4.cxx.
Image adaptors can also be used to perform simple pixel-wise computations on image data. The
following example illustrates how to use the itk::ImageAdaptor for image thresholding.
A pixel accessor for image thresholding requires that the accessor maintain the threshold value.
Therefore, it must also implement the assignment operator to set this internal parameter.
class ThresholdingPixelAccessor
{
public:
using InternalType = unsigned char;
using ExternalType = unsigned char;
ThresholdingPixelAccessor() {};
ExternalType Get( const InternalType & input ) const
200
Chapter 7. Image Adaptors
{
return (input > m_Threshold) ? 1 : 0;
}
void SetThreshold( const InternalType threshold )
{
m_Threshold = threshold;
}
ThresholdingPixelAccessor &
operator=( const ThresholdingPixelAccessor & vpa ) = default;
private:
InternalType m_Threshold{0};
};
}
The Get() method returns one if the input pixel is above the threshold and zero otherwise. The
assignment operator transfers the value of the threshold member variable from one instance of the
pixel accessor to another.
To create an image adaptor, we first instantiate an image type whose pixel type is the same as the
internal pixel type of the pixel accessor.
using PixelType = itk::ThresholdingPixelAccessor::InternalType;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
We instantiate the ImageAdaptor using the image type as the first template parameter and the pixel
accessor as the second template parameter.
using ImageAdaptorType = itk::ImageAdaptor< ImageType,
itk::ThresholdingPixelAccessor >;
ImageAdaptorType::Pointer adaptor = ImageAdaptorType::New();
The threshold value is set from the command line. A threshold pixel accessor is created and connected to the image adaptor in the same manner as in the previous example.
itk::ThresholdingPixelAccessor accessor;
accessor.SetThreshold( std::stoi( argv[3] ) );
adaptor->SetPixelAccessor( accessor );
We create a reader to load the input image and connect the output of the reader as the input to the
adaptor.
7.5. Adaptors and Writers
201
Figure 7.4: Using ImageAdaptor to perform a simple image computation. An ImageAdaptor is used to perform
binary thresholding on the input image on the left. The center image was created using a threshold of 180, while
the image on the right corresponds to a threshold of 220.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
reader->Update();
adaptor->SetImage( reader->GetOutput() );
As before, we rescale the emulated scalar image before writing it out to file. Figure 7.4 illustrates the result of applying the thresholding adaptor to a typical gray scale image using two
different threshold values. Note that the same effect could have been achieved by using the
itk::BinaryThresholdImageFilter but at the price of holding an extra copy of the image in
memory.
注意：差异
7.5 Adaptors and Writers
注意：
Image adaptors will not behave correctly when connected directly to a writer. The reason is that
writers tend to get direct access to the image buffer from their input, since image adaptors do not
have a real buffer their behavior in this circumstances is incorrect. You should avoid instantiating
the ImageFileWriter or the ImageSeriesWriter over an image adaptor type.
Part III
Development Guidelines
CHAPTER
EIGHT
HOW TO WRITE A FILTER
This purpose of this chapter is help developers create their own filter (process object). This chapter
is divided into four major parts. An initial definition of terms is followed by an overview of the
filter creation process. Next, data streaming is discussed. The way data is streamed in ITK must be
understood in order to write correct filters. Finally, a section on multi-threading describes what you
must do in order to take advantage of shared memory parallel processing.
8.1 Terminology
The following is some basic terminology for the discussion that follows. Chapter 3 provides additional background information.
• The data processing pipeline is a directed graph of process and data objects. The pipeline
inputs, operators on, and outputs data.
• A filter, or process object, has one or more inputs, and one or more outputs.
• A source, or source process object, initiates the data processing pipeline, and has one or more
outputs.
• A mapper, or mapper process object, terminates the data processing pipeline. The mapper has
one or more outputs, and may write data to disk, interface with a display system, or interface
to any other system.
• A data object represents and provides access to data. In ITK, the data object (ITK class
itk::DataObject) is typically of type itk::Image or itk::Mesh.
• A region (ITK class itk::Region) represents a piece, or subset of the entire data set.
• An image region (ITK class itk::ImageRegion) represents a structured portion of data.
ImageRegion is implemented using the itk::Index and itk::Size classes
206
Chapter 8. How To Write A Filter
• A mesh region (ITK class itk::MeshRegion) represents an unstructured portion of data.
• The LargestPossibleRegion is the theoretical single, largest piece (region) that could represent the entire dataset. The LargestPossibleRegion is used in the system as the measure of the
注意：最大可能数据大小
largest possible data size.
• The BufferedRegion is a contiguous block of memory that is less than or equal to in size to
the LargestPossibleRegion. The buffered region is what has actually been allocated by a filter
注意：小于或等于最大数据大小的连续内存块
to hold its output.
注意：滤波器实际要求处理的数据大
小
• The RequestedRegion is the piece of the dataset that a filter is required to produce. The RequestedRegion is less than or equal in size to the BufferedRegion. The RequestedRegion may
differ in size from the BufferedRegion due to performance reasons. The RequestedRegion
may be set by a user, or by an application that needs just a portion of the data.
• The modified time (represented by ITK class itk::TimeStamp) is a monotonically increasing integer value that characterizes a point in time when an object was last modified.
• Downstream is the direction of dataflow, from sources to mappers.
注意：以时间的形式来表示一个对象
最后被修改的点
• Upstream is the opposite of downstream, from mappers to sources.
• The pipeline modified time for a particular data object is the maximum modified time of all
upstream data objects and process objects.
• The term information refers to metadata that characterizes data. For example, index and
dimensions are information characterizing an image region.
8.2 Overview of Filter Creation
Filters are defined with respect to the type of
data they input (if any), and the type of data
Gaussian
they output (if any). The key to writing a ITK
Reader
Filter
filter is to identify the number and types of inDataObject
ProcessObject
ProcessObject
put and output. Having done so, there are ofFigure 8.1: Relationship between DataObject and ten superclasses that simplify this task via class
derivation. For example, most filters in ITK
ProcessObject.
take a single image as input, and produce a
single image on output. The superclass itk::ImageToImageFilter is a convenience class that
注意：基类--提供了编写滤波器所需
provide most of the functionality needed for such a filter.
Image
要的基本接口，可以从他进行派生；
Some common base classes for new filters include:
• ImageToImageFilter: the most common filter base for segmentation algorithms. Takes
an image and produces a new image, by default of the same dimensions. Override
GenerateOutputInformation to produce a different size.
8.3. Streaming Large Data
207
• UnaryFunctorImageFilter: used when defining a filter that applies a function to an image.
• BinaryFunctorImageFilter: used when defining a filter that applies an operation to two
images.
• ImageFunction: a functor that can be applied to an image, evaluating f (x) at each point in
the image.
• MeshToMeshFilter: a filter that transforms meshes, such as tessellation, polygon reduction,
and so on.
• LightObject: abstract base for filters that don’t fit well anywhere else in the class hierarchy.
Also useful for “calculator” filters; i.e. a sink filter that takes an input and calculates a result
which is retrieved using a Get() method.
Once the appropriate superclass is identified, the filter writer implements the class defining the
methods required by most all ITK objects: New(), PrintSelf(), and protected constructor, copy
constructor, delete, and operator=, and so on. Also, don’t forget standard type aliases like Self,
Superclass, Pointer, and ConstPointer. Then the filter writer can focus on the most important
parts of the implementation: defining the API, data members, and other implementation details of
the algorithm. In particular, the filter writer will have to implement either a GenerateData() (nonthreaded) or ThreadedGenerateData() and DynamicThreadedGenerateData() methods. (See
Section 3.2.7 for an overview of multi-threading in ITK.)
An important note: the GenerateData() method is required to allocate memory for the output. The
ThreadedGenerateData() method is not. In default implementation (see itk::ImageSource, a
superclass of itk::ImageToImageFilter) GenerateData() allocates memory and then invokes
DynamicThreadedGenerateData() or ThreadedGenerateData().
One of the most important decisions that the developer must make is whether the filter can stream
data; that is, process just a portion of the input to produce a portion of the output. Often superclass
behavior works well: if the filter processes the input using single pixel access, then the default
behavior is adequate. If not, then the user may have to a) find a more specialized superclass to
derive from, or b) override one or more methods that control how the filter operates during pipeline
execution. The next section describes these methods.
8.3 Streaming Large Data
The data associated with multi-dimensional images is large and becoming larger. This trend is due
to advances in scanning resolution, as well as increases in computing capability. Any practical segmentation and registration software system must address this fact in order to be useful in application.
ITK addresses this problem via its data streaming facility. 注意：处理大数据
In ITK, streaming is the process of dividing data into pieces, or regions, and then processing this
data through the data pipeline. Recall that the pipeline consists of process objects that generate data
注意：1、确定基类
2、定义类所需
的必要方法--New（）
等
3、定义标准类
型别名--Self等
4、根据算法定
义处理过
程--GenerateData
（）
208
Chapter 8. How To Write A Filter
Display
Renderer
Image
Reader
Image
Gaussian
Filter
Thresholding
Image
Writer
Image
File
Image
File
Figure 8.2: The Data Pipeline
objects, connected into a pipeline topology. The input to a process object is a data object (unless
the process initiates the pipeline and then it is a source process object). These data objects in turn
are consumed by other process objects, and so on, until a directed graph of data flow is constructed.
Eventually the pipeline is terminated by one or more mappers, that may write data to storage, or
interface with a graphics or other system. This is illustrated in figures 8.1 and 8.2.
A significant benefit of this architecture is that the relatively complex process of managing pipeline
execution is designed into the system. This means that keeping the pipeline up to date, executing
only those portions of the pipeline that have changed, multi-threading execution, managing memory
allocation, and streaming is all built into the architecture. However, these features do introduce
complexity into the system, the bulk of which is seen by class developers. The purpose of this
chapter is to describe the pipeline execution process in detail, with a focus on data streaming.
8.3.1 Overview of Pipeline Execution
The pipeline execution process performs several important functions.
1. It determines which filters, in a pipeline of filters, need to execute. This prevents redundant
execution and minimizes overall execution time.
2. It initializes the (filter’s) output data objects, preparing them for new data. In addition, it
determines how much memory each filter must allocate for its output, and allocates it.
3. The execution process determines how much data a filter must process in order to produce an
output of sufficient size for downstream filters; it also takes into account any limits on memory
8.3. Streaming Large Data
209
Image
Image
Gaussian
Filter
Reader
Image
Thresholding
Update()
Update()
Update()
GenerateData()
GenerateData()
GenerateData()
Figure 8.3: Sequence of the Data Pipeline updating mechanism
or special filter requirements. Other factors include the size of data processing kernels, that
affect how much data input data (extra padding) is required.
4. It subdivides data into subpieces for multi-threading. (Note that the division of data into
subpieces is exactly same problem as dividing data into pieces for streaming; hence multithreading comes for free as part of the streaming architecture.)
5. It may free (or release) output data if filters no longer need it to compute, and the user requests
that data is to be released. (Note: a filter’s output data object may be considered a “cache”.
If the cache is allowed to remain (ReleaseDataFlagOff()) between pipeline execution, and
the filter, or the input to the filter, never changes, then process objects downstream of the filter
注意：可以缓存滤波器中将会被释放
just reuse the filter’s cache to re-execute.)
的数据；供下游滤波器使用
To perform these functions, the execution process negotiates with the filters that define the pipeline.
Only each filter can know how much data is required on input to produce a particular output. For
example, a shrink filter with a shrink factor of two requires an image twice as large (in terms of
its x-y dimensions) on input to produce a particular size output. An image convolution filter would
require extra input (boundary padding) depending on the size of the convolution kernel. Some filters
require the entire input to produce an output (for example, a histogram), and have the option of
requesting the entire input. (In this case streaming does not work unless the developer creates a filter
that can request multiple pieces, caching state between each piece to assemble the final output.)
Ultimately the negotiation process is controlled by the request for data of a particular size (i.e.,
region). It may be that the user asks to process a region of interest within a large image, or that
memory limitations result in processing the data in several pieces. For example, an application may
compute the memory required by a pipeline, and then use itk::StreamingImageFilter to break
the data processing into several pieces. The data request is propagated through the pipeline in the
upstream direction, and the negotiation process configures each filter to produce output data of a
particular size.
注意：
210
Chapter 8. How To Write A Filter
The secret to creating a streaming filter is to understand how this negotiation process works,
and how to override its default behavior by using the appropriate virtual functions defined in
itk::ProcessObject. The next section describes the specifics of these methods, and when to
override them. Examples are provided along the way to illustrate concepts.
8.3.2 Details of Pipeline Execution
Typically pipeline execution is initiated when a process object receives the
ProcessObject::Update() method invocation. This method is simply delegated to the output of the filter, invoking the DataObject::Update() method. Note that this behavior is typical
of the interaction between ProcessObject and DataObject: a method invoked on one is eventually
and data
LargestPossibleRegion
delegated to the other. Inorigin,
this spacing,
way the
request from the pipeline is propagated upstream,
initiating data flow that returns downstream.
The DataObject::Update() method in turn invokes three other methods:
• DataObject::UpdateOutputInformation()
• DataObject::PropagateRequestedRegion()
• DataObject::UpdateOutputData()
UpdateOutputInformation()
The UpdateOutputInformation() method determines the pipeline modified time. It may set the
RequestedRegion and the LargestPossibleRegion depending on how the filters are configured. (The
RequestedRegion is set to process all the data, i.e., the LargestPossibleRegion, if it has not been set.)
The UpdateOutputInformation() propagates upstream through the entire pipeline and terminates at
the sources.
During UpdateOutputInformation(),
filters have a chance to override the
ProcessObject::GenerateOutputInformation() method (GenerateOutputInformation()
is invoked by UpdateOutputInformation()).
The default behavior is for the
GenerateOutputInformation() to copy the metadata describing the input to the output
(via DataObject::CopyInformation()). Remember, information is metadata describing the
output, such as the origin, spacing, and LargestPossibleRegion (i.e., largest possible size) of an
image.
A good example of this behavior is itk::ShrinkImageFilter. This filter takes an input image
and shrinks it by some integral value. The result is that the spacing and LargestPossibleRegion of the
output will be different to that of the input. Thus, GenerateOutputInformation() is overloaded.
注意：滤波器改变输入图像的信息
（origin, spacing, and
LargestPossibleRegion等）时，需
重载这个函数--默认是使用的
RequestedRegion？？？
8.3. Streaming Large Data
211
PropagateRequestedRegion()
The PropagateRequestedRegion() call propagates upstream to satisfy a data request. In typical
application this data request is usually the LargestPossibleRegion, but if streaming is necessary, or
the user is interested in updating just a portion of the data, the RequestedRegion may be any valid
region within the LargestPossibleRegion.
The function of PropagateRequestedRegion() is, given a request for data (the amount is specified
by RequestedRegion), propagate upstream configuring the filter’s input and output process object’s
to the correct size. Eventually, this means configuring the BufferedRegion, that is the amount of
data actually allocated.
The reason for the buffered region is this: the output of a filter may be consumed by more than
one downstream filter. If these consumers each request different amounts of input (say due to kernel
requirements or other padding needs), then the upstream, generating filter produces the data to satisfy
both consumers, that may mean it produces more data than one of the consumers needs.
The ProcessObject::PropagateRequestedRegion() method invokes three methods that the filter developer may choose to overload.
• EnlargeOutputRequestedRegion(DataObject *output) gives the (filter) subclass a
chance to indicate that it will provide more data than required for the output. This can happen,
for example, when a source can only produce the whole output (i.e., the LargestPossibleRegion).
• GenerateOutputRequestedRegion(DataObject *output) gives the subclass a chance to
define how to set the requested regions for each of its outputs, given this output’s requested
region. The default implementation is to make all the output requested regions the same. A
subclass may need to override this method if each output is a different resolution. This method
is only overridden if a filter has multiple outputs.
• GenerateInputRequestedRegion() gives the subclass a chance to request a larger requested region on the inputs. This is necessary when, for example, a filter requires more
data at the “internal” boundaries to produce the boundary values - due to kernel operations or
other region boundary effects.
itk::RGBGibbsPriorFilter is an example of a filter that needs to invoke
EnlargeOutputRequestedRegion().
The designer of this filter decided that the filter should operate on all the data.
Note that a subtle interplay between this method
and GenerateInputRequestedRegion() is occurring here.
The default behavior of
GenerateInputRequestedRegion() (at least for itk::ImageToImageFilter) is to set the
input RequestedRegion to the output’s ReqestedRegion. Hence, by overriding the method
EnlargeOutputRequestedRegion() to set the output to the LargestPossibleRegion, effectively
sets the input to this filter to the LargestPossibleRegion (and probably causing all upstream filters to
process their LargestPossibleRegion as well. This means that the filter, and therefore the pipeline,
212
Chapter 8. How To Write A Filter
does not stream. This could be fixed by reimplementing the filter with the notion of streaming built
in to the algorithm.)
itk::GradientMagnitudeImageFilter is an example of a filter that needs to invoke
GenerateInputRequestedRegion(). It needs a larger input requested region because a kernel
is required to compute the gradient at a pixel. Hence the input needs to be “padded out” so the filter
has enough data to compute the gradient at each output pixel.
UpdateOutputData()
UpdateOutputData() is the third and final method as a result of the Update() method. The purpose
of this method is to determine whether a particular filter needs to execute in order to bring its output
up to date. (A filter executes when its GenerateData() method is invoked.) Filter execution occurs
when a) the filter is modified as a result of modifying an instance variable; b) the input to the filter
changes; c) the input data has been released; or d) an invalid RequestedRegion was set previously
and the filter did not produce data. Filters execute in order in the downstream direction. Once a filter
executes, all filters downstream of it must also execute.
DataObject::UpdateOutputData() is delegated to the DataObject’s source (i.e., the ProcessObject that generated it) only if the DataObject needs to be updated. A comparison of modified time,
pipeline time, release data flag, and valid requested region is made. If any one of these conditions indicate that the data needs regeneration, then the source’s ProcessObject::UpdateOutputData()
is invoked. These calls are made recursively up the pipeline until a source filter object is encountered, or the pipeline is determined to be up to date and valid. At this point, the recursion unrolls,
and the execution of the filter proceeds. (This means that the output data is initialized, StartEvent is
invoked, the filters GenerateData() is called, EndEvent is invoked, and input data to this filter may
be released, if requested. In addition, this filter’s InformationTime is updated to the current time.)
The developer will never override UpdateOutputData(). The developer need only write the
GenerateData() method (non-threaded) or DynamicThreadedGenerateData() method. A discussion on threading follows in the next section.
8.4 Threaded Filter Execution
Filters that can process data in pieces can typically multi-process using the data parallel, shared
memory implementation built into the pipeline execution process. To create a multi-threaded
filter, simply define and implement a DynamicThreadedGenerateData(). For example, a
itk::ImageToImageFilter would create the method:
void DynamicThreadedGenerateData( const OutputImageRegionType&
outputRegionForThread ) override;
The key to threading is to generate output for the output region given as the parameter. In ITK, this
8.5. Filter Conventions
213
is simple to do because an output iterator can be created using the region provided. Hence the output
can be iterated over, accessing the corresponding input pixels as necessary to compute the value of
the output pixel.
Multi-threading requires caution when performing I/O (including using cout or cerr) or invoking
events. A safe practice is to allow only the invoking thread to perform I/O or generate events. If
more than one thread tries to write to the same place at the same time, the program can behave badly,
and possibly even deadlock or crash.
DynamicThreadedGenerateData signature allows number of pieces (output regions) to be processed to be different, usually bigger than the number of real threads executing the work. In turn, this
allows load balancing. The number of work units controls filter parallelism, and the name ‘threads’
is reserved for real threads as exposed by itk::MultiThreaderBase and its descendants.
8.5 Filter Conventions
In order to fully participate in the ITK pipeline, filters are expected to follow certain conventions, and
provide certain interfaces. This section describes the minimum requirements for a filter to integrate
into the ITK framework.
注意：
A filter should define public types for the class itself (Self) and its Superclass, and const and
non-const smart pointers, thus:
using
using
using
using
Self = ExampleImageFilter;
Superclass = ImageToImageFilter;
Pointer = SmartPointer;
ConstPointer = SmartPointer;
The Pointer type is particularly useful, as it is a smart pointer that will be used by all client code
to hold a reference-counted instantiation of the filter.
Once the above types have been defined, you can use the following convenience macros, which
permit your filter to participate in the object factory mechanism, and to be created using the canonical
::New():
/** Method for creation through the object factory. */
itkNewMacro(Self);
/** Run-time type information (and related methods). */
itkTypeMacro(ExampleImageFilter, ImageToImageFilter);
The default constructor should be protected, and provide sensible defaults (usually zero) for all
parameters. The copy constructor and assignment operator should not implemented in order to
prevent instantiating the filter without the factory methods (above). They should be declared in
注意：
214
Chapter 8. How To Write A Filter
the public section using the ITK DISALLOW COPY AND ASSIGN macro (see Section C.17 on page
373).
Finally, the template implementation code (in the .hxx file) should be included, bracketed by a test
for manual instantiation, thus:
#ifndef ITK_MANUAL_INSTANTIATION
#include "itkExampleFilter.hxx"
#endif
8.5.1 Optional
A filter can be printed to an std::ostream (such as std::cout) by implementing the following
method:
void PrintSelf( std::ostream& os, Indent indent ) const;
and writing the name-value pairs of the filter parameters to the supplied output stream. This is
particularly useful for debugging.
8.5.2 Useful Macros
Many convenience macros are provided by ITK, to simplify filter coding. Some of these are described below:
itkStaticConstMacro Declares a static variable of the given type, with the specified initial value.
itkGetMacro Defines an accessor method for the specified scalar data member. The convention is
for data members to have a prefix of m .
itkSetMacro Defines a mutator method for the specified scalar data member, of the supplied type.
This will automatically set the Modified flag, so the filter stage will be executed on the next
Update().
itkBooleanMacro Defines a pair of OnFlag and OffFlag methods for a boolean variable m Flag.
itkGetConstObjectMacro, itkSetObjectMacro Defines an accessor and mutator for an ITK object. The Get form returns a smart pointer to the object.
Much more useful information can be learned from browsing the source
Code/Common/itkMacro.h and for the itk::Object and itk::LightObject classes.
in
8.6. How To Write A Composite Filter
215
Composite
Source
Stage1
Stage2
Stage...n
Sink
Figure 8.4: A Composite filter encapsulates a number of other filters.
8.6 How To Write A Composite Filter
In general, most ITK filters implement one particular algorithm, whether it be image filtering, an
information metric, or a segmentation algorithm. In the previous section, we saw how to write new
filters from scratch. However, it is often very useful to be able to make a new filter by combining
two or more existing filters, which can then be used as a building block in a complex pipeline. This
approach follows the Composite pattern [3], whereby the composite filter itself behaves just as a
regular filter, providing its own (potentially higher level) interface and using other filters (whose
detail is hidden to users of the class) for the implementation. This composite structure is shown in
Figure 8.4, where the various Stage-n filters are combined into one by the Composite filter. The
Source and Sink filters only see the interface published by the Composite. Using the Composite
pattern, a composite filter can encapsulate a pipeline of arbitrary complexity. These can in turn be
nested inside other pipelines.
8.6.1 Implementing a Composite Filter
There are a few considerations to take into account when implementing a composite filter. All the
usual requirements for filters apply (as discussed above), but the following guidelines should be
considered:
1. The template arguments it takes must be sufficient to instantiate all of the component filters.
Each component filter needs a type supplied by either the implementor or the enclosing class.
For example, an ImageToImageFilter normally takes an input and output image type (which
may be the same). But if the output of the composite filter is a classified image, we need to
either decide on the output type inside the composite filter, or restrict the choices of the user
when she/he instantiates the filter.
2. The types of the component filters should be declared in the header, preferably with
protected visibility. This is because the internal structure normally should not be visible to
users of the class, but should be to descendent classes that may need to modify or customize
the behavior.
注意：使用多个
滤波器组合成新
的滤波器
216
Chapter 8. How To Write A Filter
CompositeExampleImageFilter
Reader
Writer
Gradient
Threshold
Rescale
Figure 8.5: Example of a typical composite filter. Note that the output of the last filter in the internal pipeline
must be grafted into the output of the composite filter.
3. The component filters should be private data members of the composite class, as in
FilterType::Pointer.
4. The default constructor should build the pipeline by creating the stages and connect them
together, along with any default parameter settings, as appropriate.
5. The input and output of the composite filter need to be grafted on to the head and tail (respectively) of the component filters.
This grafting process is illustrated in Figure 8.5.
8.6.2 A Simple Example
The source code for this section can be found in the file
CompositeFilterExample.cxx.
The composite filter we will build combines three filters: a gradient magnitude operator, which will
calculate the first-order derivative of the image; a thresholding step to select edges over a given
strength; and finally a rescaling filter, to ensure the resulting image data is visible by scaling the
intensity to the full spectrum of the output image type.
Since this filter takes an image and produces another image (of identical type), we will specialize
the ImageToImageFilter:
Next we include headers for the component filters:
#include "itkGradientMagnitudeImageFilter.h"
#include "itkThresholdImageFilter.h"
#include "itkRescaleIntensityImageFilter.h"
Now we can declare the filter itself. It is within the ITK namespace, and we decide to make it
use the same image type for both input and output, so that the template declaration needs only one
parameter. Deriving from ImageToImageFilter provides default behavior for several important
aspects, notably allocating the output image (and making it the same dimensions as the input).
8.6. How To Write A Composite Filter
217
namespace itk
{
template < typename TImage >
class CompositeExampleImageFilter :
public ImageToImageFilter< TImage, TImage >
{
public:
ITK_DISALLOW_COPY_AND_ASSIGN(CompositeExampleImageFilter);
Next we have the standard declarations, used for object creation with the object factory:
using
using
using
using
Self = CompositeExampleImageFilter;
Superclass = ImageToImageFilter< TImage, TImage >;
Pointer = SmartPointer< Self >;
ConstPointer = SmartPointer< const Self >;
Here we declare an alias (to save typing) for the image’s pixel type, which determines the type of
the threshold value. We then use the convenience macros to define the Get and Set methods for this
parameter.
using ImageType = TImage;
using PixelType = typename ImageType::PixelType;
itkGetMacro( Threshold, PixelType );
itkSetMacro( Threshold, PixelType );
Now we can declare the component filter types, templated over the enclosing image type:
protected:
using ThresholdType = ThresholdImageFilter< ImageType >;
using GradientType = GradientMagnitudeImageFilter< ImageType, ImageType >;
using RescalerType = RescaleIntensityImageFilter< ImageType, ImageType >;
The component filters are declared as data members, all using the smart pointer types.
typename GradientType::Pointer
typename ThresholdType::Pointer
typename RescalerType::Pointer
PixelType m_Threshold;
};
} // end namespace itk
m_GradientFilter;
m_ThresholdFilter;
m_RescaleFilter;
218
Chapter 8. How To Write A Filter
The constructor sets up the pipeline, which involves creating the stages, connecting them together,
and setting default parameters.
template< typename TImage >
CompositeExampleImageFilter< TImage >
::CompositeExampleImageFilter()
{
m_Threshold = 1;
m_GradientFilter = GradientType::New();
m_ThresholdFilter = ThresholdType::New();
m_ThresholdFilter->SetInput( m_GradientFilter->GetOutput() );
m_RescaleFilter = RescalerType::New();
m_RescaleFilter->SetInput( m_ThresholdFilter->GetOutput() );
m_RescaleFilter->SetOutputMinimum(
NumericTraits::NonpositiveMin());
m_RescaleFilter->SetOutputMaximum(NumericTraits::max());
}
The GenerateData() is where the composite magic happens.
First, connect the first component filter to the inputs of the composite filter (the actual input, supplied
by the upstream stage). At a filter’s GenerateData() stage, the input image’s information and
pixel buffer content have been updated by the pipeline. To prevent the mini-pipeline update from
propagating upstream, the input image is disconnected from the pipeline by grafting its contents to
a new itk::Image pointer.
This implies that the composite filter must implement pipeline methods that indicate the
itk::ImageRegion’s it requires and generates, like GenerateInputRequestedRegion(),
GenerateOutputRequestedRegion(),
GenerateOutputInformation()
and
EnlargeOutputRequestedRegion(), according to the behavior of its component filters.
Next, graft the output of the last stage onto the output of the composite, which ensures the requested
region is updated and the last stage populates the output buffer allocated by the composite filter. We
force the composite pipeline to be processed by calling Update() on the final stage. Then, graft the
output back onto the output of the enclosing filter, so it has the result available to the downstream
filter.
template< typename TImage >
void
CompositeExampleImageFilter< TImage >
::GenerateData()
{
typename ImageType::Pointer input = ImageType::New();
input->Graft( const_cast< ImageType * >( this->GetInput() ));
m_GradientFilter->SetInput( input );
m_ThresholdFilter->ThresholdBelow( this->m_Threshold );
m_RescaleFilter->GraftOutput( this->GetOutput() );
m_RescaleFilter->Update();
8.6. How To Write A Composite Filter
219
this->GraftOutput( m_RescaleFilter->GetOutput() );
}
Finally we define the PrintSelf method, which (by convention) prints the filter parameters. Note
how it invokes the superclass to print itself first, and also how the indentation prefixes each line.
template< typename TImage >
void
CompositeExampleImageFilter< TImage >
::PrintSelf( std::ostream& os, Indent indent ) const
{
Superclass::PrintSelf(os,indent);
os << indent << "Threshold:" << this->m_Threshold
<< std::endl;
}
} // end namespace itk
It is important to note that in the above example, none of the internal details of the pipeline were
exposed to users of the class. The interface consisted of the Threshold parameter (which happened
to change the value in the component filter) and the regular ImageToImageFilter interface. This
example pipeline is illustrated in Figure 8.5.
CHAPTER
NINE
HOW TO CREATE A MODULE
The Insight Toolkit is organized into logical units of coherent functionality called modules. These
modules are self-contained in a directory, whose components are organized into subdirectories with
standardized names. A module usually has dependencies on other modules, which it declares. A
module is defined with CMake scripts that inform the build system of its contents and dependencies.
The modularization effort significantly improves the extensibility of the toolkit and lowers the barrier
to contribution.
Modules are organized into:
注意：模块的组织结构--切记；
• The top level directory.
• The include directory.
• The src directory.
• The test directory.
• The wrapping directory.
This chapter describes how to create a new module. The following sections are organized by the
different directory components of the module. The chapter concludes with a section on how to add
a third-party library dependency to a module.
Note that the Insight Toolkit community has adopted a Coding Style guideline for the sake of consistentcy and readability of the code. Such guideline is described in Chapter C.
9.1 Name and dependencies
The top level directory of a module is used to define a module’s name and its dependencies. Two
files are required:
222
Chapter 9. How To Create A Module
1. CMakeLists.txt
2. itk-module.cmake
The information described in these files is used to populate .cmake
files in the ITK module registry.
The module registry is located at /lib/cmake/5.0/Modules/
in a build tree or /lib/cmake/5.0/Modules/ in an install tree. These module files declare information
about the module and what is required to use it. This includes its module dependencies, C++ include
directories required to build against it, the libraries to link against, and CMake code required to use
it in a CMake configured project.
9.1.1 CMakeLists.txt
When CMake starts processing a module, it begins with the top level CMakeLists.txt file. At a
minimum, the CMakeLists.txt should contain
cmake_minimum_required(VERSION 3.10.2 FATAL_ERROR)
cmake_policy(VERSION 3.10.2)
project(MyModule)
set(MyModule_LIBRARIES MyModule)
if(NOT ITK_SOURCE_DIR)
find_package(ITK REQUIRED)
list(APPEND CMAKE_MODULE_PATH ${ITK_CMAKE_DIR})
include(ITKModuleExternal)
else()
itk_module_impl()
endif()
where MyModule is the name of the module.
The CMake variable LIBRARIES should be set to the names of the libraries, if any,
that clients of the module need to link. This will be the same name as the library generated with
the add library command in a module’s src directory, described in further detail in the Libraries
Section 9.3.
The path if(NOT ITK SOURCE DIR) is used when developing a module outside of the ITK source
tree, i.e. an External module. An External module can be made available to the community by
adding it to Modules/Remote/*.remote.cmake Remote module index in the ITK repository per
Section 10.1.
The CMake macro itk module impl is defined in the file CMake/ITKModuleMacros.cmake.
It will initiate processing of the remainder of a module’s CMake scripts.
The script
ITKModuleExternal calls itk module impl internally.
9.1. Name and dependencies
223
9.1.2 itk-module.cmake
The itk-module.cmake is also a required CMake script at the top level of a module, but this file is
used to declare
1. The module name.
2. Dependencies on other modules.
3. Modules properties.
4. A description of the module.
In this file, first set a CMake variable with the module’s description followed by a call to the
itk module macro, which is already defined by the time the script is read. For example,
itk-module.cmake for the ITKCommon module is
set(DOCUMENTATION "This module contains the central classes of the ITK
toolkit. They include, basic data structures \(such as Points, Vectors,
Images, Regions\) the core of the process objects \(such as base
classes for image filters\) the pipeline infrastructure classes, the support
for multi-threading, and a collection of classes that isolate ITK from
platform specific features. It is anticipated that most other ITK modules will
depend on this one.")
itk_module(ITKCommon
ENABLE_SHARED
PRIVATE_DEPENDS
ITKDoubleConversion
COMPILE_DEPENDS
ITKKWSys
ITKVNLInstantiation
TEST_DEPENDS
ITKTestKernel
ITKMesh
ITKImageIntensity
ITKIOImageBase
DESCRIPTION
"${DOCUMENTATION}"
)
The description for the module should be escaped as a CMake string, and it should be formatted
with Doxygen markup. This description is added to ITK’s generated Doxygen documentation when
the module is added to the Remote module index. The description should describe the purpose and
content of the module and reference an Insight Journal article for further information.
A module name is the only required positional argument to the itk module macro. Named options
that take one or argument are:
224
Chapter 9. How To Create A Module
DEPENDS Modules that will be publicly linked to this module. The header’s used are added to
include/*.{h,hxx} files.
PRIVATE DEPENDS Modules that will be privately linked to this module. The header’s used are
only added to src/*.cxx files.
COMPILE DEPENDS Modules that are needed at compile time by this module. The header’s
used are added to include/*{h,hxx} files but there is not a library to link against.
TEST DEPENDS Modules that are needed by this modules testing executables. The header’s
used are added to test/*.cxx files.
DESCRIPTION Free text description of the module.
Public dependencies are added to the module’s INTERFACE LINK LIBRARIES, which is a list of
transitive link dependencies. When this module is linked to by another target, the libraries listed (and
recursively, their link interface libraries) will be provided to the target also. Private dependencies
are linked to by this module, but not added to INTERFACE LINK LIBRARIES.
Compile Dependencies are added to CMake’s list of dependencies for the current module, ensuring
that they are built before the current module, but they will not be linked either publicly or privately.
They are only used to support the building of the current module.
The following additional options take no arguments:
EXCLUDE FROM DEFAULT Exclude this module from collection of modules enabled with
the ITK BUILD DEFAULT MODULES CMake option.
ENABLE SHARED Build this module as a shared library if the BUILD SHARED LIBS CMake
option is set.
All External and Remote modules should set the EXCLUDE FROM DEFAULT option.
9.2 Headers
Headers for the module, both *.h declaration headers and *.hxx template definition headers, should
be added to the include directory. No other explicit CMake configuration is required.
This path will automatically be added to the build include directory paths for libraries (9.3) and tests
(9.4) in the module and when another module declares this module as a dependency.
When a module is installed, headers are installed into a single directory common to all ITK header
files.
When BUILD TESTING is enabled, a header test is automatically created. This test simply builds
a simple executable that #includes all header files in the include directory. This ensures that all
included headers can be found, which tests the module’s dependency specification per Section 9.1.
9.3. Libraries
225
9.3 Libraries
Libraries generated by a module are created from source files with the .cxx extension in a module’s
src directory. Some modules are header-only, and they will not generate any libraries; in this case,
the src directory is omitted. When present, the src directory should contain a CMakeLists.txt
file that describes how to build the library. A minimal CMakeLists.txt file is as follows.
set(AModuleName_SRCS
itkFooClass.cxx
itkBarClass.cxx
)
itk_module_add_library(AModuleName ${AModuleName_SRCS})
The itk module add library macro will create a library with the given sources. The macro will
also link the library to the libraries defined by the module dependency specification per Section 9.1.
Additionally, the macro will set CMake target properties associated with the current module to the
given target.
If the ENABLE SHARED option is set on a module, a shared library will be generated when the CMake
option BUILD SHARED LIBS is enabled. A library symbol export specification header is also generated for the module. For a module with the name AModuleName, the generated header will have the
name AModuleNameExport.h. Include the export header in the module source headers, and add the
export specification macro to the contained classes. The macro name in this case would be called
注意：外部link库
AModuleName EXPORT. For example, the file itkFooClass.h would contain
#include "AModuleNameExport.h"
namespace itk
{
class AModuleName_EXPORT FooClass
{
...
Modules that do not build a library in their src directory or do not have export specifications on
their class declarations should not set ENABLE SHARED.
9.4 Tests
Regression tests for a module are placed in the test directory. This directory will contain a
CMakeLists.txt with the CMake configuration, test sources, and optional Input and Baseline
directories, which contain test input and baseline image datasets, respectively. Placement of the
input and baseline image datasets within a given module directory is preferred over placement in
226
Chapter 9. How To Create A Module
the general Testing/Data directory; this ensures that a module’s data is only downloaded when
the module is enabled. An exception to this rule may be widely used input datasets, such as the
cthead1.png image.
An example CMake configuration for a test directory is shown below.
itk_module_test()
set(ModuleTemplateTests
itkMinimalStandardRandomVariateGeneratorTest.cxx
itkLogNormalDistributionImageSourceTest.cxx
)
CreateTestDriver(ModuleTemplate "${ModuleTemplate-Test_LIBRARIES}" "${ModuleTemplateTests}")
itk_add_test(NAME itkMinimalStandardRandomVariateGeneratorTest
COMMAND ModuleTemplateTestDriver itkMinimalStandardRandomVariateGeneratorTest
)
itk_add_test(NAME itkLogNormalDistributionImageSourceTest
COMMAND ModuleTemplateTestDriver --without-threads
--compare
${ITK_TEST_OUTPUT_DIR}/itkLogNormalDistributionImageSourceTestOutput.mha
DATA{Baseline/itkLogNormalDistributionImageSourceTestOutput.mha}
itkLogNormalDistributionImageSourceTest
${ITK_TEST_OUTPUT_DIR}/itkLogNormalDistributionImageSourceTestOutput.mha
)
The CMakeLists.txt file should start with a call to the itk module test macro. Next, the test
sources are listed. The naming convention for unit test files is itkTest.cxx. Each
test file should be written like a command line executable, but the name of the main function should
be replaced with the name of the test. The function should accept int argc, char * argv[] as
arguments. To reduce the time required for linking and to provide baseline comparison functionality, all tests are linked to into a single test driver executable. To generate the executable, call the
CreateTestDriver macro.
Tests are defined with the itk add test macro. This is a wrapper around the CMake add test
command that will resolve content links in the DATA macro. Testing data paths are given inside the
DATA macro. Content link files, stored in the source code directory, are replaced by actual content
files in the build directory when CMake downloads the ITKData target at build time. A content
link file has the same name as its target, but a .sha512 extension is added, and the .sha512 file’s
contents are only the SHA512 hash of its target. Content links for data files in a Git distributed
version control repository prevent repository bloat. To obtain content links, register an account at
https://data.kitware.com. Upload images to your account’s My folders/Public folder. Once
the image has been uploaded, click on the item’s link, then click the Show info icon. A Download
key file icon will be available to download the content link. Place this file in the repository tree
注意：？？？
where referenced by the DATA macro.
When a test requires a new (or modified) input or baseline image dataset, the corresponding content
9.4. Tests
227
link files have to be provided as well. Image datasets provided should be kept as small as possible.
As a rule of thumb, their size should be under 50 kB.
Test commands should call the test driver executable, followed by options for the test, followed by
the test function name, followed by arguments that are passed to the test. The test driver accepts
options like --compare (or --compare-MD5 when using the MD5SUM hash) to compare output
images to baselines or options that modify tolerances on comparisons. An exhaustive list of options
is displayed in itkTestDriverInclude.h.
A few rules must be acknowledged to actually write a units test file itkTest.cxx for
a given ITK class:
1. All class methods must be exercised.
2. Test cases with values for member variables different from the default ones should be provided. The usefulness of this rule is especially manifest for boolean members, whose value
usually determines whether a large portion of code is exercised or not.
3. Test cases to reach the exception cases within the class should be provided.
4. Regression tests must be included for methods returning a value.
5. When a test detects a failure condition it must return the EXIT FAILURE value; if a test exits
normally, it must return the EXIT SUCCESS value.
In any case, ITK provides with a number of classes and macros that ease the process of writing tests
and checking the expected results. The following is an exhaustive list of such tools:
• itkTestingMacros.h: it contains a number of macros that allow testing of basic object
properties:
– EXERCISE BASIC OBJECT METHODS(): verifies whether the class and superclass
names provided match the RTTI, and exercises the PrintSelf() method. Since the
PrintSelf() method prints all class member variables, this macro, when exercised,
can identify uninitialized member variables.
– TEST SET GET VALUE(): once a member variable value has been set using the corresponding Set macro, this macro verifies that the value provided to the Set() method was
effectively assigned to the member variable by comparing it to the value returned by the
Get() value.
– TEST SET GET BOOLEAN(): exercises the Set()/Get(), and On()/Off() methods of
class applied to a boolean member variable.
• TRY EXPECT NO EXCEPTION(): exercises a method which is expected to return with no errors. It is only required for methods that are known to throw exceptions, such as I/O operations, filter updates, etc.
228
Chapter 9. How To Create A Module
• TRY EXPECT EXCEPTION(): exercises a method in the hope of detecting an exception. This
macro allows a test to continue its execution when setting test cases bound to hit a class’
exception cases. It is only required for methods that are known to throw exceptions, such as
I/O operations, filter updates, etc.
• itkMath.h: contains a series of static methods used for basic type comparison. Methods are available to perform fuzzy floating point equality comparison, e.g.
itk::Math::FloatAlmostEquals(), to handle expected cross-platform differences.
A test may have some input arguments. When a test does not need any input argument (e.g., it generates a synthetic input image), the main argument names may either
be omitted (int itkTest( int, char* [] )), or the itkNotUsed macro can
be used (int itkTest( int itkNotUsed( argc ), char *itkNotUsed( argv
) [] )), to avoid compiler warnings about unused variables.
The number of input arguments provided must be checked at the beginning of the test. If a test
requires a fixed number of input arguments, then the argument number check should verify the exact
number of arguments.
It is essential that a test is made quantitative, i.e., the methods’ returned values and the test’s output
must be compared to a known ground-truth. As mentioned, ITK contains a series of methods to
compare basic types. ITK also provide a powerful regression tool for a test that checks the validity
of a process over an image, which is the most common case in ITK. To this end, the test is expected
to write its output to a file. The first time the test is run, the output is expected to be manually placed
within the test module’s Baseline folder. Hence, when CTest is executed, the distance between
the test’s output and the expected output (i.e., the baseline) is computed. If the distance is below a
configurable tolerance, the regression test is marked as a success.
9.5 Wrapping
Wrapping for programming languages like Python can be added to a module through a simple configuration in the module’s wrapping directory. While wrapping is almost entirely automatic, configuration is necessary to add two pieces of information,
1. The types with which to instantiate templated classes.
2. Class dependencies which must be wrapped before a given class.
When wrapping a class, dependencies, like the base class and other types used in the wrapped class’s
interface, should also be wrapped. The wrapping system will emit a warning when a base class
or other required type is not already wrapped to ensure proper wrapping coverage. Since module
dependencies are wrapped by the build system before the current module, class wrapping build order
is already correct module-wise. However, it may be required to wrap classes within a module in a
specific order; this order can be specified in the wrapping/CMakeLists.txt file.
9.5. Wrapping
229
Many ITK classes are templated, which allows an algorithm to be written once yet compiled into
optimized binary code for numerous pixel types and spatial dimensions. When wrapping these
templated classes, the template instantiations to wrap must be chosen at build time. The template
that should be used are configured in a module’s *.wrap files. Wrapping is configured by calling
CMake macros defined in the ITK/Wrapping/TypedefMacros.cmake file.
9.5.1 CMakeLists.txt
The wrapping/CMakeLists.txt file calls three macros, and optionally set a variable, WRAPPER SUBMODULE ORDER. The following example is from the ITKImageFilterBase module:
itk_wrap_module(ITKImageFilterBase)
set(WRAPPER_SUBMODULE_ORDER
itkRecursiveSeparableImageFilter
itkFlatStructuringElement
itkKernelImageFilter
itkMovingHistogramImageFilterBase
)
itk_auto_load_submodules()
itk_end_wrap_module()
The itk wrap module macro takes the current module name as an argument. In some cases,
classes defined in the *.wrap files within a module may depend each other. The WRAPPER SUBMODULE ORDER variable is used to declare which submodules should be wrapped first and the
order they should be wrapped.
9.5.2 Class wrap files
Wrapping specification for classes is written in the module’s *.wrap CMake script files. These
files call wrapping CMake macros, and they specify which classes to wrap, whether smart pointer’s
should be wrapped for the the class, and which template instantiations to wrap for a class.
Overall toolkit class template instantiations are parameterized by the CMake build configuration
variables shown in Table 9.1. The wrapping configuration refers to these settings with the shorthand
values listed in the second column.
Class wrap files call sets of wrapping macros for the class to be wrapped. The macros are
often called in loops over the wrapping variables to instatiate the desired types. The following example demonstates wrapping the itk::ImportImageFilter class, taken from the
ITK/Modules/Core/Common/wrapping/itkImportImageFilter.wrap file.
itk_wrap_class("itk::ImportImageFilter" POINTER)
foreach(d ${ITK_WRAP_IMAGE_DIMS})
230
Chapter 9. How To Create A Module
CMake variable
ITK WRAP IMAGE DIMS
ITK WRAP VECTOR COMPONENTS
ITK WRAP double
ITK WRAP float
ITK WRAP complex double
ITK WRAP complex float
ITK WRAP vector double
ITK WRAP vector float
ITK WRAP covariate vector double
ITK WRAP covariate vector float
ITK WRAP signed char
ITK WRAP signed short
ITK WRAP signed long
ITK WRAP unsigned char
ITK WRAP unsigned short
ITK WRAP unsigned long
ITK WRAP rgb unsigned char
ITK WRAP rgb unsigned short
ITK WRAP rgba unsigned char
ITK WRAP rgba unsigned short
Wrapping shorthand value
List of unsigned integers
List of unsigned integers
D
F
CD
CF
VD
VF
CVD
CVF
SC
SS
SL
UC
US
UL
RGBUC
RGBUS
RGBAUC
RGBAUS
Table 9.1: CMake wrapping type configuration variables and their shorthand value in the wrapping configuration.
9.5. Wrapping
231
foreach(t ${WRAP_ITK_SCALAR})
itk_wrap_template("${ITKM_${t}}${d}" "${ITKT_${t}},${d}")
endforeach()
endforeach()
itk_end_wrap_class()
Wrapping Variables
Instantiations for classes are determined by looping over CMake lists that collect sets of shorthand
wrapping values, namely,
• ITK WRAP IMAGE DIMS
• ITK WRAP IMAGE DIMS INCREMENTED
• ITK WRAP IMAGE VECTOR COMPONENTS
• ITK WRAP IMAGE VECTOR COMPONENTS INCREMENTED
• WRAP ITK USIGN INT
• WRAP ITK SIGN INT
• WRAP ITK INT
• WRAP ITK REAL
• WRAP ITK COMPLEX REAL
• WRAP ITK SCALAR
• WRAP ITK VECTOR REAL
• WRAP ITK COV VECTOR REAL
• WRAP ITK VECTOR
• WRAP ITK RGB
• WRAP ITK RGBA
232
Chapter 9. How To Create A Module
• WRAP ITK COLOR
• WRAP ITK ALL TYPES
Templated classes are wrapped as type aliases for particular instantiations. The type aliases are
named with a name mangling scheme for the template parameter types. The mangling of common
types are stored in CMake variables listed in Table 9.2, Table 9.3, and Table 9.4. Mangling variables
start with the prefix ITKM and their corresponding C++ type variables start with the prefix ITKT .
注意：？？？不是很明白
Wrapping Macros
There are a number of a wrapping macros called in the wrapping/*.wrap files. Macros are specialized for classes that use itk::SmartPointers and templated classes.
For non-templated classes, the itk wrap simple class is used. This macro takes fully qualified name of the class as an argument. Lastly, the macro takes an optional argument that can
have the values POINTER, POINTER WITH CONST POINTER, or POINTER WITH SUPERCLASS. If
this argument is passed, then the type alias classname::Pointer, classname::Pointer and
classname::ConstPointer, or classname::Pointer and classname::Superclass::Pointer
are wrapped. Thus, the wrapping configuration for itk::Object is
itk_wrap_simple_class("itk::Object" POINTER)
When wrapping templated classes, three or more macro calls are required. First, itk wrap class is called. Again, its arguments are the fully qualified followed by an option argument that can have the value POINTER, POINTER WITH CONST POINTER, POINTER WITH SUPERCLASS, POINTER WITH 2 SUPERCLASSES, EXPLICIT SPECIALIZATION, POINTER WITH EXPLICIT SPECIALIZATION, ENUM, or AUTOPOINTER. Next, a series of calls are made to macros that
declare which templates to instantiate. Finally, the itk end wrap class macro is called, which has
no arguments.
The most general template wrapping macro is itk wrap template. Two arguments are required.
The first argument is a mangled suffix to be added to the class name, which uniquely identifies the instantiation. This argument is usually specified at least partially with ITKM mangling
variables. The second argument is the is template instantiation in C++ form. This argument is
usually specified at least partially with ITKT C++ type variables. For example, wrapping for
itk::ImageSpatialObject, which templated a dimension and pixel type, is configured as
itk_wrap_class("itk::ImageSpatialObject" POINTER)
# unsigned char required for the ImageMaskSpatialObject
UNIQUE(types "UC;${WRAP_ITK_SCALAR}")
foreach(d ${ITK_WRAP_IMAGE_DIMS})
foreach(t ${types})
9.5. Wrapping
233
Mangling
C++ Type
CMake Variable
ITKM B
ITKT B
Value
B
bool
Mangling
C++ Type
ITKM UC
ITKT UC
UC
unsigned char
Mangling
C++ Type
ITKM US
ITKT US
US
unsigned short
Mangling
C++ Type
ITKM UI
ITKT UI
UI
unsigned integer
Mangling
C++ Type
ITKM UL
ITKT UL
UL
unsigned long
Mangling
C++ Type
ITKM SC
ITKT SC
SC
signed char
Mangling
C++ Type
ITKM SS
ITKT SS
SS
signed short
Mangling
C++ Type
ITKM SI
ITKT SI
SI
signed integer
Mangling
C++ Type
ITKM UL
ITKT UL
UL
signed long
Mangling
C++ Type
ITKM F
ITKT F
F
float
Mangling
C++ Type
ITKM D
ITKT D
D
double
Table 9.2: CMake wrapping mangling variables, their values, and the corresponding CMake C++ type variables
and their values for plain old datatypes (PODS).
234
Chapter 9. How To Create A Module
CMake Variable
Value
Mangling
C++ Type
ITKM C${type}
ITKT C${type}
C${type}
std::complex<${type} >
Mangling
C++ Type
ITKM A${type}
ITKT A${type}
A${type}
itk::Array<${type} >
Mangling
ITKM FA${ITKM ${type}}${dim}
ITKT FA${ITKM ${type}}${dim}
FA${ITKM ${type}}${dim}
Mangling
C++ Type
ITKM RGB${dim}
ITKT RGB${dim}
RGB${dim}
itk::RGBPixel<${dim} >
Mangling
C++ Type
ITKM RGBA${dim}
ITKT RGBA${dim}
RGBA${dim}
itk::RGBAPixel<${dim} >
Mangling
ITKM V${ITKM ${type}}${dim}
ITKT V${ITKM ${type}}${dim}
V${ITKM ${type}}${dim}
ITKM CV${ITKM ${type}}${dim}
ITKT CV${ITKM ${type}}${dim}
CV${ITKM ${type}}${dim}
ITKM VLV${ITKM ${type}}${dim}
ITKT VLV${ITKM ${type}}${dim}
VLV${ITKM ${type}}${dim}
ITKM SSRT${ITKM ${type}}${dim}
ITKT SSRT${ITKM ${type}}${dim}
SSRT${ITKM ${type}}${dim}
C++ Type
C++ Type
Mangling
C++ Type
Mangling
C++ Type
Mangling
C++ Type
itk::FixedArray<${ITKT ${type}}, ${dim} >
itk::Vector<${ITKT ${type}}, ${dim} >
itk::CovariantVector<${ITKT ${type}}, ${dim} >
itk::VariableLengthVector<${ITKT ${type}},
${dim} >
itk::SymmetricSecondRankTensor<${ITKT ${type}}, ${dim} >
Table 9.3: CMake wrapping mangling variables, their values, and the corresponding CMake C++ type variables
and their values for other ITK pixel types.
9.5. Wrapping
235
CMake Variable
Value
Mangling
C++ Type
ITKM O${dim}
ITKT O${dim}
O${dim}
itk::Offset<${dim} >
Mangling
ITKM CI${ITKM ${type}}${dim}
ITKT CI${ITKM ${type}}${dim}
CI${ITKM ${type}}${dim}
ITKM P${ITKM ${type}}${dim}
ITKT P${ITKM ${type}}${dim}
P${ITKM ${type}}${dim}
ITKM I${ITKM ${type}}${dim}
ITKT I${ITKM ${type}}${dim}
I${ITKM ${type}}${dim}
ITKM VI${ITKM ${type}}${dim}
ITKT VI${ITKM ${type}}${dim}
VI${ITKM ${type}}${dim}
Mangling
C++ Type
ITKM SO${dim}
ITKT SO${dim}
SO${dim}
itk::SpatialObject<${dim} >
Mangling
C++ Type
ITKM SE${dim}
ITKT SE${dim}
SE${dim}
itk::FlatStructuringElement<${dim} >
Mangling
C++ Type
ITKM H${ITKM ${type}}
ITKT H${ITKM ${type}}
H${ITKM ${type}}
itk::Statistics::Histogram<${ITKT${type}} >
Mangling
C++ Type
ITKM ST
ITKT ST
Depends on platform
itk::SizeValueType
Mangling
C++ Type
ITKM IT
ITKM IT
Depends on platform
itk::IdentifierType
Mangling
C++ Type
ITKM OT
ITKT OT
Depends on platform
itk::OffsetValueType
C++ Type
Mangling
C++ Type
Mangling
C++ Type
Mangling
C++ Type
itk::ContinuousIndex<${ITKT ${type}}, ${dim} >
itk::Point<${ITKT ${type}}, ${dim} >
itk::Image<${ITKT ${type}}, ${dim} >
itk::VectorImage<${ITKT ${type}}, ${dim} >
Table 9.4: CMake wrapping mangling variables, their values, and the corresponding CMake C++ type variables
and their values for basic ITK types.
236
Chapter 9. How To Create A Module
itk_wrap_template("${d}${ITKM_${t}}" "${d},${ITKT_${t}}")
endforeach()
endforeach()
itk_end_wrap_class()
In addition to itk wrap template, there are template wrapping macros specialized for wrapping
image filters. The highest level macro is itk wrap image filter, which is used for wrapping
image filters that need one or more image parameters of the same type. This macro has two required arguments. The first argument is a semicolon delimited CMake list of pixel types. The
second argument is the number of image template arguments for the filter. An optional third argument is a dimensionality condition to restrict the dimensions that the filter can be instantiated.
The dimensionality condition can be a number indicating the dimension allowed, a semicolon
delimited CMake list of dimensions, or a string of the form n+, where n is a number, to indicate that instantiations are allowed for dimension n and above. The wrapping specification for
itk::ThresholdMaximumConnectedComponentsImageFilter is
itk_wrap_class("itk::ThresholdMaximumConnectedComponentsImageFilter" POINTER)
itk_wrap_image_filter("${WRAP_ITK_INT}" 1 2+)
itk_end_wrap_class()
注意：指定参数
注意：通过定义的宏将ITK封装至其
他语言
If it is desirable or required to instantiate an image filter with different image types, the itk wrap image filter combinations macro is applicable. This macro takes a variable number of parameters, where each parameter is a list of the possible image pixel types for the corresponding filter
template parameters. A condition to restrict dimensionality may again be optionally passed as the
last argument. For example, wrapping for itk::VectorMagnitudeImageFilter is specified with
itk_wrap_class("itk::VectorMagnitudeImageFilter" POINTER_WITH_SUPERCLASS)
itk_wrap_image_filter_combinations("${WRAP_ITK_COV_VECTOR_REAL}" "${WRAP_ITK_SCALAR}")
itk_end_wrap_class()
The final template wrapping macro is itk wrap image filter types. This macro takes a variable
number of arguments that should correspond to the image pixel types in the filter’s template parameter list. Again, an optional dimensionality condition can be specified as the last argument. For
example, wrapping for itk::RGBToLuminanceImageFilter is specified with
itk_wrap_class("itk::RGBToLuminanceImageFilter" POINTER_WITH_SUPERCLASS)
if(ITK_WRAP_rgb_unsigned_char AND ITK_WRAP_unsigned_char)
itk_wrap_image_filter_types(RGBUC UC)
endif(ITK_WRAP_rgb_unsigned_char AND ITK_WRAP_unsigned_char)
if(ITK_WRAP_rgb_unsigned_short AND ITK_WRAP_unsigned_short)
itk_wrap_image_filter_types(RGBUS US)
endif(ITK_WRAP_rgb_unsigned_short AND ITK_WRAP_unsigned_short)
if(ITK_WRAP_rgba_unsigned_char AND ITK_WRAP_unsigned_char)
9.6. Third-Party Dependencies
237
itk_wrap_image_filter_types(RGBAUC UC)
endif(ITK_WRAP_rgba_unsigned_char AND ITK_WRAP_unsigned_char)
if(ITK_WRAP_rgba_unsigned_short AND ITK_WRAP_unsigned_short)
itk_wrap_image_filter_types(RGBAUS US)
endif(ITK_WRAP_rgba_unsigned_short AND ITK_WRAP_unsigned_short)
itk_end_wrap_class()
In some cases, it necessary to specify the headers required to build wrapping sources for a class. To
specify additional headers to included in the generated wrapping C++ source, use the itk wrap include macro. This macro takes the name of the header to include, and it can be called multiple
times.
By default, the class wrapping macros include a header whose filename corresponds to the name of
the class to be wrapped according to ITK naming conventions. To override the default behavior, set
the CMake variable WRAPPER AUTO INCLUDE HEADERS to OFF before calling itk wrap class.
For example,
set(WRAPPER_AUTO_INCLUDE_HEADERS OFF)
itk_wrap_include("itkTransformFileReader.h")
itk_wrap_class("itk::TransformFileReaderTemplate" POINTER)
foreach(t ${WRAP_ITK_REAL})
itk_wrap_template("${ITKM_${t}}" "${ITKT_${t}}")
endforeach()
itk_end_wrap_class()
There are a number of convenience CMake macros available to manipulate lists of template parameters. These macros take the variable name to populate with their output as the first argument followed
by input arguments. The itk wrap filter dims macro will process the dimensionality condition
previously described for the filter template wrapping macros. DECREMENT, INCREMENT are
macros that operate on dimensions. The INTERSECTION macro finds the intersection of two list
arguments. Finally, the UNIQUE macro removes duplicates from the given list. 注意：每个宏定义了不一样的转换功
能，利用宏
9.6 Third-Party Dependencies
When an ITK module depends on another ITK module, it simply lists its dependencies as described
in Section 9.1. A module can also depend on non-ITK third-party libraries. This third-party library
can be encapsulated in an ITK module – see examples in the ITK/Modules/ThirdParty directory.
Or, the dependency can be built or installed on the system and found with CMake. This section
describes how to add the CMake configuration to a module for it to find and use a third-party library
dependency.
238
Chapter 9. How To Create A Module
9.6.1 itk-module-init.cmake
The itk-module-init.cmake file, if present, is found in the top level directory of the module next
to the itk-module.cmake file. This file informs CMake of the build configuration and location of
the third-party dependency. To inform CMake about the OpenCV library, use the find package
command,
find_package(OpenCV REQUIRED)
9.6.2 CMakeList.txt
A few additions are required to the top level CMakeLists.txt of the module.
First, the itk-module-init.cmake file should be explicitly included when building the module
externally against an existing ITK build tree.
if(NOT ITK_SOURCE_DIR)
include(itk-module-init.cmake)
endif()
project(ITKVideoBridgeOpenCV)
Optionally, the dependency libraries are added to the LIBRARIES variable. Alternatively, if the module creates a library, publically link to the dependency libraries. Our ITKVideoBridgeOpenCV module example creates its own library, named ITKVideoBridgeOpenCV, and publically links to the OpenCV libraries.
CMakeLists.txt:
set(ITKVideoBridgeOpenCV_LIBRARIES ITKVideoBridgeOpenCV)
src/CMakeLists.txt:
target_link_libraries(ITKVideoBridgeOpenCV LINK_PUBLIC ${OpenCV_LIBS})
Next, CMake export code is created. This code is loaded by CMake when another project uses this
module. The export code stores where the dependency was located when the module was built, and
how CMake should find it. Two versions are required for the build tree and for the install tree.
# When this module is loaded by an app, load OpenCV too.
set(ITKVideoBridgeOpenCV_EXPORT_CODE_INSTALL "
set(OpenCV_DIR \"${OpenCV_DIR}\")
find_package(OpenCV REQUIRED)
")
9.7. Contributing with a Remote Module
239
set(ITKVideoBridgeOpenCV_EXPORT_CODE_BUILD "
if(NOT ITK_BINARY_DIR)
set(OpenCV_DIR \"${OpenCV_DIR}\")
find_package(OpenCV REQUIRED)
endif()
")
Finally, set the SYSTEM INCLUDE DIRS and SYSTEM LIBRARY DIRS, if required, to append compilation header directories and library linking directories
for this module.
set(ITKVideoBridgeOpenCV_SYSTEM_INCLUDE_DIRS ${OpenCV_INCLUDE_DIRS})
set(ITKVideoBridgeOpenCV_SYSTEM_LIBRARY_DIRS ${OpenCV_LIB_DIR})
9.7 Contributing with a Remote Module
For most ITK community members, the modularization of the toolkit is relatively transparent. The
default configuration includes all the (default) modules into the ITK library, which is used to build
their own ITK applications.
For ITK developers and code contributors, the modular structure imposes rules for organizing the
source code, building the library and contributing to the ITK source code repository.
A Module may be developed outside the main ITK repository, but it may be made available in the
ITK repository as a Remote Module. The Remote Module infrastructure enables fast dissemination
of research code through ITK without increasing the size of the main repository. The Insight Journal
(http://www.insight-journal.org/) adds support for ITK module submissions with automatic
dashboard testing (see Section 10.2 on page 244 for further details).
The source code of a Remote Module can be downloaded by CMake (with a CMake variable switch)
at ITK CMake configuration time, making it a convenient way to distribute modular source code.
9.7.1 Policy for Adding and Removing Remote Modules
A module can be added to the list of remotes if it satisfies the following criteria:
• There is a peer-reviewed article in an online, open access journal (such as the Insight Journal)
describing the theory behind and usage of the module.
• There is a nightly build against ITK master on the CDash dashboard that builds and passes
tests successfully.
• A name and contact email exists for the dashboard build. The maintainer of the dashboard
build does not necessarily need to be the original author of the Insight Journal article.
240
Chapter 9. How To Create A Module
• The license should be compatible with the rest of the toolkit. That is it should be an Open
Source Initiative-approved license1 without copyleft or non-commercial restrictions. Ideally,
it should be an Apache 2.0 license assigned to the Insight Software Consortium as found in the
rest of the toolkit. Note that the module should contain neither patented code, nor algorithms,
nor methods.
At the beginning of the release candidate phase of a release, maintainers of failing module dashboard builds will be contacted. If a module’s dashboard submission is still failing at the last release
candidate tagging, it will be removed before the final release.
Module names must be unique.
At no time in the future should a module in the main repository depend on a Remote Module.
9.7.2 Procedure for Adding a Remote Module
The repository
https://github.com/InsightSoftwareConsortium/ITKModuleTemplate
provides a useful template to be used as a starting point for a new ITK module.
The procedure to publish a new module in ITK is summarized as follows:
1. Publish an open access article describing the module in an online, open access journal like
The Insight Journal.
2. Push a topic to the ITK GitHub repository (see Section 10.1 on page 243 that adds a file named
Modules/Remote/.remote.cmake. This file must have the following:
(a) Dashboard maintainer name and email in the comments.
(b) A call to the itk fetch module CMake function
CMake/ITKModuleRemote.cmake) whose arguments are:
(documented
in
i. The name of the remote module.
Note that in each .remote.cmake, the first argument of the function itk fetch module()
is the name of the remote module, and it has to be consistent with the module name
defined in the corresponding .remote.cmake. To better
distinguish the remote modules from the internal ITK modules, the names of the
remote modules should NOT contain the “ITK” string prefix.
注意：
ii. A short description of the module with the handle to the open access article.
iii. URLs describing the location and version of the code to download. The version
should be a specific hash.
1 https://opensource.org/licenses
9.7. Contributing with a Remote Module
241
After the Remote Module has experienced sufficient testing, and community members express broad
interest in the contribution, the submitter can then move the contribution into the ITK repository via
GitHub code review.
It is possible but not recommended to directly push a module to GitHub for review without submitting to Insight Journal first.
CHAPTER
TEN
SOFTWARE PROCESS
An outstanding feature of ITK is the software process used to develop, maintain and test the toolkit.
The Insight Toolkit software continues to evolve rapidly due to the efforts of developers and users
located around the world, so the software process is essential to maintaining its quality. If you are
planning to contribute to ITK, or use the Git source code repository, you need to know something
about this process (see 2.1 on page 10 to learn more about obtaining ITK using Git). This information will help you know when and how to update and work with the software as it changes. The
following sections describe key elements of the process.
10.1 Git Source Code Repository
Git1 https://git-scm.com/ is a tool for version control. It is a valuable resource for software projects
involving multiple developers. The primary purpose of Git is to keep track of changes to software.
Git date and version stamps every addition to files in the repository. Additionally, a user may set a tag
to mark a particular of the whole software. Thus, it is possible to return to a particular state or point
of time whenever desired. The differences between any two points is represented by a “diff” file, that
is a compact, incremental representation of change. Git supports concurrent development so that two
developers can edit the same file at the same time, that are then (usually) merged together without
incident (and marked if there is a conflict). In addition, branches off of the main development trunk
provide parallel development of software.
Developers and users can check out the software from the Git repository. When developers introduce
changes in the system, Git facilitates to update the local copies of other developers and users by
downloading only the differences between their local copy and the version on the repository. This is
an important advantage for those who are interested in keeping up to date with the leading edge of
the toolkit. Bug fixes can be obtained in this way as soon as they have been checked into the system.
ITK source code, data, and examples are maintained in a Git repository. The principal advantage of
a system like Git is that it frees developers to try new ideas and introduce changes without fear of
1 \unskip\penalty\@M\vrulewidth\z@height\z@depth\dpff
244
Chapter 10. Software Process
losing a previous working version of the software. It also provides a simple way to incrementally
update code as new features are added to the repository.
The ITK community use Git, and the social coding web platform, GitHub
(https://github.com/InsightSoftwareConsortium), to facilitate a structured, orderly
method for developers to contribute new code and bug fixes to ITK. The GitHub review process allows anyone to submit a proposed change to ITK, after which it will be reviewed by other developers
before being approved and merged into ITK. For more information on how to contribute, please visit
https://github.com/InsightSoftwareConsortium/ITK/blob/master/CONTRIBUTING.md.
For information about the Git-based development workflow adopted by ITK, see the Appendix B
on page 277.
10.2 CDash Regression Testing System
One of the unique features of the ITK software process is its use of the CDash regression testing
system (http://www.cdash.org). In a nutshell, what CDash does is to provide quantifiable feedback to developers as they check in new code and make changes. The feedback consists of the
results of a variety of tests, and the results are posted on a publicly-accessible Web page (to which
we refer as a dashboard) as shown in Figure 10.1. The most recent dashboard is accessible from
https://www.itk.org/ITK/resources/testing.html). Since all users and developers of ITK
can view the Web page, the CDash dashboard serves as a vehicle for developer communication, especially when new additions to the software is found to be faulty. The dashboard should be consulted
before considering updating software via Git.
Note that CDash is independent of ITK and can be used to manage quality control for any software
project. It is itself an open-source package and can be obtained from
http://www.cdash.org
CDash supports a variety of test types. These include the following.
Compilation. All source and test code is compiled and linked. Any resulting errors and warnings
are reported on the dashboard.
Regression. Some ITK tests produce images as output. Testing requires comparing each test’s output against a valid baseline image. If the images match then the test passes. The comparison
must be performed carefully since many 3D graphics systems (e.g., OpenGL) produce slightly
different results on different platforms.
Memory. Problems relating to memory such as leaks, uninitialized memory reads, and reads/ writes
beyond allocated space can cause unexpected results and program crashes. ITK checks runtime memory access and management using Purify, a commercial package produced by Rational. (Other memory checking programs will be added in the future.)
10.2. CDash Regression Testing System
245
Figure 10.1: On-line presentation of the quality dashboard generated by CDash.
PrintSelf. All classes in ITK are expected to print out all their instance (i.e., those with associated
Set and Get methods) and their internal variables correctly. This test checks to make sure that
this is the case.
Unit. Each class in ITK should have a corresponding unit test where the class functionalities are
exercised and quantitatively compared against expected results. These tests are typically written by the class developer and should endeavor to cover all lines of code including Set/Get
methods and error handling.
Coverage. There is a saying among ITK developers: If it isn’t covered, then it’s broke. What this
means is that code that is not executed during testing is likely to be wrong. The coverage tests
identify lines that are not executed in the Insight Toolkit test suite, reporting a total percentage
covered at the end of the test. While it is nearly impossible to bring the coverage to 100%
because of error handling code and similar constructs that are rarely encountered in practice,
the coverage numbers should be 75% or higher. Code that is not covered well enough requires
additional tests.
Figure 10.1 shows the top-level dashboard web page. Each row in the dashboard corresponds to a
particular platform (hardware + operating system + compiler). The data on the row indicates the
number of compile errors and warnings as well as the results of running hundreds of small test
programs. In this way the toolkit is tested both at compile time and run time.
注意：
When a user or developer decides to update ITK source code from Git it is important to first verify
that the current dashboard is in good shape. This can be rapidly judged by the general coloration of
the dashboard. A green state means that the software is building correctly and it is a good day to
246
Chapter 10. Software Process
start with ITK or to get an upgrade. A red state, on the other hand, is an indication of instability on
the system and hence users should refrain from checking out or upgrading the source code.
Another nice feature of CDash is that it maintains a history of changes to the source code (by
coordinating with Git) and summarizes the changes as part of the dashboard. This is useful for
tracking problems and keeping up to date with new additions to ITK.
10.2.1 Developing tests
As highlighted, testing is an essential part of ITK. Regression testing on a regular basis allows ITK
to meet high code quality standards, and to enable reproducible research. Code coverage reported
daily in CDash allows us to systematically measure the degree to which the ITK source code is
reliable. Therefore, writing tests, and improving current tests and the testing infrastructure is crucial
to ITK.
There are a number of scenarios when writing tests:
• Modifying existing classes. When modifying an existing class (either due to a bug or a performance improvement), it must be checked that existing tests exercise the new code. Otherwise,
either the existing tests should be modified to include the appropriate cases for the new code
to be exercised (without harm to existing cases), or a new test file may be required.
• Contributing new classes. When contributing a new class, a unit test or a few unit tests
should be provided to check that the class is working as expected. The unit tests are expected
to exercise all of the members of the new class.
In either case, the tips and tools described in Section 9.4 were developed to improve and facilitate
the process.
10.3 Working The Process
The ITK software process functions across three cycles—the continuous cycle, the daily cycle, and
the release cycle.
The continuous cycle revolves around the actions of developers as they check code into Git. When
changed or new code is checked into Git, the CDash continuous testing process kicks in. A small
number of tests are performed (including compilation), and if something breaks, email is sent to all
developers who checked code in during the continuous cycle. Developers are expected to fix the
problem immediately.
The daily cycle occurs over a 24-hour period. Changes to the source base made during the day are
extensively tested by the nightly CDash regression testing sequence. These tests occur on different
combinations of computers and operating systems located around the world, and the results are
10.4. The Effectiveness of the Process
247
posted every day to the CDash dashboard. Developers who checked in code are expected to visit
the dashboard and ensure their changes are acceptable—that is, they do not introduce compilation
errors or warnings, or break any other tests including regression, memory, PrintSelf, and Set/Get.
Again, developers are expected to fix problems immediately.
The release cycle occurs a small number of times a year. This requires tagging and branching the
Git repository, updating documentation, and producing new release packages. Although additional
testing is performed to insure the consistency of the package, keeping the daily Git build error free
minimizes the work required to cut a release.
ITK users typically work with releases, since they are the most stable. Developers work with the Git
repository, or sometimes with periodic release snapshots, in order to take advantage of newly-added
features. It is extremely important that developers watch the dashboard carefully, and update their
software only when the dashboard is in good condition (i.e., is “green”). Failure to do so can cause
significant disruption if a particular day’s software release is unstable. 注意：1、提交代码--立即编译测
试--立即修复
2、每天进行软件测试--贴出
bug
3、每年几次发布版本
10.4 The Effectiveness of the Process
The effectiveness of this process is profound. By providing immediate feedback to developers
through email and Web pages (e.g., the dashboard), the quality of ITK is exceptionally high, especially considering the complexity of the algorithms and system. Errors, when accidentally introduced, are caught quickly, as compared to catching them at the point of release. To wait to the point
of release is to wait too long, since the causal relationship between a code change or addition and a
bug is lost. The process is so powerful that it routinely catches errors in vendor’s graphics drivers
(e.g., OpenGL drivers) or changes to external subsystems such as the VXL/VNL numerics library.
All of these tools that make up the process (CMake, Git, and CDash) are open-source. Many large
and small systems such as VTK (The Visualization Toolkit http://www.vtk.org) use the same
process with similar results. We encourage the adoption of the process in your environment.
Appendices
APPENDIX
ONE
LICENSES
A.1 Insight Toolkit License
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
252
Appendix A. Licenses
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
A.1. Insight Toolkit License
253
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form
that You distribute, all copyright,
attribution notices from the Source
excluding those notices that do not
the Derivative Works; and
of any Derivative Works
patent, trademark, and
form of the Work,
pertain to any part of
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
254
Appendix A. Licenses
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
A.1. Insight Toolkit License
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
255
256
Appendix A. Licenses
See the License for the specific language governing permissions and
limitations under the License.
A.2 Third Party Licenses
注意：绑定了许多第三方库
The Insight Toolkit bundles a number of third party libraries that are used internally. The licenses of
these libraries are as follows.
A.2.1 DICOM Parser
/*=========================================================================
Program:
Module:
Language:
Date:
Version:
DICOMParser
Copyright.txt
C++
$Date$
$Revision$
Copyright (c) 2003 Matt Turek
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright notice,
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
* The name of Matt Turek nor the names of any contributors may be used to
endorse or promote products derived from this software without specific
prior written permission.
* Modified source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS IS''
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
A.2. Third Party Licenses
257
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
=========================================================================*/
A.2.2 Double Conversion
Copyright 2006-2011, the V8 project authors. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above
copyright notice, this list of conditions and the following
disclaimer in the documentation and/or other materials provided
with the distribution.
* Neither the name of Google Inc. nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
258
Appendix A. Licenses
A.2.3 Expat
Copyright (c) 1998, 1999, 2000 Thai Open Source Software Center Ltd
and Clark Cooper
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
A.2.4 GDCM
/*=========================================================================
Program: GDCM (Grassroots DICOM). A DICOM library
Copyright (c) 2006-2016 Mathieu Malaterre
Copyright (c) 1993-2005 CREATIS
(CREATIS = Centre de Recherche et d'Applications en Traitement de l'Image)
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright notice,
A.2. Third Party Licenses
259
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
* Neither name of Mathieu Malaterre, or CREATIS, nor the names of any
contributors (CNRS, INSERM, UCB, Universite Lyon I), may be used to
endorse or promote products derived from this software without specific
prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS IS''
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
=========================================================================*/
A.2.5 GIFTI
The gifticlib code is released into the public domain. Developers are
encouraged to incorporate the library into their application, and to
contribute changes or enhancements to gifticlib.
Author: Richard Reynolds, SSCC, DIRP, NIMH, National Institutes of Health
May 13, 2008 (release version 1.0.0)
http://www.nitrc.org/projects/gifti
A.2.6 HDF5
Copyright Notice and License Terms for
HDF5 (Hierarchical Data Format 5) Software Library and Utilities
----------------------------------------------------------------------------HDF5 (Hierarchical Data Format 5) Software Library and Utilities
Copyright (c) 2006-2018, The HDF Group.
NCSA HDF5 (Hierarchical Data Format 5) Software Library and Utilities
260
Appendix A. Licenses
Copyright (c) 1998-2006, The Board of Trustees of the University of Illinois.
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted for any purpose (including commercial purposes)
provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice,
this list of conditions, and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
this list of conditions, and the following disclaimer in the documentation
and/or materials provided with the distribution.
3. Neither the name of The HDF Group, the name of the University, nor the
name of any Contributor may be used to endorse or promote products derived
from this software without specific prior written permission from
The HDF Group, the University, or the Contributor, respectively.
DISCLAIMER:
THIS SOFTWARE IS PROVIDED BY THE HDF GROUP AND THE CONTRIBUTORS
"AS IS" WITH NO WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED. IN NO EVENT SHALL THE HDF
You are under no obligation whatsoever to provide any bug fixes, patches, or upgrades to the
--------------------------------------------------------------------------------------------------------------------------------------------------------Limited portions of HDF5 were developed by Lawrence Berkeley National
Laboratory (LBNL). LBNL's Copyright Notice and Licensing Terms can be
found here: COPYING_LBNL_HDF5 file in this directory or at
http://support.hdfgroup.org/ftp/HDF5/releases/COPYING_LBNL_HDF5.
---------------------------------------------------------------------------------------------------------------------------------------------------------
Contributors:
National Center for Supercomputing Applications (NCSA) at
the University of Illinois, Fortner Software, Unidata Program Center (netCDF), The Independen
----------------------------------------------------------------------------Portions of HDF5 were developed with support from the Lawrence Berkeley
A.2. Third Party Licenses
261
National Laboratory (LBNL) and the United States Department of Energy
under Prime Contract No. DE-AC02-05CH11231.
----------------------------------------------------------------------------Portions of HDF5 were developed with support from the University of
California, Lawrence Livermore National Laboratory (UC LLNL).
The following statement applies to those portions of the product and must
be retained in any redistribution of source code, binaries, documentation,
and/or accompanying materials:
This work was partially produced at the University of California,
Lawrence Livermore National Laboratory (UC LLNL) under contract
no. W-7405-ENG-48 (Contract 48) between the U.S. Department of Energy
(DOE) and The Regents of the University of California (University)
for the operation of UC LLNL.
DISCLAIMER:
This work was prepared as an account of work sponsored by an agency of
the United States Government. Neither the United States Government nor
the University of California nor any of their employees, makes any
warranty, express or implied, or assumes any liability or responsibility
for the accuracy, completeness, or usefulness of any information,
apparatus, product, or process disclosed, or represents that its use
would not infringe privately- owned rights. Reference herein to any
specific commercial products, process, or service by trade name,
trademark, manufacturer, or otherwise, does not necessarily constitute
or imply its endorsement, recommendation, or favoring by the United
States Government or the University of California. The views and
opinions of authors expressed herein do not necessarily state or reflect
those of the United States Government or the University of California,
and shall not be used for advertising or product endorsement purposes.
----------------------------------------------------------------------------HDF5 is available with the SZIP compression library but SZIP is not part
of HDF5 and has separate copyright and license terms. See SZIP Compression
in HDF Products (www.hdfgroup.org/doc_resource/SZIP/) for further details.
-----------------------------------------------------------------------------
262
Appendix A. Licenses
A.2.7 JPEG
The authors make NO WARRANTY or representation, either express or implied,
with respect to this software, its quality, accuracy, merchantability, or
fitness for a particular purpose. This software is provided "AS IS", and you,
its user, assume the entire risk as to its quality and accuracy.
This software is copyright (C) 1991-2010, Thomas G. Lane, Guido Vollbeding.
All Rights Reserved except as specified below.
Permission is hereby granted to use, copy, modify, and distribute this
software (or portions thereof) for any purpose, without fee, subject to these
conditions:
(1) If any part of the source code for this software is distributed, then this
README file must be included, with this copyright and no-warranty notice
unaltered; and any additions, deletions, or changes to the original files
must be clearly indicated in accompanying documentation.
(2) If only executable code is distributed, then the accompanying
documentation must state that "this software is based in part on the work of
the Independent JPEG Group".
(3) Permission for use of this software is granted only if the user accepts
full responsibility for any undesirable consequences; the authors accept
NO LIABILITY for damages of any kind.
These conditions apply to any software derived from or based on the IJG code,
not just to the unmodified library. If you use our work, you ought to
acknowledge us.
Permission is NOT granted for the use of any IJG author's name or company name
in advertising or publicity relating to this software or products derived from
it. This software may be referred to only as "the Independent JPEG Group's
software".
We specifically permit and encourage the use of this software as the basis of
commercial products, provided that all warranty or liability claims are
assumed by the product vendor.
ansi2knr.c is included in this distribution by permission of L. Peter Deutsch,
sole proprietor of its copyright holder, Aladdin Enterprises of Menlo Park, CA.
ansi2knr.c is NOT covered by the above copyright and conditions, but instead
by the usual distribution terms of the Free Software Foundation; principally,
that you must include source code if you redistribute it. (See the file
A.2. Third Party Licenses
263
ansi2knr.c for full details.) However, since ansi2knr.c is not needed as part
of any program generated from the IJG code, this does not limit you more than
the foregoing paragraphs do.
The Unix configuration script "configure" was produced with GNU Autoconf.
It is copyright by the Free Software Foundation but is freely distributable.
The same holds for its supporting scripts (config.guess, config.sub,
ltmain.sh). Another support script, install-sh, is copyright by X Consortium
but is also freely distributable.
The IJG distribution formerly included code to read and write GIF files.
To avoid entanglement with the Unisys LZW patent, GIF reading support has
been removed altogether, and the GIF writer has been simplified to produce
"uncompressed GIFs". This technique does not use the LZW algorithm; the
resulting GIF files are larger than usual, but are readable by all standard
GIF decoders.
We are required to state that
"The Graphics Interchange Format(c) is the Copyright property of
CompuServe Incorporated. GIF(sm) is a Service Mark property of
CompuServe Incorporated."
A.2.8 KWSys
KWSys - Kitware System Library
Copyright 2000-2016 Kitware, Inc. and Contributors
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Kitware, Inc. nor the names of Contributors
may be used to endorse or promote products derived from this
software without specific prior written permission.
264
Appendix A. Licenses
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-----------------------------------------------------------------------------The following individuals and institutions are among the Contributors:
* Insight Software Consortium
See version control history for details of individual contributions.
A.2.9 MetaIO
MetaIO - Medical Image I/O
The following license applies to all code, without exception,
in the MetaIO library.
/*=========================================================================
Copyright 2000-2014 Insight Software Consortium
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of the Insight Software Consortium nor the names of
A.2. Third Party Licenses
265
its contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
=========================================================================*/
/*=========================================================================
Copyright (c) 1999-2007 Insight Software Consortium
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright notice,
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
* The name of the Insight Software Consortium, nor the names of any
consortium members, nor of any contributors, may be used to endorse or
promote products derived from this software without specific prior written
permission.
* Modified source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS ``AS IS''
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
266
Appendix A. Licenses
ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
=========================================================================*/
A.2.10 Netlib’s SLATEC
This code is in the public domain. From http://www.netlib.org/slatec/guide:
SECTION 4.
OBTAINING THE LIBRARY
The Library is in the public domain and distributed by the Energy Science
and Technology Software Center.
Energy Science and Technology Software Center
P.O. Box 1020
Oak Ridge, TN 37831
Telephone 615-576-2606
E-mail estsc%a1.adonis.mrouter@zeus.osti.gov
A.2.11 NIFTI
Niftilib has been developed by members of the NIFTI DFWG and volunteers in the
neuroimaging community and serves as a reference implementation of the nifti-1
file format.
http://nifti.nimh.nih.gov/
Nifticlib code is released into the public domain, developers are encouraged to
incorporate niftilib code into their applications, and, to contribute changes
and enhancements to niftilib.
A.2.12 NrrdIO
---------------------------------------------------------------------------
A.2. Third Party Licenses
267
License --------------------------------------------------------------------------------------------------------------------------------------------NrrdIO: stand-alone
Copyright (C) 2013,
Copyright (C) 2008,
Copyright (C) 2004,
code for basic nrrd functionality
2012, 2011, 2010, 2009 University of Chicago
2007, 2006, 2005 Gordon Kindlmann
2003, 2002, 2001, 2000, 1999, 1998 University of Utah
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any
damages arising from the use of this software.
Permission is granted to anyone to use this software for any
purpose, including commercial applications, and to alter it and
redistribute it freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must
not claim that you wrote the original software. If you use this
software in a product, an acknowledgment in the product
documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must
not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
--------------------------------------------------------------------------General information --------------------------------------------------------------------------------------------------------------------------------**
**
**
**
**
**
NOTE: These source files have been copied and/or modified from Teem,
. Teem is licensed under a weakened GNU Lesser Public
License (the weakening is to remove burdens on those releasing binaries
that statically link against Teem) . The non-reciprocal licensing defined
above applies to only the source files in the NrrdIO distribution, and not
to Teem.
NrrdIO is a modified and highly abbreviated version of the Teem. NrrdIO
contains only the source files (or portions thereof) required for
creating and destroying nrrds, and for getting them into and out of
files. The NrrdIO sources are created from the Teem sources by using
GNU Make (pre-GNUmakefile in the NrrdIO distribution).
268
Appendix A. Licenses
NrrdIO makes it very easy to add support for the NRRD file format to your
program, which is a good thing considering and design and flexibility of the
NRRD file format, and the existence of the "unu" command-line tool for
operating on nrrds. Using NrrdIO requires exactly one header file,
"NrrdIO.h", and exactly one library, libNrrdIO.
Currently, the API presented by NrrdIO is a strict subset of the Teem API.
There is no additional encapsulation or abstraction. This could be annoying
in the sense that you still have to deal with the biff (for error messages)
and the air (for utilities) library function calls. Or it could be good and
sane in the sense that code which uses NrrdIO can be painlessly "upgraded" to
use more of Teem. Also, the API documentation for the same functionality in
Teem will apply directly to NrrdIO.
NrrdIO was originally created with the help of Josh Cates in order to add
support for the NRRD file format to the Insight Toolkit (ITK).
--------------------------------------------------------------------------NrrdIO API crash course ----------------------------------------------------------------------------------------------------------------------------Please read . The functions that
are explained in detail are all present in NrrdIO. Be aware, however, that
NrrdIO currently supports ONLY the NRRD file format, and not: PNG, PNM, VTK,
or EPS.
The functionality in Teem's nrrd library which is NOT in NrrdIO is basically
all those non-trivial manipulations of the values in the nrrd, or their
ordering in memory. Still, NrrdIO can do a fair amount, namely all the
functions listed in these sections of the "Overview of rest of API" in the
above web page:
-
Basic "methods"
Manipulation of per-axis meta-information
Utility functions
Comments in nrrd
Key/value pairs
Endianness (byte ordering)
Getting/Setting values (crude!)
Input from, Output to files
--------------------------------------------------------------------------Files comprising NrrdIO ---------------------------------------------------
A.2. Third Party Licenses
269
--------------------------------------------------------------------------NrrdIO.h: The single header file that declares all the functions and variables
that NrrdIO provides.
sampleIO.c: Tiny little command-line program demonstrating the basic NrrdIO
API. Read this for examples of how NrrdIO is used to read and write NRRD
files.
CMakeLists.txt: to build NrrdIO with CMake
pre-GNUmakefile: how NrrdIO sources are created from the Teem
sources. Requires that TEEM_SRC_ROOT be set, and uses the following two files.
tail.pl, unteem.pl: used to make small modifications to the source files to
convert them from Teem to NrrdIO sources
mangle.pl: used to generate a #include file for name-mangling the external
symbols in the NrrdIO library, to avoid possible problems with programs
that link with both NrrdIO and the rest of Teem.
preamble.c: the preamble describing the non-copyleft licensing of NrrdIO.
qnanhibit.c: discover a variable which, like endianness, is architecture
dependent and which is required for building NrrdIO (as well as Teem), but
unlike endianness, is completely obscure and unheard of.
encodingBzip2.c, formatEPS.c, formatPNG.c, formatPNM.c, formatText.c,
formatVTK.c: These files create stubs for functionality which is fully present
in Teem, but which has been removed from NrrdIO in the interest of simplicity.
The filenames are in fact unfortunately misleading, but they should be
understood as listing the functionality that is MISSING in NrrdIO.
All other files: copied/modified from the air, biff, and nrrd libraries of
Teem.
A.2.13 OpenJPEG
/*
*
*
*
*
Copyright (c) 2002-2012,
Universite catholique de
Copyright (c) 2002-2012,
Copyright (c) 2003-2012,
Communications and Remote Sensing Laboratory,
Louvain (UCL), Belgium
Professor Benoit Macq
Antonin Descampe
270
Appendix A. Licenses
* Copyright (c) 2003-2009, Francois-Olivier Devaux
* Copyright (c) 2005, Herve Drolon, FreeImage Team
* Copyright (c) 2002-2003, Yannick Verschueren
* Copyright (c) 2001-2003, David Janssens
* Copyright (c) 2011-2012, Centre National d'Etudes Spatiales (CNES), France
* Copyright (c) 2012, CS Systemes d'Information, France
*
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
*
notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
*
notice, this list of conditions and the following disclaimer in the
*
documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
A.2.14 PNG
This copy of the libpng notices is provided for your convenience. In case of
any discrepancy between this copy and the notices in the file png.h that is
included in the libpng distribution, the latter shall prevail.
COPYRIGHT NOTICE, DISCLAIMER, and LICENSE:
If you modify libpng you may insert additional notices immediately following
this sentence.
This code is released under the libpng license.
A.2. Third Party Licenses
271
libpng versions 1.2.6, August 15, 2004, through 1.6.9, February 6, 2014, are
Copyright (c) 2004, 2006-2014 Glenn Randers-Pehrson, and are
distributed according to the same disclaimer and license as libpng-1.2.5
with the following individual added to the list of Contributing Authors
Cosmin Truta
libpng versions 1.0.7, July 1, 2000, through 1.2.5 - October 3, 2002, are
Copyright (c) 2000-2002 Glenn Randers-Pehrson, and are
distributed according to the same disclaimer and license as libpng-1.0.6
with the following individuals added to the list of Contributing Authors
Simon-Pierre Cadieux
Eric S. Raymond
Gilles Vollant
and with the following additions to the disclaimer:
There is no warranty against interference with your enjoyment of the
library or against infringement. There is no warranty that our
efforts or the library will fulfill any of your particular purposes
or needs. This library is provided with all faults, and the entire
risk of satisfactory quality, performance, accuracy, and effort is with
the user.
libpng versions 0.97, January 1998, through 1.0.6, March 20, 2000, are
Copyright (c) 1998, 1999 Glenn Randers-Pehrson, and are
distributed according to the same disclaimer and license as libpng-0.96,
with the following individuals added to the list of Contributing Authors:
Tom Lane
Glenn Randers-Pehrson
Willem van Schaik
libpng versions 0.89, June 1996, through 0.96, May 1997, are
Copyright (c) 1996, 1997 Andreas Dilger
Distributed according to the same disclaimer and license as libpng-0.88,
with the following individuals added to the list of Contributing Authors:
John Bowler
Kevin Bracey
Sam Bushell
272
Appendix A. Licenses
Magnus Holmgren
Greg Roelofs
Tom Tanner
libpng versions 0.5, May 1995, through 0.88, January 1996, are
Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.
For the purposes of this copyright and license, "Contributing Authors"
is defined as the following set of individuals:
Andreas Dilger
Dave Martindale
Guy Eric Schalnat
Paul Schmidt
Tim Wegner
The PNG Reference Library is supplied "AS IS". The Contributing Authors
and Group 42, Inc. disclaim all warranties, expressed or implied,
including, without limitation, the warranties of merchantability and of
fitness for any purpose. The Contributing Authors and Group 42, Inc.
assume no liability for direct, indirect, incidental, special, exemplary,
or consequential damages, which may result from the use of the PNG
Reference Library, even if advised of the possibility of such damage.
Permission is hereby granted to use, copy, modify, and distribute this
source code, or portions hereof, for any purpose, without fee, subject
to the following restrictions:
1. The origin of this source code must not be misrepresented.
2. Altered versions must be plainly marked as such and must not
be misrepresented as being the original source.
3. This Copyright notice may not be removed or altered from any
source or altered source distribution.
The Contributing Authors and Group 42, Inc. specifically permit, without
fee, and encourage the use of this source code as a component to
supporting the PNG file format in commercial products. If you use this
source code in a product, acknowledgment is not required but would be
appreciated.
A.2. Third Party Licenses
273
A "png_get_copyright" function is available, for convenient use in "about"
boxes and the like:
printf("%s",png_get_copyright(NULL));
Also, the PNG logo (in PNG format, of course) is supplied in the
files "pngbar.png" and "pngbar.jpg (88x31) and "pngnow.png" (98x31).
Libpng is OSI Certified Open Source Software. OSI Certified Open Source is a
certification mark of the Open Source Initiative.
Glenn Randers-Pehrson
glennrp at users.sourceforge.net
February 6, 2014
A.2.15 TIFF
Copyright (c) 1988-1997 Sam Leffler
Copyright (c) 1991-1997 Silicon Graphics, Inc.
Permission to use, copy, modify, distribute, and sell this software and
its documentation for any purpose is hereby granted without fee, provided
that (i) the above copyright notices and this permission notice appear in
all copies of the software and related documentation, and (ii) the names of
Sam Leffler and Silicon Graphics may not be used in any advertising or
publicity relating to the software without the specific, prior written
permission of Sam Leffler and Silicon Graphics.
THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND,
EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
IN NO EVENT SHALL SAM LEFFLER OR SILICON GRAPHICS BE LIABLE FOR
ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
OF THIS SOFTWARE.
A.2.16 VNL
#ifndef vxl_copyright_h_
274
Appendix A. Licenses
#define vxl_copyright_h_
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
Copyright 2000-2013 VXL Contributors
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the names of the copyright holders nor the names of their
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
OF THE POSSIBILITY OF SUCH DAMAGE.
#endif // vxl_copyright_h_
A.2.17 ZLIB
Acknowledgments:
The deflate format used by zlib was defined by Phil Katz. The deflate
and zlib specifications were written by L. Peter Deutsch. Thanks to all the
people who reported problems and suggested various improvements in zlib;
they are too numerous to cite here.
A.2. Third Party Licenses
275
Copyright notice:
(C) 1995-2004 Jean-loup Gailly and Mark Adler
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any damages
arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it
freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not
claim that you wrote the original software. If you use this software
in a product, an acknowledgment in the product documentation would be
appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
Jean-loup Gailly
jloup@gzip.org
Mark Adler
madler@alumni.caltech.edu
If you use the zlib library in a product, we would appreciate *not*
receiving lengthy legal documents to sign. The sources are provided
for free but without warranty of any kind. The library has been
entirely written by Jean-loup Gailly and Mark Adler; it does not
include third-party code.
If you redistribute modified sources, we would appreciate that you include
in the file ChangeLog history information documenting your changes. Please
read the FAQ for more information on the distribution of modified source
versions.
APPENDIX
TWO
ITK GIT WORKFLOW
This chapter describes the workflow adopted by the ITK community to develop the software. The
adopted Git-based branchy workflow is an efficient and flexible model to develop modern software.
B.1 Git Setup
Visit the main Git download site, and depending on your operating system, follow the guidelines.
B.1.1 Windows
Git comes in two flavors on Windows:
• A Windows native application installer
• A Cygwin package
Choose one and stick with it. They do not get along well in a given work tree on disk (the repository
formats are compatible but the “stat cache” of the work tree is not unless core.filemode is false).
Git for Windows
Download the “git for windows” executable from the git for windows site. You want to download
the file that is named something like
Git-2.14.2.2-64-bit.exe
Note that the filename changes as new versions are released.
278
Appendix B. ITK Git Workflow
Run the installer.
When prompted, choose to not modify the PATH and choose the
core.autocrlf=true option. Launch the Git Bash tool to get a command line shell with Git.
Cygwin
Install the following packages:
• git: Git command-line tool
• gitk: Graphical history browser
• git-completion: Bash shell completion rules
Launch a Cygwin command prompt to get a command line shell with Git.
B.1.2 macOS
Xcode 4
If you have Xcode 4 installed, you already have git installed.
Verify with:
which git
/usr/bin/git
git --version
git version 1.7.4.4
OS X Installer
Download an installer from code.google.com.
MacPorts
Enter these commands:
sudo port selfupdate
sudo port install git-core +doc
B.2. Workflow
279
B.1.3 Linux
Popular Linux distributions already come with packages for Git. Typically the packages are called:
• git-core: Git command-line tool
• git-doc: Git documentation
• gitk: Graphical history browser
B.2 Workflow
B.2.1 A Primer
This primer details a possible workflow for using Git and ITK. There are many ways to use Git and
it is a very flexible tool. This page details my particular way of working with Git, and is certainly
not the last word. It is also rambling collection of tips and experiences that I’ve picked up in the last
few years of using Git.
It is worth trying to explain some high-level Git concepts. A feature (or a bug) that sets Git apart from
Subversion is its distributed nature. In practice, that means that ITK needs to “bless” a repository
for it to be the “official” source of ITK. This has already been done at the ITK GitHub repository.
Another Git concept is that of a commit. Git uses a SHA1 hash to uniquely identify a change
set. The hash is (almost) guaranteed to be unique across your project, and even across all projects
everywhere and for all time.
Quoting from the excellent Pro Git book1:
A lot of people become concerned at some point that they will, by random happenstance, have two objects in their repository that hash to the same SHA-1 value. What
then?
If you do happen to commit an object that hashes to the same SHA-1 value as a previous
object in your repository, Git will see the previous object already in your Git database
and assume it was already written. If you try to check out that object again at some
point, you’ll always get the data of the first object.
However, you should be aware of how ridiculously unlikely this scenario is. The SHA1 digest is 20 bytes or 160 bits. The number of randomly hashed objects needed to
ensure a 50% probability of a single collision is about 280 (the formula for determining
collision probability is p = (n(n − 1)/2) ∗ (1/2160)). 280 is 1.2 × 1024 or 1 million
billion billion. That’s 1,200 times the number of grains of sand on the earth.
1 https://git-scm.com/book/en/v2
280
Appendix B. ITK Git Workflow
Here’s an example to give you an idea of what it would take to get a SHA-1 collision.
If all 6.5 billion humans on Earth were programming, and every second, each one was
producing code that was the equivalent of the entire Linux kernel history (1 million Git
objects) and pushing it into one enormous Git repository, it would take 5 years until
that repository contained enough objects to have a 50% probability of a single SHA-1
object collision. A higher probability exists that every member of your programming
team will be attacked and killed by wolves in unrelated incidents on the same night.
B.2.2 A Topic
This workflow is based on the branchy development workflow documented by Git help workflows.
Motivation
The primary goal of this workflow is to make release preparation and maintenance easier. We set
the following requirements, of which some are themselves worthwhile goals:
• Amortize the effort of release preparation throughout development.
• Support granular selection of features for release.
• Allow immature features to be published without delaying release.
• Keep unrelated development paths (topics) independent of one another.
• Maintain a clean shape of history (see Section B.2.2 on page 292).
Design
The design of this workflow is based on the observation that meeting the highlighted goal makes
the other goals easy. It is based on branchy development in which each branch has a well-defined
purpose.
We define two branch types:
• Topic Branch
– Commits represent changes (real work)
– Distinguished by feature (one topic per feature or fix)
– Named locally by each developer (describe purpose of work)
– Heads not published (no named branch on server)
• Integration Branch
B.2. Workflow
281
Meaning
Symbol
Branch name
master
Current branch
*master
Commit with parent in same
branch
...
C1
...
Commit with two parents
(merge)
C1
C2
Table B.1: Git graphs notation.
C1
master
C2
C3
C4
C5
topic
Figure B.1: Topic branch.
– Commits represent merges (merge topics together)
– Distinguished by stability (release maintenance, release preparation, development edge)
– Named everywhere
– Heads published on server
Notation
This chapter uses Git Directed Acyclic Graphs (DAG) to depict commit history:
Topic branches generally consist of a linear sequence of commits forked off an integration branch:
Integration branches generally consist of a sequence of merge commits:
Published Branches
We publish an integration branch for each stage of development:
• release: Release maintenance (high stability). Only bug fixes should be published here. Only
the release manager can push here.
282
Appendix B. ITK Git Workflow
C6
C7
C1
C2
C8
C9
C3
C4
C5
master
Figure B.2: Merge commits into the master branch.
C1
master
Figure B.3: Locate at master.
• master: Release preparation (medium stability). Only mature features and bug fixes should
be published here.
Topic branches are not published directly; their names exist only in each developer’s local repositories.
Development
We cover below the steps to take during each phase of development.
These instructions generally provide all arguments to git push commands. Some
people prefer to use git push with no additional arguments to push the current tracking branch.
Run the command
Initial Setup
git config --global push.default tracking
to establish this behavior. See the git config man-page for details.
New Topic Create a new topic branch for each separate feature or bug fix. Always start the topic
from a stable integration branch, usually master. If the topic fixes a bug in the current release, use
release. In the following section we review the steps and commands to create, develop, and publish
a topic branch based on master.
Update master to base work on the most recently integrated features.
git checkout master
B.2. Workflow
283
C1
C2
master
Figure B.4: Bring most recent changes to master.
master
C1
C2
topic
Figure B.5: Create a local topic branch.
git pull
Create the local topic branch. Use a meaningful name for topic (see Section B.2.2 on page 293).
git checkout -b topic
This is where the real work happens. Edit, stage, and commit files repeatedly as needed for your
work.
During this step, avoid the B.2.2 from an integration branch. Keep your commits focused on the
topic at hand.
edit files
git add -- files
git commit
edit files
git add -- files
git commit
When the topic is ready for publication it must be merged into master. It should be the current local
C1
C2
master
C3
topic
Figure B.6: Commit changes.
284
Appendix B. ITK Git Workflow
C1
C2
master
C3
C4
topic
Figure B.7: Commit last changes.
C1
C2
master
C3
C4
topic
Figure B.8: Checkout master and pull changes.
branch when you merge.
Switch to master and update it.
git checkout master
git pull
Merge the topic and test it.
git merge topic
See Section B.2.2 on page 305 to resolve any conflict that may arise when merging.
Finally, publish the change.
git push origin master
See Section B.2.2 on page 295 and Section B.2.2 on page 296 to resolve any conflict that may arise
when publishing the change.
Mature Topic
When a topic is ready for inclusion in the next release, we merge it into master.
C1
C2
C5
C3
C4
topic
Figure B.9: Merge topic branch into master.
master
B.2. Workflow
285
origin/master
C1
C2
C5
C3
topic
C4
master
Figure B.10: Publish the change.
C1
C2
master
C3
topic
C4
Figure B.11: Checkout master.
Update master to get the latest work by others. We will merge the topic branch into it.
git checkout master
git pull
Merge the topic and test it.
git merge topic
See Section B.2.2 on page 305 to resolve any conflict that may arise when merging the branch.
Delete the local branch.
C5
C1
C2
C6
C3
C4
topic
Figure B.12: Pull latest changes into master.
master
286
Appendix B. ITK Git Workflow
C5
C1
C2
C6
C3
C7
master
topic
C4
Figure B.13: Merge the topic branch into master.
C5
C1
C2
C6
C3
C7
master
C4
Figure B.14: Delete the local topic branch.
git branch -d topic
Finally, publish the change.
git push origin master
See Section B.2.2 on page 295 and Section B.2.2 on page 296 to resolve any conflict that may arise
when publishing the change.
Sometimes we need to continue work on an old topic that has already been merged to
an integration branch and for which we no longer have a local topic branch. To revive an old topic,
we create a local branch based on the last commit from the topic (this is not one of the merges into
Old Topic
origin/master
C5
C1
C2
C6
C3
C4
Figure B.15: Publish the change.
C7
master
B.2. Workflow
287
C5
C1
C2
C6
C3
C4
C7
master
topic b235725
Figure B.16: Local topic branch started from a given commit and switch to it.
an integration branch).
First we need to identify the commit by its hash. It is an ancestor of the integration branch into
which it was once merged, say master. Run git log with the --first-parent option to view the
integration history:
git log --first-parent master
commit 9057863...
Merge: 2948732 a348901
...
Merge branch topicA
commit 2948732...
Merge: 1094687 b235725
...
Merge branch topicB
commit 1094687...
Merge: 8267263 c715789
...
Merge branch topicC
Locate the merge commit for the topic of interest, say topicB. Its second parent is the commit from
which we will restart work (b235725 in this example).
Create a local topic branch starting from the commit identified above.
git checkout -b topic b235725
Continue development on the topic.
edit files
git add -- files
git commit
288
Appendix B. ITK Git Workflow
C5
C1
C2
C6
C7
C3
C4
C8
topic
master
Figure B.17: Commit files to topic.
C5
C1
C2
C6
C7
C3
C4
C8
master
C9
topic
Figure B.18: Further continue developing and commit files to topic.
edit files
git add -- files
git commit
When the new portion of the topic is ready, merge it into master and test.
git checkout master
git pull
git merge topic
Publish master.
git push origin master
C5
C1
C2
C6
C7
C10
master
C3
C4
C8
C9
topic
Figure B.19: Merge into master.
B.2. Workflow
289
extra-topic
C5
C6
C1
C2
C7
C8
C3
C4
0a398e5
C9
C10
origin/master
topic
Figure B.20: Fetch the upstream integration branch other-topic.
extra-topic
C5
C6
C1
C2
C7
C3
C4
C9
C10
C8
origin/master
other-topic
topic
Figure B.21: Create a local branch other-topic from commit 0a398e5.
Occasionally you may realize that you need the work from another topic to complete work on your topic. In this case your topic depends on the other topic, so merging the other
topics into yours is legitimate (see Section B.2.2 on page 295). Do not merge an integration branch
that has the other-topic branch name. Use the instructions below to merge only the other-topic
branch without getting everything else.
Dependent Topic
Fetch the upstream integration branch that has the other-topic branch, say master.
git fetch origin
Use git log --first-parent origin/master to find the commit that merges other-topic. The
commit message gives you the name of the other topic branch (we use other-topic here as a placeholder). The second parent of the commit (0a398e5 in this example) is the end of the other-topic
branch. Create a local branch from that commit.
git branch other-topic 0a398e5
Merge the other-topic branch into your topic.
290
Appendix B. ITK Git Workflow
extra-topic
C5
C6
C1
C2
C7
C3
C4
C9
C10
origin/master
C8
topic
C11
Figure B.22: Merge other-branch into topic branch.
C5
C1
C2
C3
C4
C9
C10
C11
topic
Figure B.23: topic branch shape.
git merge other-topic
git branch -d other-topic
(It is also possible to run git merge 0a398e5 and then use git commit --amend to write a nice
commit message.)
The topic branch now looks like this:
Note that after the merge, the other-topic is reachable from your topic but the extra-topic has not
been included. By not merging from the integration branch we avoided bringing in an unnecessary
dependency on the extra-topic. Furthermore, the message “Merge branch ’other-topic’ into topic” is
very informative about the purpose of the merge. Merging the whole integration branch would not
be so clear.
Each published integration branch (see Section B.2.2 on page 281)
has a defined level of stability. Express this relationship by merging more-stable branches into lessstable branches to ensure that they do not diverge. After merging a mature topic to master, we
merge master into release:
Merge Integration Branches
B.2. Workflow
291
C5
C1
C2
C6
C3
C4
C8
C9
C7
master
release
Figure B.24: Update master and release.
C5
C1
C2
C6
C3
C4
C8
C9
C7
master
C10
release
Figure B.25: Merge master into release.
Update master and then release:
git
git
git
git
checkout master
pull
checkout release
pull
Merge master into release:
git merge master
Finally, publish the change.
git push origin release
See Section B.2.2 on page 295 and Section B.2.2 on page 296 to resolve any conflict that may arise
when publishing the change.
292
Appendix B. ITK Git Workflow
C5
C1
C2
C6
C3
C4
C8
C9
C7
master
C10
release
origin/release
Figure B.26: Publish to release.
C5
C1
C2
C6
C7
C15
master
C3
C4
C8
C14
topic
C9
C11
C13
C16
release
C10
C12
Figure B.27: Complex history graph in Git.
Discussion
History Shape The history graphs produced by this workflow may look complex compared to the
fully linear history produced by a rebase workflow (used by CVS and Subversion):
However, consider the shape of history along each branch. We can view it using Git’s –first-parent
option. It traverses history by following only the first parent of each merge. The first parent is
the commit that was currently checked out when the git merge command was invoked to create the
merge commit. By following only the first parent, we see commits that logically belong to a specific
branch.
git log --first-parent topic
B.2. Workflow
293
C2
C3
C4
C8
C14
topic
Figure B.28: Parent commit in topic branch.
C5
C1
C2
C4
C14
C6
C7
C15
master
Figure B.29: Parent commit in master branch.
git log --first-parent master
git log --first-parent release
Each branch by itself looks linear and has only commits with a specific purpose. The history behind
each commit is unique to that purpose. Topic branches are independent, containing only commits
for their specific feature or fix. Integration branches consist of merge commits that integrate topics
together.
Note that achieving the nice separation of branches requires understanding of the above development
procedure and strict adherence to it.
Naming Topics This section uses the placeholder topic in place of a real topic name. In practice,
substitute for a meaningful name. Name topics like you might name functions: concise but precise.
A reader should have a general idea of the feature or fix to be developed given just the branch name.
C4
C14
C9
C10
C11
C13
C16
C12
Figure B.30: Parent commit in release branch.
release
294
Appendix B. ITK Git Workflow
Note that topic names are not published as branch heads on the server, so no one will ever see a
branch by your topic name unless they create it themselves. However, the names do appear in the
default merge commit message:
git checkout master
git merge topic
git show
...
Merge branch 'topic' into master
...
These merge commits appear on the integration branches and should therefore describe the changes
they integrate. Running git log --first-parent as described in Section B.2.2 will show only
these merge commits, so their messages should be descriptive of the changes made on their topics.
If you did not choose a good branch name, or feel that the merge needs more explanation than the
branch name provides, amend the commit to update the message by hand:
git commit --amend
Merge branch 'topic' into master
(edit the message)
Avoid the “urge to merge” from an integration branch into your topic. Keep commits on your topic focused on the feature or fix under development.
Urge to Merge
Merge your work with others when you are finished with it by merging into an
integration branch as documented above. Avoid habitual merges from an integration branch; doing
so introduces unnecessary dependencies and complicates the shape of history (see Section B.2.2 on
page 292).
Habitual Merges
Many developers coming from centralized version control systems have trained themselves to regularly update their work tree from the central repository (e.g. “cvs update”). With those version
control systems this was a good habit because they did not allow you to commit without first integrating your work with the latest from the server. When integrating the local and remote changes
resulted in conflicts, developers were forced to resolve the conflicts before they could commit. A
mistake during conflict resolution could result in loss of work because the local changes might have
been lost. By regularly updating from the server, developers hoped to avoid this loss of work by
resolving conflicts incrementally.
Developers using Git do not face this problem. Instead, one should follow a simple motto: “commit
first, integrate later”. There is no risk that your work will be lost during conflict resolution because
all your changes have been safely committed before attempting to merge. If you make a mistake
while merging, you always have the option to throw away the merge attempt and start over with a
clean tree.
B.2. Workflow
295
One reason to merge other work into your topic is when you realize that your
topic depends on it. See Section B.2.2 on page 289 for help with this case.
Legitimate Merges
Occasionally one may merge directly from master if there is a good reason. This is rare, so bring
up the reason on your project discussion forum first. Never merge release into a topic under any
circumstances!!!
Troubleshooting
Here we document problems one might encounter while following the workflow instructions above.
This is not a general Git troubleshooting page.
Trouble Merging
Trouble Pushing
Remote End Hung up Unexpectedly
Pushing may fail with this error:
git push
fatal: The remote end hung up unexpectedly
This likely means that you have set a push URL for the remote repository. You can see the URL to
which it tries to push using -v:
git push -v
Pushing to git://public.kitware.com/Project.git
fatal: The remote end hung up unexpectedly
The git:// repository URL may not be used for pushing; it is meant for efficient read-only anonymous access only. Instead you need to configure a SSH-protocol URL for pushing:
git config remote.origin.pushurl git@public.kitware.com:Project.git
(Note that pushurl requires Git ¿= 1.6.4. Use just url for Git ¡ 1.6.4.). The URL in the above
example is a placeholder. In practice, use th push URL documented for your repository.
The above assumes that you want to push to the same repository that you originally cloned. To push
elsewhere, see help for git push and git remote.
296
Appendix B. ITK Git Workflow
C1
C2
master
C3
C5
release
C6
C8
C7
C4
topic
origin/release
other-topic
Figure B.31: Unreachable origin/release branch.
When trying to publish new merge commits on an integration branch, perhaps
release, the final push may fail:
Non-Fast-Forward
git push origin release
To ...
! [rejected]
release -> release (non-fast-forward)
error: failed to push some refs to '...'
To prevent you from losing history, non-fast-forward updates were rejected
Merge the remote changes before pushing again. See the 'Note about
fast-forwards' section of 'git push --help' for details.
This means that the server’s release refers to a commit that is not reachable from the release you
are trying to push:
This is the Git equivalent to when cvs commit complains that your file is not up-to-date, but now it
applies to the whole project and not just one file. Git is telling you that it cannot update release on
the server to point at your merge commit because that would throw away someone else’s work (such
as other-topic). There are a few possible causes, all of which mean you have not yet integrated your
work with the latest from upstream:
• You forgot to run git pull before git merge so you did not have everything from upstream.
• Someone else managed to merge and push something into release since you last ran git
pull.
Some Git guides may tell you to just git pull again to merge upstream work into yours. That
approach is not compatible with the goals of this workflow. We want to preserve a clean shape of
history (see Section B.2.2 on page 292).
The solution is to throw away your previous merge and try again, but this time start from the latest
upstream work:
B.2. Workflow
297
C1
C2
master
C3
C6
C8
C7
release
C4
topic
origin/release
other-topic
Figure B.32: Start from latest upstream commit.
C1
C2
C6
C7
C8
master
C3
C4
C9
release
topic
other-topic
Figure B.33: Merge the topic branch into the local release branch.
git reset --hard origin/release
git merge topic
Now your release can reach the upstream work as well as yours. Publish it.
git push origin release
See git rerere to help avoid resolving the same conflicts on each merge attempt.
One goal of this workflow is to preserve a clean shape of
history (see Section B.2.2 on page 292). This means that a --first-parent traversal of an integration branch, such as master, should see only the merge commits that integrate topics into the
branch:
First-Parent Sequence Not Preserved
298
Appendix B. ITK Git Workflow
C1
C2
master
C3
C6
C8
origin/release
C9
C7
topic
C4
release
other-topic
Figure B.34: Publish the release branch.
C1
C5
...
C2
C3
C7
C4
C6
C8
master
Figure B.35: Traversal of the master integration branch.
B.2. Workflow
299
F
G
...
E
D
C
H
J
U
T
master@1
B
A
M
master
topic
Figure B.36: Server’s integration branch history shape.
F
G
K
...
E
D
H
J
C
B
A
master@1
Figure B.37: First parent traversal of master before update.
The commits on the individual topic branches are not included in the traversal. This provides a
medium-level overview of the development of the project.
We enforce the shape of history on the server’s integration branches using an update hook at pushtime. Each update must point its branch at a new commit from which a first-parent traversal
reaches the old head of the branch:
A first-parent traversal of master from before the update (master@1) sees A B C D:
A first-parent traversal of master from after the update sees M A B C D:
The above assumes correct history shape. Now, consider what happens if merge M is incorrectly
made on the topic branch:
Now a first-parent traversal of master from after the update sees M’ T U B C D:
This not only shows details of the topic branch, but skips over A altogether! Our update hooks will
F
G
K
...
E
D
H
J
T
C
B
A
Figure B.38: First parent traversal of master after update.
M
master
300
Appendix B. ITK Git Workflow
F
G
K
...
E
D
C
B
H
J
U
T
M’
master@1
A
master
topic
Figure B.39: Incorrect merge of M on branch topic.
F
G
...
E
D
C
B
H
J
U
T
M’
master
topic
Figure B.40: First-parent traversal of master branch.
B.2. Workflow
301
F
G
...
E
H
J
origin/master
D
C
B
master
Figure B.41: Checkout master.
origin/master
F
G
...
E
D
H
J
U
C
B
master
Figure B.42: Commit on master branch.
reject the push in this case because the new master cannot see the old one in a first-parent traversal.
There are a few possible causes and solutions to the above problem, but all involve non-strict compliance with the workflow instructions. A likely cause is that you did not create a local topic branch
but instead committed directly on master and then pulled from upstream before pushing:
wrong$ git checkout master
wrong$ edit files
wrong$ git add files
wrong$ git commit
wrong$ edit files
wrong$ git add files
wrong$ git commit
wrong$ git push origin master
Rejected as non-fast-forward (see Section B.2.2 on page 296).
wrong$ git pull
302
Appendix B. ITK Git Workflow
origin/master
F
G
...
E
D
C
H
J
U
T
B
A
M
master
Figure B.43: Additional commit on master branch.
F
G
...
E
D
C
B
H
J
U
T
M’
A
origin/master
master
Figure B.44: Pull from upstream.
wrong$ git push origin master
Rejected with the first parent sequence not preserved error (see Section B.2.2 on page 297).
The solution in this case is to recreate the merge on the proper branch.
First, create a nicely-named topic branch starting from the first-parent of the incorrect merge.
git branch topic 'masterˆ1'
Then reset your local master to that from upstream.
F
G
...
E
D
C
B
H
J
U
T
M’
A
origin/master
master
topic
Figure B.45: Create a topic branch starting from the first-parent of the incorrect merge.
B.2. Workflow
303
origin/master
F
G
...
E
D
C
H
J
U
T
B
A
topic
master
Figure B.46: Reset the local master branch to upstream master.
F
G
...
E
D
C
H
J
U
T
B
A
M
master
topic
Figure B.47: Merge the topic branch.
git reset --hard origin/master
Now create the correct merge commit as described in the workflow instructions above.
git merge topic
git push origin master
git branch -d topic
Topics Must Be Merged
origin/master
F
G
...
E
D
C
H
J
U
T
B
A
Figure B.48: Delete the local topic branch.
M
master
304
Appendix B. ITK Git Workflow
C1
C2
C3
master
C4
C5
other-topic
C6
C7
topic
Figure B.49: Conflicting topic branches.
C1
C2
C3
C8
C4
C5
other-topic
C6
C7
topic
master
Figure B.50: Topic-to-single branch resolution approach.
Conflicts
This section documents conflict resolution in a topic-based branchy workflow.
Whenever two paths of development make different changes to the same initial content conflicts
may occur when merging the branches.
Consider two conflicting topic branches, topic and other-topic, with the
latter already merged to master:
Single Integration Branch
An attempt to merge topic into master will fail with conflicts. One may use the following approaches
to resolve the situation:
• Merge the topic to the branch
• Merge the branch to the topic
If one performs the merge in a local work tree it is possible to simply
resolve the conflicts and complete the merge:
Topic-to-Single-Branch
Branch-to-Topic Since a developer works on a topic branch locally one may simply merge the
conflicting integration branch into the topic and resolve the conflicts:
In order to maintain a good shape of history one may then merge the topic into the integration branch
without allowing a fast-forward (merge --no-ff):
B.2. Workflow
305
C1
C2
C7
master
C3
C4
other-topic
C5
C6
topic
C8
Figure B.51: Branch-to-Topic resolution approach.
C1
C2
C7
C3
C4
C5
C6
C9
master
other-topic
topic
C8
Figure B.52: Merge disallowing fast-forward (--no-ff).
In a workflow using multiple integration branches one must deal
differently with conflicting topics. Consider two conflicting topic branches, topic and other-topic,
with the latter already merged to release:
Multiple Integration Branches
An attempt to merge topic into release will fail with conflicts. One may use the following approaches
to resolve the situation:
• Merge the topic to the branch (see Section B.2.2).
• Merge one topic into the other (see Section B.2.2).
• Merge both topics into a resolution topic (see Section B.2.2).
C1
C2
master
C3
C4
topic
C5
C6
other-topic
...
C7
release
Figure B.53: Conflicting topic branches.
306
Appendix B. ITK Git Workflow
C1
C2
master
C3
C4
topic
C5
C6
other-topic
...
C7
C8
release
Figure B.54: Merge locally.
C1
C2
C9
master
C3
C4
topic
C5
C6
other-topic
...
C7
C8
release
Figure B.55: Merge topic.
Note that one may not merge the branch into the topic as in the single-integration-branch case because release may never be merged into a topic.
Topic-to-Branch If one performs the merge in a local work tree it is possible to simply resolve the
conflicts and complete the merge:
However, the topics eventually must be merged to master. Assume topic is merged first:
An attempt to merge other-topic into master will fail with the same conflicts!
The only branch that contains a resolution to these conflicts is release, but that may not be merged
to master. Therefore one must resolve the conflicts a second time.
If the second resolution is not byte-for-byte identical to the first then the new master will not merge
cleanly into release:
Then one must resolve conflicts a third time!
This approach works with manual merging but requires care.
B.2. Workflow
307
C1
C2
C9
?
C3
C4
topic
C5
C6
other-topic
...
C7
master
C8
release
Figure B.56: Merge conflict when attempting to merge other-topic into master.
C1
C2
C9
C10
C3
C4
topic
C5
C6
other-topic
...
C7
C8
master
?
release
Figure B.57: master not merging cleanly into release if conflicts have not been resolved.
308
Appendix B. ITK Git Workflow
C1
C2
master
C3
C6
C4
C5
topic
other-topic
C7
...
C8
release
Figure B.58: Manually merge other-topic into topic.
C1
C2
master
C3
C6
C7
C4
C5
topic
C9
release
other-topic
...
C8
Figure B.59: Merge into release.
Topic-to-Topic The design (see Section B.2.2 on page 280) of our topic-based workflow guarantees
that work is always committed on topic branches and never directly on an integration branch. If
conflicts occur while merging a topic into an integration branch it means that the topic conflicts with
another topic that has already been merged.
One may manually merge the conflicting other-topic into one’s own topic and resolve the conflicts:
Then topic will merge cleanly into release:
Later, topic may be merged cleanly into master to bring in both topics (or just topic if other-topic
has already been merged):
Finally, master may be merged cleanly into release:
Note that this produces an artificial topic dependency (see Section B.2.2 on page 289) introduced by
the conflict resolution commit. See the B.2.2 approach to avoid this problem.
The B.2.2 approach introduces an artificial topic dependency because it asymmetrically favors one topic over another. Instead one may use a third topic to resolve the conflicts.
Resolution Topic
One may start a new resolve/topic/other-topic branch from topic, merge other-topic into it, and
resolve the conflicts:
B.2. Workflow
309
C1
C2
C10
C3
C6
C4
C5
topic
C9
release
master
other-topic
C7
...
C8
Figure B.60: Merge into master.
C1
C2
C10
C3
C6
C4
C5
master
topic
other-topic
C7
...
C8
C9
C11
release
Figure B.61: Merge master into release.
C1
C2
master
C3
C4
topic
C7
C5
C6
resolve topic/other-topic
other-topic
...
C8
release
Figure B.62: Start conflict resolution branch.
310
Appendix B. ITK Git Workflow
C1
C2
master
C3
C5
C6
C4
topic1
C7
resolve topic/other-topic
C9
release
other-topic
...
C8
Figure B.63: Merge into release.
C1
C2
C10
C3
C5
C6
C4
master
topic
C7
resolve topic/other-topic
C9
release
other-topic
...
C8
Figure B.64: Merge topic branch.
The resolution topic will merge cleanly into release to bring in the changes from topic through the
conflict resolution commit:
Since topic and other-topic are still independent either may be merged to master first. Assume topic
is merged first:
As in the B.2.2 approach, an attempt to merge other-topic directly into master will fail with the
original conflicts but now we have a topic containing the resolution commit independent of next.
One may merge the resolution topic to master to bring in the changes from other-topic and the
conflict resolution:
Finally, master may be merged cleanly into release:
B.2. Workflow
311
C1
C2
C10
C3
C5
C6
C4
C11
master
topic
C7
resolve topic/other-topic
C9
release
other-topic
...
C8
Figure B.65: Merge conflict resolution branch into master.
C1
C2
C10
C3
C4
C6
resolve topic/other-topic
other-topic
...
C8
master
topic
C7
C5
C11
C9
C12
Figure B.66: Merge into release.
release
312
Appendix B. ITK Git Workflow
B.2.3 Publish
Push Access
Authorized developers may publish work directly to a repository using Git’s SSH protocol.
Note that we may not grant all contributors push access to any given repository. The distributed
nature of Git allows contributors to retain authorship credit even if they do not publish changes
directly.
All publishers share the git@public.kitware.com account but each uses a unique
SSH key for authentication. If you do not have a public/private SSH key pair, generate one:
Authentication
ssh-keygen -C 'you@yourdomain.com'
Generating public/private rsa key pair.
Enter file in which to save the key (\$HOME/.ssh/id\_rsa):
Enter passphrase (empty for no passphrase): (use-a-passphrase!!)
Enter same passphrase again: (use-same-passphrase!!)
Your identification has been saved in \$HOME/.ssh/id\_rsa.
Your public key has been saved in \$HOME/.ssh/id\_rsa.pub.
To request access, fill out the Kitware Password form. Include your SSH public key, id rsa.pub,
and a reference to someone our administrators may contact to verify your privileges.
If you are familiar with generating an SSH key on Linux or macOS, you can
follow the same procedure on Windows in a Git Bash prompt. There is an ssh-keygen program
installed with Git for Windows to help you set up an SSH identity on a Windows machine. By
default it puts the .ssh directory in the HOME directory, which is typically C:
Users
Username.
SSH on Windows
Alternatively, you can also set up a “normal” Windows command prompt shell such that it will work
with Git for Windows (see Section B.1.1) on page 277, without ever invoking the Git Bash prompt
if you like. If you install Git for Windows and accept all its default options, “git” will not be in
the PATH. However, if you add C:
Program Files (x86)
Git
cmd to your PATH, then only the two commands git and gitk are available to use via *.cmd script
wrappers installed by Git for Windows. Or, if you add C:
Program Files (x86)
Git
bin to your PATH, then all of the command line tools that git installs are available.
B.2. Workflow
313
The full PuTTY2 suite of tools includes an application called PuTTYgen. If you already have a
private key created with PuTTYgen, you may export it to an OpenSSH identity file. Open the key
using PuTTYgen and choose Conversions ¿ Export OpenSSH key from the menu bar. That will allow
you to save an id rsa file for use in the .ssh directory. You can also copy and paste the public
key portion of the key from the PuTTYgen text field to save into an id rsa.pub file if you like. Or
email it to whoever needs the public side of your key pair.
If you routinely set up your own command prompt environment on Windows, using Git for
Windows from that environment is a cinch: just add the full path to either Git
cmd or Git
bin to your PATH. (Or, write your own git.cmd wrapper that is in your PATH that simply calls the
git.cmd installed with msysGit.) And make sure you have a HOME environment variable that points
to the place where the .ssh directory is.
AuthenticationTest When your SSH public key has been installed for git@public.kitware.com,
you may test your SSH key setup by running
ssh git@public.kitware.com info
If your key is correctly configured you should see a message reporting your email address followed
by a list of access permissions. If you get something like Permission denied then add -v options to
your ssh command line to diagnose the problem:
ssh -v git@public.kitware.com info
Do not attempt to git push until the ssh-only test succeeds.
Pushing Git automatically configures a new clone to refer to its origin through a remote called
origin. Initially one may fetch or pull changes from origin, but may not push changes to it.
In order to publish new commits in a repository, developers must configure a push URL for the
origin. Use git config to specify an SSH-protocol URL:
git config remote.origin.pushurl git@public.kitware.com:repo.git
The actual URL will vary from project to project. (Note that pushurl requires Git ¿= 1.6.4. Use
just url for Git ¡ 1.6.4.)
Failing to do so with result in the error message fatal: The remote end hung up unexpectedly.
2 https://www.chiark.greenend.org.uk/
˜sgtatham/putty/
314
Appendix B. ITK Git Workflow
Once your push URL is configured and your key is installed for git@public.kitware.com then
you can try pushing changes. Note that many repositories use an update hook to check commit as
documented in Section B.2.4 on page 316.
Patches
Git allows anyone to be a first-class developer on any published project. One can clone a public repository, commit locally, and publish these commits for inclusion upstream. One method of
sending commits upstream is to supply them as patches.
See these links for more help:
• Pro Git Book, Chapter 5: Distributed Git3
• Everyday Git: Integrator4
Creating Patches
Construct your commits on a local topic branch, typically started from the up-
stream master:
git checkout -b my-cool-feature origin/master
edit files
git add -- files
git commit
Begin each commit message with a short one-line summary of its change, suitable for use as an
email subject line. Then leave a blank line and describe the change in detail as one might write in
an email body.
When the patch(es) are ready for publication to upstream developers, use the git format-patch
command to construct the patch files:
git format-patch -M origin/master
Git will write out one patch file per commit. Each patch file is formatted like a raw email message
and includes enough information to reconstruct the commit message and author.
Sending Patches
The patch files created in the preceding step will be named with the form
NNNN-Subject-line-of-commit-message.patch
3 https://git-scm.com/book/en/v2
4 https://git-scm.com/docs/giteveryday
B.2. Workflow
315
where NNNN is an index for the patch within the series. These files may be attached in bug trackers
or attached to email messages.
A patch series may also be sent directly as email. Use git config --global to set sendemail.*
configuration entries that tell Git how to send email from your computer (one-time setup per user
per machine). Then use the git send-email command:
git send-email *.patch --to='Some One ' --cc='Someone Else '
One may receive patches as attachments in a bug tracker or as attachments to
email messages. Save these files to your local disk. One may also receive patches inlined in email
messages. In this case, save the whole message to your local disk (typically as .eml files). (If your
local mail client uses maildir format mailboxes each message is already its own file.)
Applying Patches
Create a local topic branch on which to replay the patch series:
git checkout -b cool-feature origin/master
Now use git am to apply the patch series as local commits:
git am --whitespace=fix /path/to/*.patch
Review the changes using
git log -p origin/master..
or the method of your choice. Note that the author of each commit is the contributor rather than
yourself. Build, test, and publish the changes normally.
If the git am command fails with a message like
Patch format detection failed.
this means that the patch was not generated with git format-patch and transmitted correctly.
Either ask the contributor to try again using the above patch creation instructions, or apply each
patch separately using git apply:
git apply --whitespace=fix /path/to/0001-First-change.patch
git commit --author='Contributor Name '
316
Appendix B. ITK Git Workflow
B.2.4 Hooks
Setup
The git commit command creates local commits. A separate git push step is needed to publish
commits to a repository. The server enforces some rules (see Section B.2.4 on page 319) on the
commits it accepts and will reject non-conforming commits. In order to push rejected commits, one
must edit history locally to repair them before publishing.
Since it is possible to create many commits locally and push them all at once, we provide local Git
hooks to help developers keep their individual commits clean. Git provides no way to enable such
hooks by default, giving developers maximum control over their local repositories. We recommend
enabling our hooks manually in all clones.
Git looks for hooks in the .git/hooks directory within the work tree of a local repository. Create a
new local repository in this directory to manage the hooks:
cd .git/hooks
git init
cd ../..
Choose one of the following methods to install or update the hooks. The hooks will then run in the
outer repository to enforce some rules on commits.
Many repositories provide a hooks branch. It will have already been fetched into your
local clone. Pull it from there:
Local Pull
git fetch origin
cd .git/hooks
git pull .. remotes/origin/hooks
cd ../..
Direct Pull
If you did not clone from a repository you may not have a hooks branch. Pull it from :
cd .git/hooks
git pull git://public.kitware.com/.git hooks
cd ../..
where .git is the name of your project repository.
Local
The above sequences maintain the following local hooks in your repository. See Git help on
https://git-scm.com/docs/githooks for more details.
B.2. Workflow
pre-commit
317
This runs during git commit. It checks identity and content of changes:
• Git user.name and user.email are set to something reasonable.
• Git’s standard whitespace checks (see help on git diff --check).
• The staged changes do not introduce any leading tabs in source files (we i.ndent with spaces)
• File modes look reasonable (no executable .cxx files, scripts with shebang lines are executable).
• File size is not too large (do not commit big data files; prints limit and instructions on rejection).
• Submodule updates are staged alone or explicitly allowed (prints instructions on rejection).
One of Git’s standard whitespace checks is to reject trailing whitespace on lines that were added or
modified. Many people consider extra space characters at the end of a line to be an unprofessional
style (including Git’s own developers), but some don ot care. Text editors typically have a mode to
highlight trailing whitespace:
• Emacs
(custom-set-variables '(show-trailing-whitespace t))
• Vim
:highlight ExtraWhitespace ctermbg=red guibg=red
:match ExtraWhitespace /\s\+\$/
• Microsoft Visual Studio To toggle viewing of white space characters, with a source file
document active, choose the menu item:
Edit > Advanced > View White Space
(2-stroke keyboard shortcut: Ctrl+R, Ctrl+W)
• Notepad++ (v7.5.1) To eliminate trailing white space, choose the menu item:
Edit > Blank Operations > Trim Trailing Space
To toggle viewing of white space characters, choose from the menu items:
318
Appendix B. ITK Git Workflow
View > Show Symbol > (multiple items, choose one...)
If you really don’t want to keep your code clean of trailing whitespace, you can disable this part of
Git’s checks locally:
git config core.whitespace "-blank-at-eol"
commit-msg
This runs during git commit. It checks the commit message format:
• The first line must be between 8 and 78 characters long. If you were writing an email to
describe the change, this would be the Subject line. Use the pre-defined prefixes (e.g. ENH: or
BUG:); they are valuable to allow the user get a fast understanding of the change.
• The first line must not have leading or trailing whitespace.
• The second line must be blank, if present.
• The third line and below may be free-form. Usually, a summary of the commit changes is
written. Try to keep paragraph text formatted in 72 columns (this is not enforced).
GUI and text-based tools that help view history typically use the first line (Subject line) from the
commit message to give a one-line summary of each commit. This allows a medium-level view of
history, but works well only if developers write good Subject lines for their commits.
Examples of improper commit messages:
Fixed
This is too short and not informative at all.
I did a really complicated change and I am trying to describe the entire thing
with a big message entered on the command line.
Some good tips on why good commit messages matter can be found in the post How to Write a Git
Commit Message5 .
Many CVS users develop the habit of using the -m commit option to specify the whole message on
the command line. This is probably because in CVS it is hard to abort a commit if it already brought
up the message editor. In Git this is trivial. Just leave the message blank and the whole commit
will be aborted. Furthermore, since commits are not published automatically it is easy to allow the
commit to complete and then fix it with git commit --amend.
5 https://chris.beams.io/posts/git-commit/
B.2. Workflow
319
Server
Many public.kitware.com repositories have server-side hooks.
The update hook runs when someone tries to update a ref on the server by pushing. The
hook checks all commits included in the push:
Update
• Commit author and committer must have valid email address domains (DNS lookup succeeds).
• Commit message does not start with WIP:. (Use the prefix locally for work-in-progress that
must be rewritten before publishing.)
• Changes to paths updated by robots (such as Utilities/kwsys) are not allowed.
• No “large” blobs may be pushed. The limit is set on a per-repository basis and is typically 1
MB or so.
• No CRLF newlines may be added in the repository (see core.autocrlf in git help
config).
• Submodules (if any) must be pushed before the references to them are pushed.
B.2.5 TipsAndTricks
Editor support
Emacs users: if you put this line in your .emacs file:
(setq auto-mode-alist (cons '("COMMIT\_EDITMSG\$" . auto-fill-mode) auto-mode-alist))
Git will automatically wrap your commit messages, which is what good Git etiquette requires.
Shell Customization
Bash users: Git comes with a set of completion options that are very useful. The
location of the file varies depending on your system:
Bash Completion
source /opt/local/share/doc/git-core/contrib/completion/git-completion.bash # Mac with git installed by Mac
source /usr/share/bash-completion/git
# Linux
source /etc/bash\_completion.d/git
# Linux debian/gentoo
320
Appendix B. ITK Git Workflow
Bash Prompt If you are using the bash shell, you can customize the prompt to show which Git
branch is active. Here are the commands for your /.bashrc file:
source
export
export
export
export
/etc/bash\_completion.d/git # or one of the appropriate paths from the above section
GIT\_PS1\_SHOWDIRTYSTATE=1
GIT\_PS1\_SHOWUNTRACKEDFILES=1
GIT\_PS1\_SHOWUPSTREAM="verbose"
PS1="[\[\e[01;34m\]\W\[\e[31m\]\$(\_\_git\_ps1 " (%s)")\[\e[00m\]]\[\e[00m\]
For more information on the options, see the comments in the top of the bash completion script.
Renaming
Git does not explicitly track renames. The command
git mv old new
is equivalent to
mv old new
git add new
git rm old
Neither approach records the rename outright. However, Git’s philosophy is “dumb add, smart
view”. It uses heuristics to detect renames when viewing history after-the-fact. It even works when
the content of a renamed file changes slightly.
In order to help Git efficiently detect the rename, it is important to remove the old file and add the
new one in one commit, perhaps by using git mv or the above 3-step procedure. If the new file
were added in one commit and the old file removed in the next, Git would report this as a copy
followed by a removal. Its copy-detection heuristics are more computationally intensive and must
be explicitly enabled with the -C option to relevant operations (such as git blame).
APPENDIX
THREE
CODING STYLE GUIDE
This chapter describes the ITK Coding Style. Developers must follow these conventions when
submitting contributions to the toolkit.
The following coding-style guidelines have been adopted by the ITK community. To a large extent
these guidelines are a result of the fundamental architectural and implementation decisions made
early in the project. For example, the decision was made to implement ITK with a C++ core using
principles of generic programming, so the rules are oriented towards this style of implementation.
Some guidelines are relatively arbitrary, such as indentation levels and style. However, an attempt
was made to find coding styles consistent with accepted practices. The point is to adhere to a
common style to assist community members of the future to learn, use, maintain, and extend ITK.
A common style greatly improves readability.
Please do your best to be an outstanding member of the ITK community. The rules described here
have been developed with the community as a whole in mind. Any contributor’s code is subject to
style review, and it will likely not be accepted until it is consistent with these guidelines.
C.1 Purpose
The following document is a description of the accepted coding style for the NLM Insight Segmentation and Registration Toolkit (ITK). Developers who wish to contribute code to ITK should read
and adhere to the standards described here.
C.2 Overview
This chapter is organized into the following sections:
• System Overview & Philosophy: coding methodologies and motivation for the resulting
style.
322
Appendix C. Coding Style Guide
• Copyright: the copyright header to be included in all files and other copyright issues.
• Citations: guidelines to be followed when citing others’ work in the documentation.
• Naming Conventions: patterns used to name classes, variables, template parameters, and
instance variables.
• Namespaces: the use of namespaces.
• Aliasing Template Parameter Typenames: guidelines on aliasing template parameter typenames in a class.
• The auto Keyword: when and when not to use the auto keyword.
• Pipelines: useful tips when writing pipelines in ITK.
• Initialization and Assignment: accepted standards for variable initialization and assignment.
• Accessing Members: patterns to be used when accessing class members.
• Code Layout and Indentation: accepted standards for arranging code including indentation
style.
• Empty Arguments in Methods: guidelines for specifying empty argument lists.
• Ternary Operator: accepted standards for using the ternary operator.
• Using Standard Macros (itkMacro.h): use of standard macros in header files.
• Exception Handling: how to add exception handling to the system.
• Messages: accepted guidelines to output messages to the error and standard outputs.
• Concept Checking: specifics on the use of concept checking in ITK.
• Printing Variables: guidelines to print member variable values.
• Checking for Null: accepted standards for checking null values.
• Writing Tests: additional rules specific to writing tests in ITK.
• Doxygen Documentation System: basic Doxygen formatting instructions.
• CMake Style: guidelines to write CMake files.
• Documentation Style: a brief section describing the documentation philosophy adopted by
the Insight Software Consortium.
C.3. System Overview & Philosophy
323
This style guide is an evolving chapter.
Please discuss with the ITK community members if you wish to add, modify, or delete the rules
described in these guidelines.
See http://www.itk.org/ITK/help/mailing.html for more information about joining the ITK community members discussion. This forum is one of the best venues in which to propose changes to these
style guidelines.
C.3 System Overview & Philosophy
The following implementation strategies have been adopted by the ITK community. These directly
and indirectly affect the resulting code style. Understanding these aspects motivates the reasons for
many of the style guidelines described in this chapter.
The principle is that code is read many more times than it is written, and readability is far more
important that writability.
• Readability: the code is intended to be read by humans. Most of the time of debugging code
is spent reading code.
• Consistency: systematically following these guidelines will make the code more robust, and
easier to read, which will at term save time to community members.
• Conciseness: class, method and variable names should be concise, but self-contained. Extraordinarily long names and redundancies should be avoided.
• Language: proper English must be used when writing the code and the documentation.
• Documentation: document the code. Approximately one third of the code should be documentation.
Note as well that ITK follows American English spelling and norms.
C.3.1 Kitware Style
Kitware Style (KWStyle) pre-commit hooks enforce a number of policies on any given patch set
submitted to ITK.
C.3.2 Implementation Language
The core implementation language is C++. C++ was chosen for its flexibility, performance, and
familiarity to consortium members. ITK uses the full spectrum of C++ features including const and
324
Appendix C. Coding Style Guide
volatile correctness, namespaces, partial template specialization, operator overloading, traits, and
iterators.
Currently, C++11 and C++14 features are used in the code whenever they are available.
A growing number of ITK classes offer a Python wrapping. Note that these are wrappers on the
C++ counterparts. ITK’s Python wrappers can be easily installed. See Section 9.5 on page 228
for further details. Users requiring a Python interface for ITK classes may refer to SimpleITK
(http://www.simpleitk.org/).
Additionally, SimpleITK offers interpreted language bindings for Java, C#, R, Tcl, and Ruby.
C.3.3 Constants
ITK does not define constants with #define in header files, hence do not declare constants using
#define CONST_VALUE_NAME 3
Use instead
constexpr unsigned int ConstValueName = 3;
or
const typename OperatorType::ConstIterator opEnd = op.End();
Add the const qualifier to arguments which set the pipeline inputs and state functions, e.g.
/** Set the marker image */
void SetMaskImage(const MaskImageType *input)
{
// Process object is not const-correct so the const casting is required.
this->SetNthInput( 1, const_cast< TMaskImage * >( input ) );
}
C.3.4 Generic Programming and the STL
Compile-time binding using methods of generic programming and template instantiation is the preferred implementation style. This approach has demonstrated its ability to create efficient, flexible
code. Use of the STL (Standard Template Library) is encouraged. STL is typically used by a class,
rather than as serving as a base class for derivation of ITK classes. Other STL influences are iterators
and traits. ITK defines a large set of iterators; however, the ITK iterator style differs in many cases
from STL because STL iterators follow a linear traversal model; ITK iterators are often designed for
2D, 3D, and even n-D traversal (see Section 6 on page 149 for further details on iterators).
C.3. System Overview & Philosophy
325
Traits are used heavily by ITK. ITK naming conventions supersede STL naming conventions; this
difference is useful in that it indicates to the community member something of a boundary between
ITK and STL.
C.3.5 Portability
ITK is designed to build and is systematically tested on a set of target operating system/compiler
combinations and results are reported continuously to the dashboards using CDash. These combinations include Linux, macOS, and Windows operating systems, and various versions of compilers
for each. This ensures that the code complies with the particular requirements on each of these
environments. See Section 10.2 on page 244 for further details.
When sufficient demand or need is detected, ITK maintainers add machines to the dashboard. Note
that since the dashboard is open to submissions from remote locations, other user configurations can
be tested dynamically. For a detailed and updated view, visit the ITK dashboard.
Since some of these compilers do not support all C++ features, the ITK community has had to back
off some important C++ features (such as partial specialization) because of limitations in compilers
(e.g., MSVC 6.0).
ITK’s open source philosophy, as well as its design, and heavy use of templates has made it possible
to improve the support of many of these compilers over time. Indeed, ITK has been incorporated to
the Microsoft Visual Studio and Intel C++ Compiler (ICC) build validation suites as of April 2017.
This means that ITK is being used by these teams in their benchmarks and validation cycles before
a version of their compiler is released to the market.
C.3.6 Multi-Layer Architecture
ITK is designed with a multi-layer architecture in mind. That is, three layers: a templated layer,
a run-time layer, and an application layer. The templated (or generic) layer is written in C++ and
requires significant programming skills and domain knowledge. The run-time layer is generated
automatically using the Swig-based wrapping system to produce language bindings to Python. The
interpreted layer is easier to use than the templated layer, and can be used for prototyping and
smaller-sized application development. Finally, the application layer is not directly addressed by
ITK other than providing simple examples of applications.
C.3.7 CMake Build Environment
The ITK build environment is CMake. CMake is an open-source, advanced cross-platform build
system that enables community members to write simple makefiles (named CMakeLists.txt) that
are processed to generated native build tools for a particular operating system/compiler combination.
See the CMake web pages at http://www.cmake.org for more information.
326
Appendix C. Coding Style Guide
See Section 2.2 on page 12 for specifics about the use of CMake in ITK.
Section C.26 on page 390 provides a reference to the recommended style of makefiles in ITK.
C.3.8 Doxygen Documentation System
The Doxygen open-source system is used to generate on-line documentation. Doxygen requires the
embedding of simple comments in the code which is in turn extracted and formatted into documentation.
Note that ITK prefers the backslash (“
”) style versus the at-sign (“@”) style to write the documentation commands.
For more information about Doxygen, please visit http://www.stack.nl/ dimitri/doxygen/
C.3.9 vnl Math Library
ITK has adopted the vnl – visual numerics library. Vnl is a portion of the vxl image understanding
environment. See http://vxl.sourceforge.net/ for more information about vxl and vnl.
C.3.10 Reference Counting
ITK has adopted reference counting via so-called itk::SmartPointer to manage object references. While alternative approaches such as automatic garbage collection were considered, their
overhead due to memory requirements, performance, and lack of control as to when to delete memory, precluded these methods. SmartPointers manage the reference to objects, automatically incrementing and deleting an instance’s reference count, deleting the object when the count goes to
zero.
An important note about SmarPointers refers to their destruction: the Delete() method on an ITK
smart pointer must never be called directly; if a SmartPointer object itkSmartPtr needs to be
deleted:
itkSmartPtr = nullptr;
must be done instead. The ITK smart pointer will determine whether the object should be destroyed.
See Section 3.2.4 on page 30 for further details.
C.4. Copyright
327
C.4 Copyright
ITK has adopted a standard copyright. This copyright should be placed at the head of every source
code file. The current copyright header and license reads as follows:
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
See Chapter A.1 for further details on the ITK license.
C.5 Citations
Give credit to others’ work. If when writing some piece of code (whether it is a class, group of
classes or algorithm in a method) the theory, framework or implementation are based on some scientific work, cite the work. In general, a citation to the peer-reviewed scientific publication, including its Digital Object Identifier (DOI), is preferred. This helps avoiding issues with web links. In
absence of such a reference, it is recommended that the link to the URL is written as it is (i.e. even
if the maximum line width is exceeded). Do not use URL shortening services.
When documenting a class header, if citations are required, use the Doxygen
par References command to list the references in a separate and clearly visible paragraph.
For instance,
namespace itk
{
/** \class DiffusionTensor3DReconstructionImageFilter
* \brief This class takes as input one or more reference image (acquired in the
* absence of diffusion sensitizing gradients) and 'n' diffusion
* weighted images and their gradient directions and computes an image of
* tensors. (with DiffusionTensor3D as the pixel type). Once that is done, you
* can apply filters on this tensor image to compute FA, ADC, RGB weighted
328
Appendix C. Coding Style Guide
* maps, etc.
*
* ...
*
* \par References
* \li[1]
*
* Carl-Fredrik Westin, Stephan E. Maier, Hatsuho Mamata, Arya Nabavi, Ferenc
* Andras Jolesz, and Ron Kikinis. "Processing and visualization for Diffusion
* tensor MRI. Medical Image Analysis, 6(2):93-108, 2002
* \li[2]
*
* Carl-Fredrik Westin, and Stephan E. Maier. A Dual Tensor Basis Solution to the
* Stejskal-Tanner Equations for DT-MRI. Proceedings of the 10th International
* Society of Magnetic Resonance In Medicine (ISMRM) Scientific Meeting \&
* Exhibition, Honolulu (HW, USA), 2002.
*
* ...
*
* \sa DiffusionTensor3D SymmetricSecondRankTensor
* \ingroup MultiThreaded TensorObjects
* \ingroup ITKDiffusionTensorImage
*/
template< typename TReferenceImagePixelType,
typename TGradientImagePixelType = TReferenceImagePixelType,
typename TTensorPixelType = double,
typename TMaskImageType = Image >
class ITK_TEMPLATE_EXPORT DiffusionTensor3DReconstructionImageFilter:
public ImageToImageFilter< Image< TReferenceImagePixelType, 3 >,
Image< DiffusionTensor3D< TTensorPixelType >, 3 > >
{
...
};
} // end namespace itk
The recommended bibliography style for citations is the LATEXplain style.
Or in a method body,
template< unsigned int VDimension >
void Solver< VDimension >
::ApplyBC( int dimension, unsigned int matrix )
{
...
// Store the appropriate value in bc correction vector (-K12*u2)
//
// See
// http://titan.colorado.edu/courses.d/IFEM.d/IFEM.Ch04.d/IFEM.Ch04.pdf
// chapter 4.1.3 (Matrix Forms of DBC Application Methods) for more
C.6. Naming Conventions
329
// info.
m_LinearSystem->AddVectorValue(*cc, -d * fixedvalue, 1);
}
...
}
C.6 Naming Conventions
In general, names are constructed by using case change to indicate separate words, as in TimeStamp.
Other general rules that must be followed in naming ITK constructs are:
• Underscores are not used (with the sole exception of enums and member variables).
• Variable names are chosen carefully with the intention to convey the meaning behind the code.
• Names are generally spelled out; use of abbreviations is discouraged. While this does result
in long names, it self-documents the code (e.g. use Dimension, point, size, or vector,
instead of D, pt, sz, or vec, respectively). Abbreviations are allowable when in common use,
and should be in uppercase as in RGB, or ID for “identifier”.)
The above general conventions must be followed in all cases. Depending on whether the name is a
• class
• file
• variable
• other name
variations on this theme result as explained in the following subsections.
C.6.1 ITK
The acronym for the NLM Insight Segmentation and Registration Toolkit must always be written in
capitals, i.e. ITK, when referring to it, e.g. in class documentation.
C.6.2 Naming Namespaces
Namespaces must be written in lowercase, e.g.
330
Appendix C. Coding Style Guide
namespace itk
{
...
} // end namespace itk
C.6.3 Naming Classes
Classes are:
• Named beginning with a capital letter.
• Placed in the appropriate namespace, typically itk:: (see Section C.7 on page 343).
• Named according to the following general rule:
class name =
In this formula, the name of the algorithm or process (possibly with an associated adjective
or adverb) comes first, followed by an input type (if the class is a filter), and completed by a
concept name.
A concept is an informal classification describing what a class does. There are many concepts in
ITK, the more common or important being:
• Accessor: Access and convert between types.
• Adaptor: Provide access to a portion of a complex pixel type.
• Boundary: The boundary of a cell.
• Calculator: Compute information.
• Classifier: Classify a pixel.
• Container: A container of objects such as points or cells.
• Estimator: Estimate a value or condition.
• Factory: Object factories are used to create instances.
• Filter: A class that participates in the data processing pipeline. Filters typically take one or
more inputs and produce one or more outputs.
• Function: Evaluate a function at a given position.
C.6. Naming Conventions
331
• Identifier: A unique ID for accessing points, cells, or other entities.
• Interface: Classes that specify an abstract interface.
• Interpolator: Interpolate data values, for example at non-pixel values.
• Iterator: Traverse data in various ways (e.g., forward, backward, within a region, etc.)
• Mapper: Transform data from one form into another.
• Metric: Compute similarity between two objects.
• Operator: A class that applies a user-specified function to a region.
• Optimizer: A class that performs numerical optimization.
• Pointer: A itk::SmartPointer to an instance of a class. Almost all instances in ITK are
referred to via SmartPointers.
• Reader: A class that reads a single data object (e.g., image or mesh).
• Reference: A type that refers to another object.
• Region: A subset of a data object, such as an image region.
• Source: A filter that initiates the data processing pipeline such as a reader or a procedural data
generator.
• Threader: A class that manages multi-threading.
• Traits: A collection of template parameters used to control the instantiation of other classes.
• Transform: Various types of transformations including affine, procedural, and so on.
• Writer: A filter that terminates the data processing pipeline by writing data to disk or to a
communications port.
The naming of classes is an art form; please review existing names to catch the spirit of the naming
convention.
Conventions adopted in ITK for naming classes include:
• The “To” convention (such as in itk::ImageToImageFilter) is generally used for base
classes, and when a filter converts from one data type to another. Derived classes do not
continue the “To” convention. Classes like itk::HistogramToTextureFeaturesFilter,
or itk::ImageToHistogram do not produce an image as outputs, but change the data type
or produce a set of features. The expectation of an ITK filter name is that it maintains the
same type (even when changing the dimensionality as when changing from an itk::Image
to a itk::VectorImage) unless it has the “To” naming conventions.
332
Appendix C. Coding Style Guide
• Adding the Base appendix to a base class name is generally discouraged.
Example names include:
• ShrinkImageFilter
• TriangleCell
• ScalarImageRegionIterator
• NeighborhoodIterator
• MapContainer
• DefaultImageTraits
• BackwardDifferenceOperator
C.6.4 Naming Files
Files should have the same name as the class, with an “itk” prepended.
Header files are named .h, while implementation files are named either .cxx or .hxx, depending
on whether they are implementations of templated classes.
For example, the class itk::Image
• is declared in the file itkImage.h and
• is defined in the file itkImage.hxx (because itk::Image is templated).
The class itk::Object
• is declared in the file itkObject.h and
• is defined in the file itkObject.cxx.
Naming Tests
Following the TestDriver philosophy, see Section 9.4 on page 225, test files must be named with
the same name used to name the main method contained in the test file (.cxx). This name should
generally be indicative of the class tested, e.g.
int itkTobogganImageFilterTest( int argc, char *argv[] )
C.6. Naming Conventions
333
for a test that checks the itk::TobogganImageFilter class, and contained in the test file named
itkTobogganImageFilterTest.cxx.
Note that all test files should start with the lowercase itk prefix. Hence, the main method name
in a test is the sole exception to the method naming convention of starting all method names with
capitals (see C.6.6).
A test’s input argument number should always be named argc, and the input arguments argv for
the sake of consistency.
If due to some constraint (e.g. nature of input images, number of input images, dimensionality) a
class has multiple test files with minor changes in its content, the test files should be named following
the convention
test filename =
In this formula, the filename comes first, and is completed by the variation tested, conveying the
meaning behind the test, e.g.
itkSimpleImageRegistrationTest.cxx
itkSimpleImageRegistrationTestWithMaskAndSampling.cxx
When the same test file is used by multiple tests in the corresponding CMakeLists.txt, for example,
with different parameters, these different tests should be named following the convention
test name =
In this formula, the filename comes first, and is completed by the variation tested, conveying the
meaning behind the test, e.g.
itk_add_test(NAME itkHConcaveImageFilterTestFullyConnectedOff
COMMAND ITKMathematicalMorphologyTestDriver
--compare-MD5
${ITK_TEST_OUTPUT_DIR}/itkHConcaveImageFilterTestFullyConnectedOff.png
bd1b5ab47f54cd97b5c6b454bee130e2
itkHConcaveImageFilterTest DATA{${ITK_DATA_ROOT}/Input/Input-RA-Short.nrrd}
${ITK_TEST_OUTPUT_DIR}/itkHConcaveImageFilterTestFullyConnectedOff.png 2000
0)
itk_add_test(NAME itkHConcaveImageFilterTestFullyConnectedOn
COMMAND ITKMathematicalMorphologyTestDriver
--compare-MD5
${ITK_TEST_OUTPUT_DIR}/itkHConcaveImageFilterTestFullyConnectedOn.png
c7116406ded975955965226f6a69e28d
itkHConcaveImageFilterTest DATA{${ITK_DATA_ROOT}/Input/Input-RA-Short.nrrd}
${ITK_TEST_OUTPUT_DIR}/itkHConcaveImageFilterTestFullyConnectedOn.png 2000 1)
If the test checks features that span multiple classes or other general features, the filename should
adhere to the general convention of conveying the meaning behind the code, e.g.
334
Appendix C. Coding Style Guide
int itkSingleLevelSetWhitakerImage2DWithCurvatureTest( int argc, char* argv[] )
means that the itk::WhitakerSparseLevelSetImage class is tested on two dimensions, using
the itk::LevelSetEquationCurvatureTerm class to represent the curvature term in the level-set
evolution PDE.
However, in these last cases, readability of the test name is important, and too long test names are
discouraged.
See Section 9.4 on page 225 for further details about the ITK testing framework.
C.6.5 Examples
C.6.6 Naming Methods and Functions
Global functions and class methods, either static or class members, are named beginning with
a capital letter. The biggest challenge when naming methods and functions is to be consistent with existing names. For example, given the choice between ComputeBoundingBox() and
CalculateBoundingBox(), the choice is ComputeBoundingBox() because “Compute” is used
elsewhere in the system in similar settings (i.e. the concepts described in Section C.6.3 should
be used whenever possible).
Note that in the above example CalcBoundingBox() is not allowed because it is not spelled out.
Method argument names should be included in their declaration.
When declaring class methods, it is generally recommended to follow a logical order, not alphabetical, and group them by blocks separated by empty lines for the sake of readability.
The definition of the methods should follow this same order.
C.6.7 Naming Class Data Members
Class data members are prepended with m as in m Size. This clearly indicates the origin of data
members, and differentiates them from all other variables. Furthermore, it is a key requirement for
the correct application of the ITK macros (such as the Get##name and Set##name methods).
RadiusType m_Radius;
When declaring class data members, it is generally recommended to follow a logical order, not
alphabetical, and group them by blocks separated by empty lines for the sake of readability.
C.6. Naming Conventions
335
C.6.8 Naming Enums
Enumeration list (enum) must declare its identifier before the enum-list is specified, it must start
with capitals and be written with case change, it should generally add the Type appendix (e.g.
MyEnumType), and the enum-list must be specified in capitals. The documentation or comments
added for each enum-list entry, if necessary, need not to be aligned.
/** Weight types. */
enum WeightType {
GOURAUD, // Uniform weights
THURMER, // Angle on a triangle at the given vertex
AREA // Doc alignment not needed
};
No typedef keyword shall be added to an enum.
Enum-lists do not need to have their corresponding integral value specified (i.e. GOURAUD = 0,).
When calling an enum-list entry from within a class, it should be called with the Self:: class alias.
while( !m_OperationQ.empty() )
{
switch( m_OperationQ.front() )
{
case Self::SET_PRIORITY_LEVEL:
m_PriorityLevel = m_LevelQ.front();
m_LevelQ.pop();
break;
case Self::SET_LEVEL_FOR_FLUSHING:
m_LevelForFlushing = m_LevelQ.front();
m_LevelQ.pop();
break;
...
}
}
C.6.9 Naming Local Variables
Local variables begin in lowercase. There is more flexibility in the naming of local variables, but
they should adhere to the general convention of conveying the meaning behind the code.
Please remember that others will review, maintain, fix, study and extend your code. Any bread
crumbs that you can drop in the way of explanatory variable names and comments will go a long
way towards helping other community members.
336
Appendix C. Coding Style Guide
Temporary Variable Naming
Every effort should be made to properly name temporary variables of any type that may be used in
a reduced part of a method, such as in
...
// Resize the schedules
ScheduleType schedule( m_NumberOfLevels, ImageDimension );
schedule.Fill( 0 );
m_Schedule = schedule;
...
For such temporary variables whose naming would be overly wordy to express their meaning or may
be misleading, or may be re-used at multiple stages within a method (e.g. using the name output
for intermediate results), the name tmp can be used.
...
ValueType dimension = static_cast< ValueType >( ImageDimension );
NormalVectorFilterType normalVectorFilter = NormalVectorFilterType::New();
...
normalVectorFilter->SetIsoLevelLow( -m_CurvatureBandWidth - dimension );
normalVectorFilter->SetIsoLevelHigh( m_CurvatureBandWidth + dimension );
...
// Move the pixel container and image information of the image we are working
// on into a temporary image to use as the input to the mini-pipeline. This
// avoids a complete copy of the image.
typename OutputImageType::Pointer output = this->GetOutput();
typename OutputImageType::Pointer tmp = OutputImageType::New();
tmp->SetRequestedRegion( output->GetRequestedRegion() );
tmp->SetBufferedRegion( output->GetBufferedRegion() );
tmp->SetLargestPossibleRegion( output->GetLargestPossibleRegion() );
tmp->SetPixelContainer( output->GetPixelContainer() );
tmp->CopyInformation( output );
typename SparseImageType::Pointer sparseNormalImage =
normalVectorFilter->GetOutput();
this->ComputeCurvatureTarget( tmp, sparseNormalImage );
m_LevelSetFunction->SetSparseTargetImage( sparseNormalImage );
Variable Initialization
A basic type variable declared and not being assigned immediately within a method should be initialized to its zero value.
Note the weight variable in the following example:
template< typename TInputImage >
double
C.6. Naming Conventions
337
WarpHarmonicEnergyCalculator< TInputImage >
::EvaluateAtNeighborhood( ConstNeighborhoodIteratorType & it ) const
{
vnl_matrix_fixed< double, ImageDimension, VectorDimension > J;
PixelType next, prev;
double weight = 0;
for( unsigned int i = 0; i < ImageDimension; ++i )
{
next = it.GetNext(i);
prev = it.GetPrevious(i);
weight = 0.5 * m_DerivativeWeights[i];
for( unsigned int j = 0; j < VectorDimension; ++j )
{
J[i][j] = weight * ( static_cast< double >( next[j] )
- static_cast< double >( prev[j] ) );
}
}
const double norm = J.fro_norm();
return norm * norm;
}
Take into account that many ITK variables, such as itk::ImageRegion class instances initialize
themselves to zero, so they do not need to be initialized unless required. The following declaration
would create a matrix with all zero by default:
// Define the dimension of the images
constexpr unsigned int ImageDimension = 2;
...
// Declare the type of the size
using SizeType = itk::Size< ImageDimension >;
SizeType size;
size[0] = 100;
size[1] = 100;
// Declare the type of the index to access images
using IndexType = itk::Index< ImageDimension >;
IndexType start;
start[0] = 0;
start[1] = 0;
// Declare the type of the Region
using RegionType = itk::ImageRegion< ImageDimension >;
338
Appendix C. Coding Style Guide
RegionType region;
region.SetIndex( start );
region.SetSize( size );
Control Statement Variable Naming
Control statement variables names should be clear and concise. For simple counters over arrays,
lists, maps elements or matrices, the i, j, k order is preferred, i.e. when requiring to walking over
multiple dimensions, over, for example ii.
If more than three nested control statements are required, there is probably a better design that can
be implemented.
For iterators, inIt and outIt are recommended if both input and output structures are involved.
Otherwise, it can be used.
Variable Scope
Control statement variables should have a local scope. Hence, instead of declaring a method-scope
variable and re-using it,
unsigned int i;
for ( i = 0; i < ImageDimension; ++i )
{
Something();
}
...
for ( i = 0; i < ImageDimension; ++i )
{
SomethingElse();
}
it is recommended to limit the scope of the variables to the control statements in which they are
required:
unsigned int i = 0;
for ( unsigned int i = 0; i < ImageDimension; ++i )
{
Something();
}
...
for ( unsigned int i = 0; i < ImageDimension; ++i )
{
SomethingElse();
}
C.6. Naming Conventions
339
C.6.10 Naming Template Parameters
Template parameters follow the usual rules with naming except that they should start with either the
capital letter “T” or “V”. Type parameters (such as the pixel type) begin with the letter “T” while
value template parameters (such as the dimensionality) begin with the letter “V”.
template< typename TPixel, unsigned int VImageDimension = 2 >
class ITK_TEMPLATE_EXPORT Image:public ImageBase< VImageDimension >
For template parameters the use of typename is preferred over class.
Very early
C++ compilers did not have a typename keyword, and class was purposed for declaring template parameters.
It was later discovered that this lead to ambiguity in some
valid code constructs, and the typename key word was added.
It is often agreed
(http://blogs.msdn.com/b/slippman/archive/2004/08/11/212768.aspx) that typename is marginally
more expressive in its intent and ITK should consistently use typename instead of class.
C.6.11 Naming Typedefs
Type aliases are absolutely essential in generic programming. They significantly improve the readability of code, and facilitate the declaration of complex syntactic combinations. Unfortunately, creation of type aliases is tantamount to creating another programming language. Hence type aliases
must be used in a consistent fashion. The general rule for type alias names is that they end in the
word “Type”. For example,
using PixelType = TPixel;
However, there are many exceptions to this rule that recognize that ITK has several important concepts that are expressed partially in the names used to implement the concept. An iterator is a
concept, as is a container or pointer. These concepts are used in preference to Type at the end of a
type alias as appropriate. For example,
using PixelContainer = typename ImageTraits::PixelContainer;
Here “Container” is a concept used in place of “Type”. ITK currently identifies the following concepts used when naming type aliases:
• Self as in
using Self = Image;
All classes should define this type alias.
340
Appendix C. Coding Style Guide
• Superclass as in
using Superclass = ImageBase< VImageDimension >;
All classes should define the Superclass type alias.
• Pointer as in a smart pointer to an object as in
using Pointer = SmartPointer< Self >;
All classes should define the Pointer type alias.
• Container is a type of container class.
• Iterator an iterator over some container class.
• Identifier or id such as a point or cell identifier.
C.6.12 Naming Constants
Constants must start with capital letters, e.g.
constexpr unsigned int CodeAxisField = 14;
C.6.13 Using Operators to Pointers
The indirection unary operator (*) must be placed next to the variable, e.g.
int itkTransformFileReaderTest( int argc, char *argv[] )
or
const InputImageType *inputPtr = this->GetInput();
The reference or address unary operator (&) must be placed next to the variable, e.g.
const typename FixedImageType::RegionType &fixedRegion =
m_FixedImage->GetLargestPossibleRegion();
or
C.6. Naming Conventions
341
::PrintSelf( std::ostream &os, Indent indent ) const
C.6.14 Using Operators to Arrays
The subscript operator ([]) must be placed next to the variable, e.g.
int itkGaborKernelFunctionTest( int argc, char *argv[] )
or
unsigned int GetSplitInternal( unsigned int dim,
unsigned int i, unsigned int numberOfPieces, IndexValueType regionIndex[],
SizeValueType regionSize[] ) const override;
C.6.15 Using Underscores
Do not use undersocres. The only exception is when defining preprocessor variables and macros
(which are discouraged). In this case, underscores are allowed to separate words.
C.6.16 Include Guards
An include guard’s case must mimic the one used for a file, with the file extension separated by an
undersore
#ifndef itkImage_h
#define itkImage_h
// Class declaration code
#endif
and
#ifndef itkImage_hxx
#define itkImage_hxx
// Template class implementation code
#endif
Note that include guards in implementation files are to be used only for templated classes.
342
Appendix C. Coding Style Guide
C.6.17 Preprocessor Directives
Some of the worst code contains many preprocessor directives and macros such as
#if defined(__APPLE__) && (__clang_major__ == 3)
&& (__clang_minor__ == 0) && defined(NDEBUG) && defined(__x86_64__)
cc = -1.0 * itk::Math::sqr(1.0 / (cc + itk::Math::eps) );
#else
cc = -1.0 * itk::Math::sqr(1.0 / cc);
#endif
Do not use them except in a very limited sense (to support minor differences in compilers or operating systems). If a class makes extensive use of preprocessor directives, it is a candidate for
separation into multiple sub-classes.
However, if such directives are to be used, they should start in column one, regardless of the required
indentation level of the code they contain.
C.6.18 Header Includes
Headers in ITK must be included using quotes (“ ”)
#include "itkImageRegion.h"
Only the required headers should be included. If an included header already includes a header for a
class also used in the current file, the header for that class should not be included.
Header includes are preferred over forward declarations. Forward declarations are only used to
prevent circular dependencies.
C.6.19 Const Correctness
As a general rule, the const type qualifier must be used for:
• Arguments which set the pipeline inputs, e.g.
/** Set the marker image. */
void SetMaskImage( const MaskImageType *input )
{
// Process object is not const-correct so the const casting is required.
this->SetNthInput( 1, const_cast< TMaskImage * >( input ) );
}
/** Set the input image. */
void SetInput1( const InputImageType *input )
C.7. Namespaces
343
{
this->SetInput( input );
}
/** Set the marker image. */
void SetInput2( const MaskImageType *input )
{
this->SetMaskImage( input );
}
• Accessor/state functions, e.g.
bool GetUseVectorBasedAlgorithm() const
{
return HistogramType::UseVectorBasedAlgorithm();
}
C.7 Namespaces
All classes should be placed in the itk:: namespace. Additional sub-namespaces are being designed to support special functionality, e.g.
namespace itk
{
namespace fem
{
...
} // end namespace fem
} // end namespace itk
Please see current documentation to determine if there is a sub-namespace relevant to a specific
situation. Normally sub-namespaces are used for helper ITK classes.
Code should not use using namespace. This is to avoid namespace conflicts, but, more importantly,
to improve readability.
When declaring or defining members of the itk:: namespace, for example, the itk:: namespace
prefix should not be added. That is, code within namespace itk {... }” should not use itk::.
The :: global namespace should be used when referring to a global function, e.g.
// Execute the filter
clock_t start = ::clock();
m_Filter->UpdateLargestPossibleRegion();
clock_t stop = ::clock();
344
Appendix C. Coding Style Guide
It helps clarifying exactly which method is exactly being invoked and where it originates.
Note that itk:: should only be used outside the itk:: namespace.
C.8 Aliasing Template Parameter Typenames
The public class typename’s should be limited to the types that are required to be available by other
classes. The typename’s can clutter a class API and can restrict future refactoring that changes the
types when unnecessary.
For instance,
template< typename TPixel, unsigned int VImageDimension = 2 >
class ITK_TEMPLATE_EXPORT Image : public ImageBase< VImageDimension >
{
public:
/** Standard class type alias. */
using Self = Image;
using Superclass = ImageBase< VImageDimension >;
using Pointer = SmartPointer< Self >;
using ConstPointer = SmartPointer< const Self >;
using ConstWeakPointer = WeakPointer< const Self >;
...
/** Pixel type alias support. Used to declare pixel type in filters
* or other operations. */
using PixelType = TPixel;
...
or
template< typename TImage >
class ITK_TEMPLATE_EXPORT ImageRegionIterator : public ImageRegionConstIterator< TImage >
{
public:
/** Standard class type alias. */
using Self = ImageRegionIterator;
using Superclass = ImageRegionConstIterator< TImage >;
/** Types inherited from the Superclass */
using IndexType = typename Superclass::IndexType;
using SizeType = typename Superclass::SizeType;
...
C.9. Pipelines
345
C.9 Pipelines
The following is a set of useful tips that must be taken into account when developing ITK code:
• Do call Update() before using the pipeline output.
• Do call UpdateLargestPossibleRegion() when reusing a reader.
When reusing a reader you must call:
reader->UpdateLargestPossibleRegion()
instead of the usual:
reader->Update()
Otherwise the extent of the previous image is kept, and in some cases lead to Exceptions
being thrown if the second image is smaller than the first one.
• Do not assume inputImage->SetRequestedRegion( smallRegion ) will make the filter
faster! The filter might run on the entire input image regardless. To make it run on a smaller
block, get a new itk::RegionOfInterestImageFilter, say ROIfilter, and do:
ROIfilter->SetInput( inputImage );
ROIfilter->SetRegionOfInterest( smallRegion );
CCfilter->SetInput( ROIfilter->GetOutput() );
• On a newly-manually-created image, do initialize the pixel values if you expect them to be so!
ITK does not initialize the image buffer when you call Allocate(). It is your responsibility
to initialize the pixel values, either by calling Allocate( true ) or filling the image buffer
as in:
image->FillBuffer( 0 ); // Initialize it to all dark.
C.10 The auto Keyword
Available since C++11, the auto keyword specifies that a variable’s type is automatically deduced
from its initializer.
The auto keyword should be used to specify a type in the following cases:
• The type is duplicated on the left side of an assigment when it is mandated on the right side,
e.g. when there is an explicit cast or initializing with new or ITK’s ::New().
• When obtaining container elements, when the element type is obvious from the type of the
container.
346
Appendix C. Coding Style Guide
• When the type does not matter because it is not being used for anything other than equality
comparison.
• When declaring iterators in a for loop.
• When a trailing return type is used in a function declaration.
• When creating lambda functions.
All other cases should not use auto, but a semantic type name should be used that conveys meaning,
as described in Section C.6 and Section C.6.11 on page 339.
Application or example code that uses ITK, as opposed to the toolkit itself, may use auto more
liberally.
C.11 Initialization and Assignment
All member variables must be initialized in the class constructor. For such purpose, initialization
lists are preferred
template< typename TInputImage, typename TOutputImage >
SpecializedFilter< TInputImage, TOutputImage >
::SpecializedFilter() :
m_ForegroundValue( NumericTraits::max() ),
m_BackgroundValue( NumericTraits::ZeroValue() ),
m_NumPixelComponents( 0 ),
m_NoiseSigmaIsSet( false ),
m_SearchSpaceList( ListAdaptorType::New() )
{
// By default, turn off automatic kernel bandwidth sigma estimation
this->KernelBandwidthEstimationOff();
}
over assignment:
template< typename TInputImage, typename TOutputImage >
SpecializedFilter< TInputImage, TOutputImage >
::SpecializedFilter()
{
m_ForegroundValue = NumericTraits::max();
m_BackgroundValue = NumericTraits::ZeroValue();
m_NumPixelComponents = 0;
m_UseSmoothDiscPatchWeights = false;
m_SearchSpaceList = ListAdaptorType::New();
// By default, turn off automatic kernel bandwidth sigma estimation
C.12. Accessing Members
347
this->KernelBandwidthEstimationOff();
}
Nevertheless, there may be some exceptions to the initialization list rule. In some situations where
it can be foreseen that the corresponding Set##name or ##nameOn/##nameOff may be
• overloaded by some classes in the future, or
• deprecated, and a warning thrown when it is called to help migration,
initialization through the corresponding Set##name or ##nameOn/##nameOff method is recommended instead of directly manipulating the data member.
Smart pointers need not to be initialized, since they initialize themselves to the null pointer, so they
are the sole exception to the above rule.
Note that all numeric data members must be initialized using the appropriate ITK’s NumericTraits
static method.
C.12 Accessing Members
The C++ keyword this must be used when calling a class’ own methods:
template< typename TInputImage, typename TOutputImage >
void
ExpandImageFilter< TInputImage, TOutputImage >
::GenerateInputRequestedRegion()
{
// Call the superclass' implementation of this method
Superclass::GenerateInputRequestedRegion();
// Get pointers to the input and output
InputImageType * inputPtr =
const_cast< InputImageType * >( this->GetInput() );
const OutputImageType * outputPtr = this->GetOutput();
...
}
The use of the explicit this-> pointer helps clarifying which method is exactly being invoked and
where it originates.
The value of a member variables or data within a class must be retrieved calling the variable name
directly, i.e. the use of its getter method (i.e. GetMyVariable()) is discouraged for such purpose.
Similarly, the use of the this keyword when calling self data members is discouraged, i.e.
348
Appendix C. Coding Style Guide
template< typename TInputImage, typename TOutputImage >
void
BinaryContourImageFilter< TInputImage, TOutputImage >
::PrintSelf( std::ostream & os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
os << indent << "FullyConnected: "
...
<< m_FullyConnected << std::endl;
}
is preferred over
template< typename TInputImage, typename TOutputImage >
void
BinaryContourImageFilter< TInputImage, TOutputImage >
::PrintSelf( std::ostream & os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
os << indent << "FullyConnected: "
...
<< this->m_FullyConnected << std::endl;
}
C.13 Code Layout and Indentation
The following are the accepted ITK code layout rules and indentation style. After reading this
section, you may wish to visit many of the source files found in ITK. This will help crystallize the
rules described here.
C.13.1 General Layout
• Each line of code should take no more than 200 characters.
• Break the code across multiple lines as necessary.
• Use lots of white space to separate logical blocks of code, intermixed with comments.
• To a large extent the structure of code directly expresses its implementation.
• The appropriate indentation level is two spaces for each level of indentation.
• Do not use tabs; set up your editor to insert spaces. Using tabs may look good in your editor
but will wreak havoc in others.
C.13. Code Layout and Indentation
349
• The declaration of variables within classes, methods, and functions should be one declaration
per line:
int i = 0;
int j = 0;
char* stringName;
A short code snippet in ITK might look like:
if( condition )
{
unsigned int numberOfIterations = 100;
filter->SetNumberOfIterations( numberOfIterations );
filter->Update();
filter->Print( std::cout );
}
The body of a method must always be indented, starting with an indentation of two white spaces,
and indenting the rest of the body as described in this section.
C.13.2 Class Layout
Classes are declared (.h) using the following guidelines:
• Begin with the Copyright notice.
• Follow with include guards (e.g #ifndef itkBoxImageFilter h)).
• Follow with the necessary includes. Include only what is necessary to avoid dependency
problems.
• Place the class in the correct namespace.
• public methods come first.
• protected methods follow.
• private members come last.
• public data members are forbidden.
• End the namespaces.
• Templated classes require a special preprocessor directive to control the manual instantiation
of templates. See the example below and look for ITK MANUAL INSTANTIATION.
• Close the include guards.
350
Appendix C. Coding Style Guide
The class layout looks something like this:
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
#ifndef itkImage_h
#define itkImage_h
#include
#include
#include
#include
"itkImageBase.h"
"itkPixelTraits.h"
"itkDefaultImageTraits.h"
"itkDefaultDataAccessor.h"
namespace itk
{
/** \class Image
* \brief Templated N-dimensional image class.
*
* Detailed documentation...
*/
template< typename TPixel, unsigned int VImageDimension=2,
typename TImageTraits=DefaultImageTraits< TPixel, VImageDimension > >
class Image: public ImageBase< VImageDimension >
{
public:
...
protected:
...
private:
...
};
} // end namespace itk
#ifndef ITK_MANUAL_INSTANTIATION
C.13. Code Layout and Indentation
351
#include "itkImage.hxx"
#endif
#endif // itkImage_h
Many of the guidelines for the class declaration file are applied to the class definition (.hxx, .cxx)
file:
• Begin with the Copyright notice.
• Follow with include guards
itkBoxImageFilter hxx)).
in
case
of
templated
classes
(e.g
#ifndef
• Follow with the necessary includes. Include only what is necessary to avoid dependency
problems.
• Place the class in the correct namespace.
• The constructor come first.
• The destructor follows.
• The PrintSelf member come last.
• End the namespaces.
• Close the include guards if present.
The class definition layout looks something like this:
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
/*=========================================================================
*
* Portions of this file are subject to the VTK Toolkit Version 3 copyright.
352
Appendix C. Coding Style Guide
*
* Copyright (c) Ken Martin, Will Schroeder, Bill Lorensen
*
* For complete copyright, license and disclaimer of warranty information
* please refer to the NOTICE file at the top of the ITK source tree.
*
*=========================================================================*/
#ifndef itkImage_hxx
#define itkImage_hxx
#include "itkImage.h"
#include "itkProcessObject.h"
#include
namespace itk
{
template< typename TPixel, unsigned int VImageDimension >
Image< TPixel, VImageDimension >
::Image()
{
m_Buffer = PixelContainer::New();
}
template< typename TPixel, unsigned int VImageDimension >
Image< TPixel, VImageDimension >
˜::Image()
{
}
...
template< typename TPixel, unsigned int VImageDimension >
void
Image< TPixel, VImageDimension >
::PrintSelf( std::ostream &os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
os << indent << "PixelContainer: " << std::endl;
m_Buffer->Print( os, indent.GetNextIndent() );
}
} // end namespace itk
#endif
Note that ITK headers are included first, and system or third party libraries follow.
C.13. Code Layout and Indentation
353
C.13.3 Method Definition
Methods are defined across multiple lines. This is to accommodate the extremely long definitions
possible when using templates. The starting and ending brace should be in column one, and the
following order must be followed:
• The first line is the template declaration.
• The second line is the method return type.
• The third line is the class qualifier.
• And the fourth line is the name of the method.
e.g.
template< typename TPixel, unsigned int VImageDimension, typename TImageTraits >
const double *
Image< TPixel, VImageDimension, TImageTraits >
::GetSpacing() const
{
...
}
The same rules apply for non-templated classes:
void
Bruker2DSEQImageIO
::PrintSelf(std::ostream & os, Indent indent) const
{
...
}
C.13.4 Use of Braces
Braces in Control Sequences
Braces must be used to delimit the scope of an if, for, while, switch, or other control structure.
for( unsigned int i = 0; i < ImageDimension; ++i )
{
...
}
or when using an if:
354
Appendix C. Coding Style Guide
if( condition )
{
...
}
else if( otherCondition )
{
...
}
else
{
...
}
In switch statement cases, the constant-expression statement bodies should not be enclosed with
braces:
switch( m_OperationQ.front() )
{
case Self::SET_PRIORITY_LEVEL:
m_PriorityLevel = m_LevelQ.front();
m_LevelQ.pop();
break;
case Self::SET_LEVEL_FOR_FLUSHING:
m_LevelForFlushing = m_LevelQ.front();
m_LevelQ.pop();
break;
...
default:
break;
}
In do-while statements the opening/closing braces must lie on a line of their own:
do
{
k += 1;
p *= rand->GetVariate();
}
while( p > L );
Braces in Arrays
When initializing an array, no space shall be left between the first value and the opening brace, and
the last argument and closing brace:
C.13. Code Layout and Indentation
355
// Define the image size in image coordinates, and origin and spacing in
// physical coordinates.
SizeType size = {{20, 20, 20}};
double origin[3] = {0.0, 0.0, 0.0};
double spacing[3] = {1, 1, 1};
C.13.5 Indentation and Tabs
The ITK style bans the use of tabs. Contributors should configure their editors to use white spaces
instead of tabs. The size of the indent in ITK is fixed to two white spaces.
template< typename TInputImage, typename TOutputImage = TInputImage >
class ITK_TEMPLATE_EXPORT SpecializedFilter :
public ImageToImageFilter< TInputImage, TOutputImage >
{
public:
using Self = SpecializedFilter;
using Superclass = ImageToImageFilter< TInputImage, TOutputImage >;
using Pointer = SmartPointer< Self >;
using ConstPointer = SmartPointer< const Self >;
/** Method for creation through the object factory. */
itkNewMacro( Self );
/** Run-time type information (and related methods) */
itkTypeMacro( SpecializedFilter, ImageToImageFilter );
...
};
or for the implementation of a given method:
template< typename TInputImage, typename TOutputImage >
void
SpecializedFilter< TInputImage, TOutputImage >
::GenerateData() const
{
// Allocate the outputs.
this->AllocateOutputs();
// Create a process accumulator for tracking the progress of this minipipeline.
ProgressAccumulator::Pointer progress = ProgressAccumulator::New();
progress->SetMiniPipelineFilter( this );
...
}
ITK uses the Whitesmiths indentation style, with the braces associated with a control statement on
356
Appendix C. Coding Style Guide
the next line, indented. Thus, source code in the body of the brackets must be aligned along with the
brackets.
while( x == y )
{
Something();
}
C.13.6 White Spaces
As a general rule, a single white space should be used to separate every word.
However, no white space shall be added between type names, keywords, and method names and the
following marks:
• An opening angle bracket (<) and the template keyword.
• An opening round bracket (() and its preceding word (e.g. in method declarations and definitions).
• An opening/closing brace ({/}) and its subsequent/preceding word (e.g. when initializing an
array).
• An opening or closing square bracket ([, ] ) and its contents (e.g. when specifying the index
of an array).
• The constant expression termination colon in a switch statement (e.g. case Self::SET PRIORITY LEVEL:).
• Semicolons (;) and their preceding word (e.g. end of a statement, etc.).
To the contrary, a single white space should be added between
• An opening angle bracket (<) and the subsequent template, variable or keyword.
• A closing angle bracket (>) and the preceding and subsequent words (such as in type aliases).
• An opening/closing round bracket ((/)) and its subsequent/preceding word (e.g. in method
argument lists).
• Individual members in a list separated by commas (e.g. SizeType size = 20, 20, 20, or
data[i, j]). The comma must be always placed next to a given element, and be followed
by the single white space.
• Operators(i.e. +, -, =, ==, +=, <<, etc. ) and the left-hand and right-hand arguments.
• The ternary operators (?, :) and the left-hand condition, and right-hand values.
C.13. Code Layout and Indentation
357
• Control statements (i.e. if, for, while, switch, etc.) and their conditional statements or
arguments.
• The different parts of a for control statement
for( unsigned int i = 0; i < ImageDimension; ++i )
{
...
}
• A method call parentheses and its content
this->SomeMethod( param, a+b );
Thus, for a class declaration we would write
template< typename TInputImage, typename TOutputImage = TInputImage >
class ITK_TEMPLATE_EXPORT SpecializedFilter :
public ImageToImageFilter< TInputImage, TOutputImage >
{
public:
using Self = SpecializedFilter;
using Superclass = ImageToImageFilter< TInputImage, TOutputImage >;
using Pointer = SmartPointer< Self >;
using ConstPointer = SmartPointer< const Self >;
/** Method for creation through the object factory. */
itkNewMacro( Self );
/** Run-time type information (and related methods) */
itkTypeMacro( SpecializedFilter, ImageToImageFilter );
...
};
And for a class constructor we would write
template< typename TInputImage, typename TOutputImage >
SpecializedFilter< TInputImage, TOutputImage >
::SpecializedFilter() :
m_ForegroundValue( NumericTraits::max() ),
m_BackgroundValue( NumericTraits::ZeroValue() ),
m_NumPixelComponents( 0 ),
m_NoiseSigmaIsSet( false ),
m_SearchSpaceList( ListAdaptorType::New() )
{
// By default, turn off automatic kernel bandwidth sigma estimation
this->KernelBandwidthEstimationOff();
}
Trailing white spaces are not allowed in ITK.
358
Appendix C. Coding Style Guide
C.13.7 Grouping
Unnecessary parentheses for grouping hinder reading the code. The C++ precedence and associativity (the order in which the operands are evaluated) of operators must be taken into account to avoid
using unnecessary parentheses.
As a general principle, these apply to condition expressions, and statements where mathematical,
logical or bitwise operations are performed.
Conditional Expressions
In conditional expressions contained in control statements (e.g. if, for, while, etc.) composed by
multiple operands (e.g. joined using the logical operators), assignments and other constructs where
such expressions are involved, the use of excessive parentheses is discouraged.
For example, the style below:
if( modelFile == "tri3-q.meta" && ( s == 2 || s == 1 ) )
{
...
}
else
{
...
}
is recommended over:
if( ( modelFile == "tri3-q.meta" ) && ( ( s == 2 ) || ( s == 1 ) ) )
{
...
}
else
{
...
}
Assignments
In assignments, the operator precedence and associativity rules apply to help keeping the code readable and void of operators in-excess. In assignments that do not involve long expressions that would
otherwise be hard and time-consuming to interpret, the use of parentheses should be avoided.
For example, in
C.13. Code Layout and Indentation
359
sum[dim] += ( component * weight );
grouping is not necessary, as only a single operator exists in the right-hand operand. Hence, instead
of writing the above code, community members should rather write:
sum[dim] += component * weight;
Return Statements
In return statements, using parentheses should be avoided when they are not strictly necessary for
the evaluation of the returned term. For example, when returning variables or method calls, as in:
OutputType Evaluate(const PointType & point) const override
{
ContinuousIndexType index;
this->GetInputImage()->TransformPhysicalPointToContinuousIndex( point,
index);
// No thread info passed in, so call method that doesn't need thread
// identifier.
return this->EvaluateDerivativeAtContinuousIndex( index );
}
The same principle applies when returning the result of an algebraic, logical or bitwise operation
that does not require using parentheses to specify evaluation preference, such as in:
template< typename TPoint >
double
SimpleSignedDistance( const TPoint & p )
{
...
return accum - radius;
}
instead of writing return ( accum - radius );.
Or in:
bool
RealTimeStamp::operator>( const Self & other ) const
{
if( this->m_Seconds > other.m_Seconds )
{
return true;
}
360
Appendix C. Coding Style Guide
if( this->m_Seconds < other.m_Seconds )
{
return false;
}
return this->m_MicroSeconds > other.m_MicroSeconds;
}
Or in:
IntegerType mixBits(const IntegerType & u, const IntegerType & v) const
{
return hiBit(u) | loBits(v);
}
C.13.8 Alignment
In every ITK file, the following code parts always start in column one:
• Copyright notice.
• Namespace opening/closing braces.
• Include guards.
The following specific parts of a class declaration always start in column one:
• Class documentation.
• Template declaration.
• Class declaration, including its opening/closing braces.
• Manual instantiation preprocessor directives.
• Access modifiers.
For instance,
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
C.13. Code Layout and Indentation
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
#ifndef itkImage_h
#define itkImage_h
#include "itkImageBase.h"
namespace itk
{
/** \class Image
* \brief Templated N-dimensional image class.
*
* Detailed documentation...
*/
template< typename TPixel, unsigned int VImageDimension=2,
typename TImageTraits=DefaultImageTraits< TPixel, VImageDimension > >
class Image: public ImageBase< VImageDimension >
{
public:
...
protected:
...
private:
...
};
} // end namespace itk
#ifndef ITK_MANUAL_INSTANTIATION
#include "itkImage.hxx"
#endif
#endif // itkImage_h
In a class implementation, the following code parts always start in column one:
• Method definition,
• Method opening/closing braces.
361
362
Appendix C. Coding Style Guide
For instance,
template< typename TPixel, unsigned int VImageDimension, typename TImageTraits >
const double *
Image< TPixel, VImageDimension, TImageTraits >
::GetSpacing() const
{
...
}
Member data declarations should also be aligned when they are declared in consecutive lines.
InputPixelType m_ForegroundValue;
OutputPixelType m_BackgroundValue;
unsigned int m_MaximumIterations;
std::vector< double > m_Sensitivity;
std::vector< float > m_Overlaps;
By virtue of the principles in Section , method calls on consecutive lines should not align their
parentheses, i.e. use:
normalVectorFilter->SetIsoLevelLow( -m_CurvatureBandWidth - dimension );
normalVectorFilter->SetIsoLevelHigh( m_CurvatureBandWidth + dimension );
normalVectorFilter->SetMaxIteration( m_MaxNormalIteration );
normalVectorFilter->SetUnsharpMaskingFlag( m_NormalProcessUnsharpFlag );
normalVectorFilter->SetUnsharpMaskingWeight( m_NormalProcessUnsharpWeight );
avoiding:
normalVectorFilter->SetIsoLevelLow
(
normalVectorFilter->SetIsoLevelHigh
(
normalVectorFilter->SetMaxIteration
(
normalVectorFilter->SetUnsharpMaskingFlag (
normalVectorFilter->SetUnsharpMaskingWeight(
-m_CurvatureBandWidth - dimension );
m_CurvatureBandWidth + dimension );
m_MaxNormalIteration );
m_NormalProcessUnsharpFlag );
m_NormalProcessUnsharpWeight );
The same principle applies to consecutive statements involving any type of operator. Prefer:
double weight = 0.;
double distance = 0.;
over
double weight
= 0.;
double distance = 0.;
C.13. Code Layout and Indentation
363
Lines exceeding the recommended line length in ITK that are split in several lines must be consecutive, and must be aligned with two-space indentation:
if( coeff.size() > m_MaximumKernelWidth )
{
itkWarningMacro("Kernel size has exceeded the specified maximum width of "
<< m_MaximumKernelWidth << " and has been truncated to "
<< static_cast< unsigned long >( coeff.size() ) << " elements. You can raise "
"the maximum width using the SetMaximumKernelWidth method.");
break;
}
is preferred over
if( coeff.size() > m_MaximumKernelWidth )
{
itkWarningMacro( "Kernel size has exceeded the specified maximum width of "
<< m_MaximumKernelWidth << " and has been truncated to "
<< static_cast< unsigned long >( coeff.size() ) << " elements. You can raise "
"the maximum width using the SetMaximumKernelWidth method." );
break;
}
The same principle applies to method declarations:
unsigned int GetSplitInternal( unsigned int dim,
unsigned int i, unsigned int numberOfPieces, IndexValueType regionIndex[],
SizeValueType regionSize[] ) const override;
is preferred over
unsigned int GetSplitInternal( unsigned int dim,
unsigned int i,
unsigned int numberOfPieces,
IndexValueType regionIndex[],
SizeValueType regionSize[] ) const override;
C.13.9 Line Splitting Policy
Lines exceeding the recommended line length in ITK must be split in the necessary amount of lines.
This policy is enforced by the KWStyle pre-commit hooks (see Section C.3.1 on page 323).
If a line has to be split, the following preference order is established in ITK:
• Split the right-hand operand in an assignment =, e.g.
364
Appendix C. Coding Style Guide
const typename FixedImageType::RegionType &fixedRegion =
m_FixedImage->GetLargestPossibleRegion();
• Split after the comma separator (,) in a list, e.g.
using FilterType = AddImageFilter< BiasFieldControlPointLatticeType,
BiasFieldControlPointLatticeType, BiasFieldControlPointLatticeType >
• Split before the math operator (+, *, ||, &&, etc.) in an arithmetic or logical operation:, e.g.
while( m_ElapsedIterations++ < m_MaximumNumberOfIterations[m_CurrentLevel]
&& m_CurrentConvergenceMeasurement > m_ConvergenceThreshold )
or
centerFixedIndex[k] =
static_cast< ContinuousIndexValueType >( fixedIndex[k] )
+ static_cast< ContinuousIndexValueType >( fixedSize[k] - 1 ) / 2.0;
C.13.10 Empty Lines
As a general rule, empty lines should be used to separate code blocks that semantically belong to
separate operations, or when a portion of code is too long. In the latter case, adding documentation
lines contributes to the readability of the code.
However, no empty lines shall be added between:
• The accessor type (public, protected, private) and the declaration that immediately follows it.
• An opening/closing brace ({/}) and its subsequent/preceding line (e.g. nested namespace
braces, method definition and its body, control statements, etc).
However, an empty line should exist in a header file (.h)
• Between the Copyright notice and the include guards (e.g. #ifndef itkBoxImageFilter h).
• Between the pre-processor directives and the header includes.
• Between the header includes and the ITK namespace (i.e. namespace itk).
• Between the ITK namespace brace and the class documentation.
• Between the class documentation and the class declaration.
C.13. Code Layout and Indentation
365
• Between the access modifier and its preceding declaration, unless for the first declaration of
public.
• Between method declarations (including their corresponding documentation block).
• Between a member method declaration and any member variable declaration that immediately
follows.
• Between the ITK namespace end brace }// end namespace itk and further pre-processor
directives #ifndef ITK MANUAL INSTANTIATION.
• Between the closing pre-processor directives and include guards #endif.
For instance,
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
#ifndef itkBoxImageFilter_h
#define itkBoxImageFilter_h
#include "itkImageToImageFilter.h"
#include "itkCastImageFilter.h"
namespace itk
{
/** \class BoxImageFilter
* \brief A base class for all the filters working on a box neighborhood.
*
* This filter provides the code to store the radius information about the
* neighborhood used in the subclasses.
* It also conveniently reimplement the GenerateInputRequestedRegion() so
* that region is well defined for the provided radius.
*
* \author Gaetan Lehmann. Biologie du Developpement et de la Reproduction,
* INRA de Jouy-en-Josas, France.
* \ingroup ITKImageFilterBase
366
Appendix C. Coding Style Guide
*/
template< typename TInputImage, typename TOutputImage >
class ITK_TEMPLATE_EXPORT BoxImageFilter:
public ImageToImageFilter< TInputImage, TOutputImage >
{
public:
ITK_DISALLOW_COPY_AND_ASSIGN(BoxImageFilter);
/** Standard class type alias. */
using Self = BoxImageFilter;
using Superclass = ImageToImageFilter< TInputImage, TOutputImage >;
...
protected:
BoxImageFilter();
˜BoxImageFilter() {}
void GenerateInputRequestedRegion() override;
void PrintSelf(std::ostream & os, Indent indent) const override;
private:
RadiusType m_Radius;
};
} // end namespace itk
#ifndef ITK_MANUAL_INSTANTIATION
#include "itkBoxImageFilter.hxx"
#endif
#endif // itkBoxImageFilter_h
An empty line should exist in an implementation file (.cxx, .hxx):
• Between the Copyright notice and the include guard (e.g. #ifndef itkBoxImageFilter hxx).
• Between the pre-processor directives and the header includes.
• Between the header includes and the class implementation.
• Between the ITK namespace end brace }// end namespace itk and closing include guard
#endif.
Two empty lines are recommended between method definitions for the sake of readability. For
instance,
C.13. Code Layout and Indentation
/*=========================================================================
*
* Copyright Insight Software Consortium
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
http://www.apache.org/licenses/LICENSE-2.0.txt
*
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*=========================================================================*/
#ifndef itkBoxImageFilter_hxx
#define itkBoxImageFilter_hxx
#include "itkBoxImageFilter.h"
#include "itkProgressAccumulator.h"
namespace itk
{
template< typename TInputImage, typename TOutputImage >
BoxImageFilter< TInputImage, TOutputImage >
::BoxImageFilter()
{
m_Radius.Fill( 1 ); // A good arbitrary starting point.
}
template< typename TInputImage, typename TOutputImage >
void
BoxImageFilter< TInputImage, TOutputImage >
::SetRadius(const RadiusType & radius)
{
if( m_Radius != radius )
{
m_Radius = radius;
this->Modified();
}
}
template< typename TInputImage, typename TOutputImage >
void
BoxImageFilter< TInputImage, TOutputImage >
::PrintSelf( std::ostream & os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
367
368
Appendix C. Coding Style Guide
os << indent << "Radius: " << m_Radius << std::endl;
}
} // end namespace itk
#endif // itkBoxImageFilter_hxx
Logical code blocks that must dwell in the same method but which may not be tightly related can be
separated by two empty lines at most, e.g.
int itkHMinimaImageFilterTest( int argc, char *argv[] )
{
...
hMinimaFilter->SetInput( reader->GetOutput() );
// Run the filter
TRY_EXPECT_NO_EXCEPTION( hMinimaFilter->Update() );
// Write the output
using WriterType = itk::ImageFileWriter< OutputImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[2] );
writer->SetInput( hMinimaFilter->GetOutput() );
TRY_EXPECT_NO_EXCEPTION( writer->Update() );
std::cout << "Test finished." << std::endl;
return EXIT_SUCCESS;
}
However, it is preferable to use a single empty line and use a comment block using the // character
to describe the part of the code at issue. The comment block can start and end with an empty
comment line (//), and immediately be followed by the code, as in:
int itkHMinimaImageFilterTest( int argc, char * argv[] )
{
if( argc != 5 )
{
std::cerr << "Missing parameters." << std::endl;
std::cerr << "Usage: " << std::endl;
std::cerr << argv[0]
<< " inputImageFile"
<< " outputImageFile"
<< " height"
<< " fullyConnected" << std::endl;
return EXIT_FAILURE;
}
C.13. Code Layout and Indentation
369
//
// The following code defines the input and output pixel types and their
// associated image types.
//
constexpr unsigned int Dimension = 2;
using InputPixelType = short;
using OutputPixelType = unsigned char;
...
}
or just have an empty line before and after the comment, such as in:
template< typename TInputImage, typename TMaskImage, typename TOutputImage >
typename
N4BiasFieldCorrectionImageFilter< TInputImage, TMaskImage, TOutputImage >::RealImagePointer
N4BiasFieldCorrectionImageFilter< TInputImage, TMaskImage, TOutputImage >
::SharpenImage( const RealImageType *unsharpenedImage ) const
{
const MaskImageType *maskImage = this->GetMaskImage();
const RealImageType *confidenceImage = this->GetConfidenceImage();
#if !defined( ITK_FUTURE_LEGACY_REMOVE )
const MaskPixelType maskLabel = this->GetMaskLabel();
const bool useMaskLabel = this->GetUseMaskLabel();
#endif
//
//
//
//
Build the histogram for the uncorrected image. Store copy
in a vnl_vector to utilize vnl FFT routines. Note that variables
in real space are denoted by a single uppercase letter whereas their
frequency counterparts are indicated by a trailing lowercase 'f'.
RealType binMaximum = NumericTraits< RealType >::NonpositiveMin();
RealType binMinimum = NumericTraits< RealType >::max();
ImageRegionConstIterator< RealImageType > itU(
unsharpenedImage, unsharpenedImage->GetLargestPossibleRegion() );
...
}
Empty lines are not allowed to contain white spaces in ITK.
Logical blocks may be separated by a single-line comment (// Comment) if necessary
in the implementation file (.h).
No comment line or any other separation string (e.g.
/***********************/) must be placed between the definition of two methods in the implementation file (.cxx, .hxx).
370
Appendix C. Coding Style Guide
C.13.11 New Line Character
Use std::endl to introduce a new line instead of \n in string literals, e.g.
template< typename TInputImage >
void
MinimumMaximumImageCalculator< TInputImage >
::PrintSelf( std::ostream & os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
os << indent << "Minimum: "
<< static_cast< typename NumericTraits< PixelType >::PrintType >( m_Minimum )
<< std::endl;
os << indent << "Maximum: "
<< static_cast< typename NumericTraits< PixelType >::PrintType >( m_Maximum )
<< std::endl;
os << indent << "IndexOfMinimum: " << m_IndexOfMinimum << std::endl;
os << indent << "IndexOfMaximum: " << m_IndexOfMaximum << std::endl;
itkPrintSelfObjectMacro( Image );
os << indent << "Region: " << std::endl;
m_Region.Print( os, indent.GetNextIndent() );
os << indent << "RegionSetByUser: " << m_RegionSetByUser << std::endl;
}
C.13.12 End Of File Character
The file must be terminated by a (preferably single) blank line. This policy is enforced by the
KWStyle pre-commit hooks (see Section C.3.1 on page 323).
C.14 Increment/decrement Operators
Systematically use the pre-increment(decrement) syntax, instead of the post-increment(decrement)
syntax:
for( unsigned int i = 0; i < ImageDimension; ++i )
{
...
}
Although the advantage can be very little when using it with a standard type, in the case of iterators
over potentially large structures, the optimization performed by modern compilers may involve a
significant advantage:
C.15. Empty Arguments in Methods
371
while( it != m_Container.end() )
{
...
++it;
}
C.15 Empty Arguments in Methods
The use of the void keyword is discouraged for methods not requiring input arguments. Hence, they
are declared and called with an empty opening/closing parenthesis pair:
/** Method doc. */
void methodName();
and
this->methodName();
C.16 Ternary Operator
The use of the ternary operator
for( unsigned int i = 0; i < m_NumOfThreads; ++i )
{
for( unsigned int j = ( i == 0 ? 0 : m_Boundary[i - 1] + 1 ); j <= m_Boundary[i]; ++j )
{
m_GlobalZHistogram[j] = m_Data[i].m_ZHistogram[j];
}
}
is generally discouraged in ITK, especially in cases where complicated statements have any part of
the ternary operator.
Thus, the above should be expanded to
for( unsigned int i = 0; i < m_NumOfThreads; ++i )
{
if( i == 0 )
{
for( unsigned int j = 0; j <= m_Boundary[i]; ++j )
{
m_GlobalZHistogram[j] = m_Data[i].m_ZHistogram[j];
}
372
Appendix C. Coding Style Guide
}
else
{
for( unsigned int j = m_Boundary[i - 1] + 1; j <= m_Boundary[i]; ++j )
{
m_GlobalZHistogram[j] = m_Data[i].m_ZHistogram[j];
}
}
}
or, performing a code refactoring:
for( unsigned int j = 0; j <= m_Boundary[i]; ++j )
{
m_GlobalZHistogram[j] = m_Data[0].m_ZHistogram[j];
}
for( unsigned int i = 1; i < m_NumOfThreads; ++i )
{
for( unsigned int j = m_Boundary[i - 1] + 1; j <= m_Boundary[i]; ++j )
{
m_GlobalZHistogram[j] = m_Data[i].m_ZHistogram[j];
}
}
However, in simple constructs, such as when initializing a variable that could be const, e.g.
for( unsigned int i = 0; j < ImageDimension; i++ )
{
const elementSign = ( m_Step[i] > 0 ) ? 1.0 : -1.0;
flipMatrix[i][i] = elementSign;
}
a ternary operator can have significant advantage in terms of the reading speed over the alternative
if-else statement with duplicated syntax:
for( unsigned int i = 0; j < ImageDimension; i++ )
{
if( m_Step[i] > 0 )
{
elementSign = 1.0;
}
else
{
elementSign = -1.0;
}
flipMatrix[i][i] = elementSign;
}
And hence, the ternary operator is accepted in such cases.
C.17. Using Standard Macros
373
C.17 Using Standard Macros
There are several macros defined in the file itkMacro.h. These macros should be used because
they perform several important operations that if not done correctly can cause serious, hard to debug
problems in the system.
These operations are:
• Object modified time is properly managed.
• Debug information is printed.
• Reference counting is handled properly.
• Disallow copy semantics by deleting copy constructor and assignment operator.
Some of the more important object macros are:
• itkNewMacro(T): Creates the static class method New(void) that interacts with the object
factory to instantiate objects. The method returns a SmartPointer properly reference
counted.
• itkTypeMacro(thisClass, superclass): Adds standard methods a class, mainly type information. Adds the GetNameOfClass() method to the class.
• ITK DISALLOW COPY AND ASSIGN(TypeName): Disallow copying by declaring copy constructor and assignment operator deleted. This must be declared in the public section.
• itkDebugMacro(x): If debug is set on a subclass of itk::Object, prints debug information
to the appropriate output stream.
• itkStaticConstMacro(name, type, value): Creates a static const member of type
type and sets it to the value value.
• itkSetMacro(name, type): Creates a method SetName() that takes an argument of type
type.
• itkGetMacro(name, type): Creates a method GetName() that returns a non-const value of
type type.
• itkGetConstMacro(name, type): Creates a method GetName() that returns a const value
of type type.
• itkSetStringMacro(name): Creates a method SetName() that takes an argument of type
const char*.
• itkGetStringMacro(name): Creates a method GetName() that returns an argumentof type
const char*.
374
Appendix C. Coding Style Guide
• itkBooleanMacro(name): Creates two methods named NameOn and NameOff that set
true/false boolean values.
• itkSetObjectMacro(name, type): Creates a method SetName() that takes argument type
type *. For ITK objects, itkSetObjectMacro must be used in lieu of itkSetMacro.
• itkGetConstObjectMacro(name, type): Creates a method named GetName() that returns
a itk::SmartPointer to a type type.
• itkSetConstObjectMacro(name, type): Creates a method SetName() that takes an argument of type const type *.
• itkGetConstObjectMacro(name, type): Creates a method named GetName() that returns
a const itk::SmartPointer to a type type.
• itkSetClampMacro(name, type, min, max): Creates a method named SetName() that
takes an argument of type type constraining it to the [min, max] closed interval.
Furthermore, the ITK symbol visibility is governed by some macros using the following rules:
• $ModuleName EXPORT: export for non-templated classes.
• ITK TEMPLATE EXPORT: export for templated classes.
• ITK FORWARD EXPORT: export for forward declarations.
This supports the all the combinations of:
• macOS, Linux and Windows operating systems,
• the shared BUILD SHARED LIBS ON and OFF static linking modes,
• explicit and implicit template instantiation,
• the
CMake
CMAKE CXX VISIBILITY PRESET
-fvisibility=hidden),
flag
set
to
hidden
(i.e.
• the CMake flag CMAKE WINDOWS EXPORT ALL SYMBOLS:BOOL=ON.
Please review this file and become familiar with these macros.
All classes must declare the basic macros for object creation and run-time type information (RTTI):
/** Method for creation through the object factory. */
itkNewMacro( Self );
/** Run-time type information (and related methods). */
itkTypeMacro( Image, ImageBase );
C.18. Exception Handling
375
Basic types (e.g. int, double, etc.) must be returned by value using the method defined through
the itkGetMacro macro; member data pointers that must not be modified should be returned using
the method defined through the itkGetConstMacro.
When using a macro that accepts a statement, a semi-colon (;) is not required for the argument, e.g.
TRY_EXPECT_NO_EXCEPTION( writer->Update() );
C.18 Exception Handling
ITK exceptions are defined in itkExceptionObject.h. Derived exception classes include:
• itkImageFileReaderException.h for exceptions thrown while trying to read image files
(i.e. DICOM files, JPEG files, metaimage files, etc.).
• itkMeshFileReaderException.h for exceptions thrown while trying to read mesh files.
• itkMeshFileWriterrException.h for exceptions thrown while trying to write mesh files.
Methods throwing exceptions must indicate so in their declaration as in:
/** Initialize the components related to supporting multiple threads. */
virtual void MultiThreadingInitialize(void) throw ( ExceptionObject );
When a code block is liable to throw an exception, a try/catch block must be used to deal with the
exception. The following rules apply to such blocks
• The exception object should generally be redirected to the error output std::cerr and be
trailed with a line break std::endl.
• In classes that have a member to store the error messages (e.g. m ExceptionMessage), the
exception description must be obtained using GetDescription() and be assigned to the exception message member.
• Otherwise, the error shall be re-thrown to the caller.
For instance,
try
{
// Code liable to throw an exception
}
catch( ExceptionObject & exc )
{
376
Appendix C. Coding Style Guide
std::cerr << exc << std::endl;
}
catch( std::exception & exc )
{
std::cerr << exc.what() << std::endl;
}
For instance,
try
{
m_ExceptionMessage = "";
this->TestFileExistanceAndReadability();
}
catch( ExceptionObject & exc )
{
m_ExceptionMessage = exc.GetDescription();
}
For instance,
// Do the optimization
try
{
m_Optimizer->StartOptimization();
}
catch( ExceptionObject & exc )
{
// An error has occurred in the optimization.
// Update the parameters.
m_LastTransformParameters = m_Optimizer->GetCurrentPosition();
// Pass the exception to the caller
throw exc;
}
Exceptions can also be thrown outside try/catch blocks, when there is sufficient evidence for that,
e.g. a filename to be read is empty. In such cases, depending on the exception class the following
information should be included:
• the file
• the line
• the error message
of the related exception, e.g.
C.18. Exception Handling
377
if( m_FileName == "" )
{
throw MeshFileReaderException( __FILE__, __LINE__, "FileName must be specified", ITK_LOCATION );
}
See Section 3.2.5 on page 31 for details about error handling in ITK.
C.18.1 Errors in Pipelines
When in a function an element must have a given value, a check must ensure that such condition is
met. The condition is generally being non-null (e.g. for I/O images), or different from zero (e.g. for
sizes, etc.).
When the I/O objects are not set, the ITK itkAssertInDebugAndIgnoreInReleaseMacro macro
is used: the ITK processing framework should handle this situation and throw the appropriate exception (e.g. the itk::ProcessObject class), such macro assertion is preferred over an exception,
e.g.
template< typename TInputImage, typename TOutputImage >
void
BinShrinkImageFilter< TInputImage, TOutputImage >
::GenerateInputRequestedRegion()
{
// Call the superclass' implementation of this method.
Superclass::GenerateInputRequestedRegion();
// Get pointers to the input and output.
InputImageType * inputPtr =
const_cast< InputImageType * >( this->GetInput() );
const OutputImageType * outputPtr = this->GetOutput();
itkAssertInDebugAndIgnoreInReleaseMacro( inputPtr != nullptr );
itkAssertInDebugAndIgnoreInReleaseMacro( outputPtr );
...
}
e.g
template< typename TInputImage, typename TOutputImage >
void
PatchBasedDenoisingBaseImageFilter< TInputImage, TOutputImage >
::SetPatchWeights( const PatchWeightsType& weights )
{
itkAssertOrThrowMacro( this->GetPatchLengthInVoxels() == weights.GetSize(),
"Unexpected patch size encountered while setting patch weights" );
378
Appendix C. Coding Style Guide
...
}
The itkAssertInDebugAndIgnoreInReleaseMacro macro is useful for logic checks in performance sections that should never be violated. itkAssertOrThrowMacro is fine for non-performance
critical sections where it would be helpful to also add an error message.
C.19 Messages
C.19.1 Messages in Macros
Messages written for debugging purposes which are deemed to be appropriate to remain in the code,
and those reported when raising exceptions should be using the output stream operator (<<) to add
the message to any previous string in the buffer. Messages should start with capitals and should not
finish with a period. Self-contained sentences must be streamed.
itkDebugMacro( << "Computing Bayes Rule" );
or
itkExceptionMacro( << "The size of the mask differs from the input image" );
C.19.2 Messages in Tests
ITK tests are run automatically, and hence, results are not read by humans. Although at times it
may be beneficial (e.g. when a large number of regressions are done in a test, checking different
image types, or using different approaches), tests should not generally contain messages sent to the
standard output.
One of the general exceptions is a message at the end of the test signaling that the test has ended:
...
std::cout << "Test finished.";
return EXIT_SUCCESS;
In case of test failure, this allows ITK maintainers to know whether the issue came from the test
execution or from a potential regression against a baseline.
When failures are to be reported in a test (i.e. if an insufficient number of test arguments are provided
or a regression fails), messages must be redirected to the error output.
C.19. Messages
379
When an insufficient number of parameters are found, the test arguments should be written in medial
capitals, starting with lower cases.
if( argc != 3 )
{
std::cerr << "Missing parameters." << std::endl;
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImage outputImage" << std::endl;
return EXIT_FAILURE;
}
If the length of the message is longer than 200 characters, the argument list must be split in its
individual components, leaving a white space at the beginning of each line, and the std::cerr
redirection should only exist on the first line. A final line break must be always added. Optional
arguments must be enclosed in square brackets [].
if( argc < 3 )
{
std::cerr << "Missing parameters." << std::endl;
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImage"
<< " outputImage"
<< " [foregroundValue]
<< " [backgroundValue]" << std::endl;
return EXIT_FAILURE;
}
When a regression fails in a check, it must be clearly stated that the test failed, and details about the
method that failed to return the correct value, as well as the expected and returned values, must be
provided:
bool tf = colors->SetColor( 0, 0, 0, 0, name );
if( tf != true )
{
std::cerr << "Test failed!" << std::endl;
std::cerr << "Error in itk::ColorTable::SetColor" << std::endl;
std::cerr << "Expected: " << true << ", but got: "
<< tf << std::endl;
return EXIT_FAILURE;
}
If any index is involved (i.e. the test failure stems from a given index position when checking the
values of a list, image, etc.), the index at issue must be specified in the message:
// Check the content of the result image
const OutputImageType::PixelType expectedValue =
static_cast< OutputImageType::PixelType >( valueA * valueB );
const OutputImageType::PixelType epsilon = 1e-6;
380
Appendix C. Coding Style Guide
while( !oIt.IsAtEnd() )
{
if( !itk::Math::FloatAlmostEqual( oIt.Get(), expectedValue, 10, epsilon ) )
{
std::cerr.precision( static_cast< int >( itk::Math::abs( std::log10( epsilon ) ) ) );
std::cerr << "Test failed!" << std::endl;
std::cerr << "Error in pixel value at index [" << oIt.GetIndex() << "]" << std::endl;
std::cerr << "Expected value " << expectedValue << std::endl;
std::cerr << " differs from " << oIt.Get();
std::cerr << " by more than " << epsilon << std::endl;
return EXIT_FAILURE;
}
++oIt;
}
C.20 Concept Checking
C.21 Printing Variables
All member variables, regardless of whether they are publicly exposed or not, must be printed in a
class’ PrintSelf method. Besides being an important sanity check that allows to identify uninitialized variables, it allows to know the state of a class instance at any stage.
The basic conventions for printing member variables are:
• Each variable must be printed on a new line and be indented.
• The name of the variable must immediately follow to the indentation.
• The Superclass must always be printed.
Thus, the general layout for printing member variables is:
Superclass::PrintSelf( os, indent );
os << indent << "" << << std::endl;
The following additional conventions apply to printing member variables:
• When printing constructs such as matrices, double indentation should be used to print its
contents using itk::Indent::GetNextIndent().
• Objects that can be null (such as
itkPrintSelfObjectMacro macro.
itk::SmartPointer) must be printed using the
C.22. Checking for Null
381
• Without harm to the previous convention, constructs such as images can be printed using the
Print method.
• Objects that have been declared as a type alias must be casted statically using the
NumericTraits< Type >::PrintType > helper formatting.
• The order of the variables should be the same used in their declaration.
For instance,
template< typename TInputImage >
void
MinimumMaximumImageCalculator< TInputImage >
::PrintSelf( std::ostream & os, Indent indent ) const
{
Superclass::PrintSelf( os, indent );
os << indent << "Minimum: "
<< static_cast< typename NumericTraits< PixelType >::PrintType >( m_Minimum )
<< std::endl;
os << indent << "Maximum: "
<< static_cast< typename NumericTraits< PixelType >::PrintType >( m_Maximum )
<< std::endl;
os << indent << "IndexOfMinimum: " << m_IndexOfMinimum << std::endl;
os << indent << "IndexOfMaximum: " << m_IndexOfMaximum << std::endl;
itkPrintSelfObjectMacro( Image );
os << indent << "Region: " << std::endl;
m_Region.Print( os, indent.GetNextIndent() );
os << indent << "RegionSetByUser: " << m_RegionSetByUser << std::endl;
}
C.22 Checking for Null
ITK’s itk::SmartPointer constructs can be checked against the null pointer using either the
syntax
itkSmartPtr.IsNull();
or
itkSmartPtr == nullptr;
The latter, being more explicit, is preferred over the former.
382
Appendix C. Coding Style Guide
C.23 Writing Tests
The following section provides additional rules that apply to writing tests in ITK.
C.23.1 Code Layout in Tests
The following general layout is recommended for ITK unit tests:
• Input argument number check.
• Input image read (or generation).
• foo class instantiation and basic object checks (e.g. EXERCISE BASIC OBJECT METHODS).
• foo class properties’ input argument read and test (e.g.
itkTestingMacro.h, such as TEST SET GET OBJECT, etc.).
using the macros in
• foo class Update().
• Regression checks.
• Output image write.
Note that constant declarations (e.g. image dimensions, etc.) and type alias declarations (e.g. pixel
and image types, etc.) should be local to where they are used for the sake of readability. If the test
main body uses them, they should be put after the input argument number check section.
C.23.2 Regressions in Tests
Tests should run as long as possible to report as much failures as possible before returning
int itkAbsImageFilterAndAdaptorTest( int, char* [] )
{
int testStatus = EXIT_SUCCESS;
...
// Check the content of the result image.
const OutputImageType::PixelType epsilon = 1e-6;
ot.GoToBegin();
it.GoToBegin();
while( !ot.IsAtEnd() )
{
std::cout.precision( static_cast< int >( itk::Math::abs( std::log10( epsilon ) ) ) );
std::cout << ot.Get() << " = ";
std::cout << itk::Math::abs( it.Get() ) << std::endl;
C.23. Writing Tests
383
const InputImageType::PixelType input = it.Get();
const OutputImageType::PixelType output = ot.Get();
const OutputImageType::PixelType absolute = itk::Math::abs(input);
if( !itk::Math::FloatAlmostEqual( absolute, output, 10, epsilon ) )
{
std::cerr.precision( static_cast< int >( itk::Math::abs( std::log10( epsilon ) ) ) );
std::cerr << "Test failed!" << std::endl;
std::cerr << "Error in pixel value at index [" << oIt.GetIndex() << "]" << std::endl;
std::cerr << "Expected value " << abs(" << input << ") = " << absolute << std::endl;
std::cerr << " differs from " << output();
std::cerr << " by more than " << epsilon << std::endl;
testStatus = EXIT_FAILURE;
}
++ot;
++it;
}
//
// Test AbsImageAdaptor
//
...
// Check the content of the diff image.
std::cout << "Comparing the results with those of an Adaptor" << std::endl;
std::cout << "Verification of the output " << std::endl;
// Create an iterator for going through the image output.
OutputIteratorType dt( diffImage, diffImage->GetRequestedRegion() );
dt.GoToBegin();
while( !dt.IsAtEnd() )
{
std::cout.precision( static_cast< int >( itk::Math::abs( std::log10( epsilon ) ) ) );
const OutputImageType::PixelType diff = dt.Get();
if( !itk::Math::FloatAlmostEqual( diff, ( OutputImageType::PixelType )0, 10, epsilon ) )
{
std::cerr.precision( static_cast< int >( itk::Math::abs( std::log10( epsilon ) ) ) );
std::cerr << "Test failed!" << std::endl;
std::cerr << "Error in pixel value at index [" << dt.GetIndex() << "]" << std::endl;
std::cerr << "Expected difference " << diff << std::endl;
std::cerr << " differs from 0 ";
std::cerr << " by more than " << epsilon << std::endl;
testStatus = EXIT_FAILURE;
}
++dt;
}
std::cout << "Test finished.";
return testStatus;
}
Note that when dealing with real numbers, a tolerance parameter must be specified in or-
384
Appendix C. Coding Style Guide
der to avoid precision issues.
Furthermore, setting the output message precision with
std::cerr.precision(int n); is recommended to allow for easy identification of the magnitude of the error.
When the magnitude of the error needs to be reported, as in the above examples, the error message
should be split into different lines, all starting with the error output redirection std::cerr << "";.
Care must be taken to appropriately add white space for a correct formatting of the message.
C.23.3 Arguments in Tests
Tests generally require input arguments, whether the filename of an input image, the output image filename for regression purposes, or a variety of other parameters to be set to a filter instance.
However, some tests are self-contained and do not need any input parameter.
In such cases, the test’s main method argument variables do not need to be specified. The generally
accepted syntax for these cases is:
int itkVersionTest( int, char* [] )
Otherwise, it may happen that some test may or may not accept arguments, depending on the implementation. In such cases, the itkNotUsed ITK macro must be used to avoid compiler warnings:
int itkGaborKernelFunctionTest( int itkNotUsed( argc ), char * itkNotUsed( argv )[] )
When a test requires input arguments, a basic sanity check on the presence of the required arguments
must be made. If the test does not have optional arguments, the exact match for the input arguments
must be checked:
if( argc != 3 )
{
std::cerr << "Missing parameters." << std::endl;
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImage outputImage " << std::endl;
return EXIT_FAILURE;
}
If the test does have optional arguments, the presence of the set of compulsory arguments must be
checked:
if( argc < 3 )
{
std::cerr << "Missing parameters." << std::endl;
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImage"
<< " outputImage"
C.24. Writing Examples
385
<< " [foregroundValue]
<< " [backgroundValue]" << std::endl;
return EXIT_FAILURE;
}
C.23.4 Test Return Value
Tests must always return a value of type int, even if bool is tempting:
int itkVersionTest( int, char* [] )
Thus, if a test requires a variable to store its exit value due to the need of multiple regressions, an
int variable must be declared:
int itkAbsImageFilterAndAdaptorTest( int, char* [] )
{
int testStatus = EXIT_SUCCESS;
...
return testStatus;
Tests must exit gracefully using the values EXIT SUCCESS (in case of success) or EXIT FAILURE
(in case of failure) defined in the stdlib.h library values. Other ways of exiting tests such as
exit(1);, exit(255);, or exit(EXIT FAILURE); are not allowed in ITK.
C.24 Writing Examples
Many ITK examples are used in this software guide to demonstrate ITK’s architecture and development.
Thanks to scripting work, parts of the *.cxx example files within special placeholders are included
in this software guide. The LATEXplaceholders available to the code for such purpose are:
• Software Guide : BeginLatex and Software Guide : EndLatex: the text within these placeholders is included in this software guide for text explanations, e.g.
//
//
//
//
//
//
Software Guide : BeginLatex
Noise present in the image can reduce the capacity of this filter to grow
large regions. When faced with noisy images, it is usually convenient to
pre-process the image by using an edge-preserving smoothing filter. Any of
the filters discussed in Section˜\ref{sec:EdgePreservingSmoothingFilters}
386
Appendix C. Coding Style Guide
//
//
//
//
//
could be used to this end. In this particular example we use the
\doxygen{CurvatureFlowImageFilter}, so we need to include its header
file.
Software Guide : EndLatex
• Software Guide : BeginCodeSnippet and Software Guide : EndCodeSnippet: the text
within these placeholders is included in this software guide for verbatim code snippets, e.g.
// Software Guide : BeginCodeSnippet
#include "itkCurvatureFlowImageFilter.h"
// Software Guide : EndCodeSnippet
Note that anything inside these gets inserted into the document; avoid blank lines or too much
whitespace. Make sure any LATEXcomments included in the code are correct in terms of grammar,
spelling, and are complete sentences.
Note that the code should not exceed 79 columns or it will go out of margins in the final document.
It is recommended that the LATEXcomment blocks are aligned to the code for the sake of readability.
C.25 Doxygen Documentation System
Doxygen is an open-source, powerful system for automatically generating documentation from
source code. To use Doxygen effectively, the developer must insert comments, delimited in a special way, that Doxygen extracts to produce the documentation. While there are a large number of
options to Doxygen, ITK community members are required at a minimum to insert Doxygen commands listed in this section.
See more at http://www.stack.nl/ dimitri/doxygen/
C.25.1 General Principles
ITK uses a subset of C-style Doxygen markdown. No other markdown style (e.g. Qt, Javadoc) shall
be used.
In ITK, documentation is placed before the documented construct (i.e. a class, a method, a variable,
etc.).
Although not the general rule, if a comment is too short or applies to a single line so that it is a clear
candidate to dwell on that line, it can be placed on the same line using the // comment style, and
leaving a single space before the statement-ending ; and the comment itself, e.g.
C.25. Doxygen Documentation System
387
template< typename TInputImage, typename TOutputImage >
BoxImageFilter< TInputImage, TOutputImage >
::BoxImageFilter()
{
m_Radius.Fill( 1 ); // A good arbitrary starting point.
}
Correct English and complete, grammatically correct sentences must be used when documenting.
Finish the sentences with a period (.).
C.25.2 Documenting Classes
Classes must be documented using the \class, \brief, and \ingroup Doxygen commands, followed by the detailed class description. The comment starts with /**, each subsequent line has an
aligned *, and the comment block terminates with a */ on a line of its own. A single white space
should exist between these keywords/characters and the documentation body, e.g.
/** \class Object
* \brief Base class for most ITK classes.
*
* Object is the second-highest level base class for most itk objects.
* It extends the base object functionality of LightObject by
* implementing debug flags/methods and modification time tracking.
*
* \ingroup Module
*/
The \ingroup and other additional Doxygen keywords must be separated from their preceding and
following lines by an empty comment * line.
Doxygen keywords that may most commonly apply to complete a class documentation are
• \note
• \sa
Math formulas in class documentation are formatted following the LATEXguidelines. For more information, please visit https://www.stack.nl/ dimitri/doxygen/manual/formulas.html.
Every class must be documented.
C.25.3 Documenting Methods
The method Doxygen documentation must be placed in the header file (.h).
388
Appendix C. Coding Style Guide
A single white space should separate the comment characters (/**, *, or */) and the comment itself.
The starting (/**) and ending (*/) comment characters must be placed on the same lines as the
comment text, and the lines with the asterisk (*) character should be aligned, e.g.
/** Provides opportunity for the data object to insure internal
* consistency before access. Also causes owning source/filter (if
* any) to update itself. The Update() method is composed of
* UpdateOutputInformation(), PropagateRequestedRegion(), and
* UpdateOutputData(). This method may call methods that throw an
* InvalidRequestedRegionError exception. This exception will leave
* the pipeline in an inconsistent state. You will need to call
* ResetPipeline() on the last ProcessObject in your pipeline in
* order to restore the pipeline to a state where you can call
* Update() again. */
virtual void Update();
The base class virtual method documentation is automatically applied for such methods in derived
class unless they are overridden. Virtual methods whose meaning or set of instructions differs from
their base class need to be documented in the derived classes. If the base class method documentation
applies, they need not to be documented in derived classes (e.g. the PrinSelf method).
Intra-method documentation must be done where necessary using single-line comment style, and
must be repeated for every line. A single white space should separate the comment character // and
the comment itself, e.g.
// We wish to copy whole lines, otherwise just use the basic implementation.
// Check that the number of internal components match.
if( inRegion.GetSize()[0] != outRegion.GetSize()[0]
|| NumberOfInternalComponents != ImageAlgorithm::PixelSize::Get( outImage ) )
{
ImageAlgorithm::DispatchedCopy( inImage, outImage, inRegion,
outRegion );
return;
}
Self-contained, complete sentences must end with a period.
Every method must be documented.
C.25.4 Documenting Data Members
Class member variables should be documented through their corresponding Get##name/Set##name
methods, using a comment block style shown in the following example:
public:
/** Set/Get the standard deviation of the Gaussian used for smoothing. */
itkSetMacro( Sigma, SigmaArrayType );
C.25. Doxygen Documentation System
389
itkGetConstMacro( Sigma, SigmaArrayType );
private:
SigmaArrayType m_Sigma;
The documentation block must be aligned to the Get##name/Set##name method indentation.
For bool type variables, the recommended way of documenting its default value is using “On” for
true and “Off” for false:
/** Set/Get direction along the gradient to search.
* Set to true to use the direction that the gradient is pointing;
* set to false for the opposite direction. Default is Off. */
itkGetConstMacro( Polarity, bool );
itkSetMacro( Polarity, bool );
itkBooleanMacro( Polarity );
Member variables that do not have either a Get##name or a Set##name method should also be
documented following the above guidelines.
C.25.5 Documenting Macros
The documentation block in a macro should start in column one, and should be placed immediately
before the macro definition, and will use /* as the starting character, immediately followed by the
body of the documentation, which shall be split into different lines starting with asterisks (*), aligned
to the preceding asterisk character, with a single white space indentation for the text, and will end
with the */ character. The macro definition should have a double indentation, e.g.
/** This macro is used to print debug (or other information). They are
* also used to catch errors, etc. Example usage looks like:
* itkDebugMacro( << "this is debug info" << this->SomeVariable ); */
#if defined( NDEBUG )
#define itkDebugMacro(x)
#define itkDebugStatement(x)
#else
#define itkDebugMacro(x)
{
if ( this->GetDebug() && ::itk::Object::GetGlobalWarningDisplay() )
{
std::ostringstream itkmsg;
itkmsg << "Debug: In " __FILE__ ", line " << __LINE__ << "\n"
<< this->GetNameOfClass() << " (" << this << "): " x
<< "\n\n";
::itk::OutputWindowDisplayDebugText( itkmsg.str().c_str() );
}
}
\
\
\
\
\
\
\
\
\
\
390
Appendix C. Coding Style Guide
C.25.6 Documenting Tests
Generally, an ITK test does not need to have a documentation block stating its purpose if this is restricted to testing a single class. However, for tests that check multiple classes or complex pipelines,
documenting its motivation and purpose, as well as its general schema, is recommended.
The documentation block should start in column one, and should be placed immediately before the
main method, and will use /* as the starting character, immediately followed by the body of the
documentation, which shall be split into different lines starting with asterisks (*), aligned to the
preceding asterisk character, with a single white space indentation for the text, and will end with the
*/ character, e.g.
/* Test the SetMetricSamplingPercentage and SetMetricSamplingPercentagePerLevel.
* We only need to explicitly run the SetMetricSamplingPercentage method because
* it invokes the SetMetricSamplingPercentagePerLevel method. */
int itkImageRegistrationSamplingTest( int, char *[] )
It is recommended to document the body of the test with single-line comment style where appropriate.
C.26 CMake Style
For writing CMake scripts, the community member is referred to the standard CMake style.
C.27 Documentation Style
The Insight Software Consortium has adopted the following guidelines for producing supplemental
documentation (documentation not produced by Doxygen):
• The common denominator for documentation is either PDF or HTML. All documents in the
system should be available in these formats, even if they are mastered by another system.
• Presentations are acceptable in Microsoft PowerPoint format.
• Administrative and planning documents are acceptable in Microsoft Word format (either
.docx or .rtf).
• Larger documents, such as the user’s or developer’s guide, are written in LATEX.
BIBLIOGRAPHY
[1] M. H. Austern. Generic Programming and the STL:. Professional Computing Series. AddisonWesley, 1999. 3.2.1
[2] K.R. Castleman. Digital Image Processing. Prentice Hall, Upper Saddle River, NJ, 1996. 6.4.1,
6.4.2
[3] E. Gamma, R. Helm, R. Johnson, and J. Vlissides. Design Patterns, Elements of Reusable
Object-Oriented Software. Professional Computing Series. Addison-Wesley, 1995. 3.2.6, 4.3.9,
8.6
[4] R.C. Gonzalez and R.E. Woods. Digital Image Processing. Addison-Wesley, Reading, MA,
1993. 6.4.1, 6.4.1, 6.4.2
[5] H. Gray. Gray’s Anatomy. Merchant Book Company, sixteenth edition, 2003. 4.1.5
[6] H. Lodish, A. Berk, S. Zipursky, P. Matsudaira, D. Baltimore, and J. Darnell. Molecular Cell
Biology. W. H. Freeman and Company, 2000. 4.1.5
[7] D. Malacara. Color Vision and Colorimetry: Theory and Applications. SPIE PRESS, 2002.
4.1.5, 4.1.5
[8] D. Musser and A. Saini. STL Tutorial and Reference Guide. Professional Computing Series.
Addison-Wesley, 1996. 3.2.1
[9] G. Wyszecki. Color Science: Concepts and Methods, Quantitative Data and Formulae. WileyInterscience, 2000. 4.1.5, 4.1.5
INDEX
Accept()
itk::Mesh, 103
AddVisitor()
itk::Mesh, 103
BoundaryFeature, 88
BufferedRegion, 206
CDash, 244
CellAutoPointer, 75
TakeOwnership(), 75, 78, 82, 84, 91
CellBoundaryFeature, 88
CellDataContainer
Begin(), 79, 83
ConstIterator, 79, 83
End(), 79, 83
Iterator, 79, 83
CellDataIterator
increment, 79, 83
Value(), 79, 83
CellInterface
iterating points, 101
PointIdsBegin(), 101
PointIdsEnd(), 101
CellInterfaceVisitor, 98, 100
requirements, 99, 100
Visit(), 99, 100
CellIterator
increment, 77
Value(), 77
CellMultiVisitorType, 103
CellsContainer
Begin(), 77, 87, 92, 96
End(), 77, 87, 92, 96
CellType
creation, 75, 78, 82, 84, 91
GetNumberOfPoints(), 77
PointIdIterator, 87, 92
PointIdsBegin(), 87, 92
PointIdsEnd(), 87, 92
Print(), 77
CellVisitor, 98–100, 102
CMake, 12
downloading, 12
Command/Observer design pattern, 32
const-correctness, 67, 69
ConstIterator, 67, 69
convolution
kernels, 178
operators, 178
convolution filtering, 177
Dashboard, 244
data object, 36, 205
data processing pipeline, 37, 205
discussion, 7
394
down casting, 77
Downloading, 10
edge detection, 174
error handling, 31
event handling, 32
exceptions, 31
factory, 29
filter, 37, 205
overview of creation, 206
forward iteration, 150
garbage collection, 30
Gaussian blurring, 180
Generic Programming, 149
generic programming, 28, 149
GetBoundaryAssignment()
itk::Mesh, 89
GetNumberOfBoundaryFeatures()
itk::Mesh, 89
GetNumberOfFaces()
TetrahedronCell, 102
GetPointId(), 100
Git, 243
Hello World, 21
image region, 205
ImageAdaptor
RGB blue channel, 196
RGB green channel, 195
RGB red channel, 194
ImageAdaptors, 191
ImageLinearIteratorWithIndex
4D images, 161
InvokeEvent(), 32
iteration region, 150
Iterators
advantages of, 149
and 4D images, 161
and bounds checking, 152
and image lines, 158
and image regions, 150, 153, 154, 156
Index
and image slices, 162
const, 150
construction of, 150, 156
definition of, 149
Get(), 152
GetIndex(), 152
GoToBegin(), 150
GoToEnd(), 150
image, 149–190
image dimensionality, 156
IsAtBegin(), 152
IsAtEnd(), 152
neighborhood, 167–190
operator++(), 151
operator+=(), 151
operator–, 151
operator-=(), 151
programming interface, 150–154
Set(), 152
SetPosition(), 152
speed, 154, 156
Value(), 153
iterators
neighborhood
and convolution, 178
ITK
advanced configuration, 15
building, 18
configuration, 14
discussion, 7
downloading release, 11
Git repository, 11, 243
history, 8
installation, 19
modules, 15
itk::ArrowSpatialObject, 120
itk::AutomaticTopologyMeshSource, 93
AddPoint(), 94
AddTetrahedron(), 94
header, 93
IdentifierArrayType, 93
IdentifierType, 93
itk::AutoPointer, 75
Index
TakeOwnership(), 75, 78, 82, 84, 91
itk::BlobSpatialObject, 120
itk::Cell
CellAutoPointer, 75
itk::CellInterface
GetPointId(), 100
itk::Command, 32
itk::CovariantVector, 72
Header, 70
Instantiation, 70
itk::PointSet, 70
itk::CylinderSpatialObject, 122
itk::DefaultStaticMeshTraits
Header, 80
Instantiation, 81
itk::DTITubeSpatialObject, 142
itk::EllipseSpatialObject, 123
itk::GaussianSpatialObject, 125
itk::GroupSpatialObject, 126
itk::Image, 36
Allocate(), 45
direction, 50
GetPixel(), 47, 54
Header, 43
Index, 44, 51
IndexType, 44
Instantiation, 43
itk::ImageRegion, 44
New(), 44
origin, 49
PhysicalPoint, 51
Pointer, 44
read, 46
RegionType, 44
SetDirection(), 50
SetOrigin(), 49
SetPixel(), 47
SetRegions(), 45
SetSpacing(), 49
Size, 44
SizeType, 44
Spacing, 49
TransformPhysicalPointToIndex(), 51
395
Vector pixel, 56
itk::ImageRandomConstIteratorWithIndex,
166–167
and statistics, 167
begin and end positions, 166
example of using, 167
ReinitializeSeed(), 167
sample size, 166
SetNumberOfSamples(), 167
itk::ImageSliceIteratorWithIndex
example of using, 163–166
IsAtEndOfSlice(), 163
IsAtReverseEndOfSlice(), 163
NextSlice(), 163
PreviousSlice(), 163
SetFirstDirection(), 163
SetSecondDirection(), 163
itk::ImageAdaptor
Header, 192, 194, 197, 199
Instantiation, 192, 194, 197, 199
performing computation, 199
RGB blue channel, 196
RGB green channel, 195
RGB red channel, 194
itk::ImageFileReader
GetOutput(), 46
Instantiation, 46
New(), 46
Pointer, 46
RGB Image, 54
SetFileName(), 46
Update(), 46
itk::ImageLinearIteratorWithIndex, 158–162
example of using, 160–161
GoToBeginOfLine(), 159
GoToEndOfLine(), 159
GoToReverseBeginOfLine(), 159
IsAtEndOfLine(), 159
IsAtReverseEndOfLine(), 159
NextLine(), 158
PreviousLine(), 159
itk::ImageMaskSpatialObject, 128
itk::ImageRegionIterator, 154–156
396
example of using, 154–156
itk::ImageRegionIteratorWithIndex,
156–158
example of using, 157–158
itk::ImageSliceIteratorWithIndex, 162–166
itk::ImageSpatialObject, 127
itk::ImportImageFilter
Header, 56
Instantiation, 56, 57
New(), 57
Pointer, 57
SetRegion(), 57
itk::LandmarkSpatialObject, 130
itk::LineCell
Header, 74
header, 83, 90
Instantiation, 74, 78, 81, 84, 85, 90, 91
SetPointId(), 85, 91
itk::LineSpatialObject, 131
itk::MapContainer
InsertElement(), 62, 64
itk::Mesh, 36, 72
Accept(), 100, 103
AddVisitor(), 99, 103
BoundaryFeature, 88
Cell data, 77
CellInterfaceVisitorImplementation, 99,
102
CellAutoPointer, 75
CellFeatureCount, 89
CellInterfaceVisitor, 98–100, 102
CellIterator, 92, 96
CellsContainer, 87, 92, 96
CellsIterators, 87
CellType, 74
CellType casting, 77
CellVisitor, 98–100, 102
Dynamic, 72
GetBoundaryAssignment(), 89
GetCellData(), 79, 82, 83
GetCells(), 77, 87, 92, 96
GetNumberOfBoundaryFeatures(), 89
GetNumberOfCells(), 76
Index
GetNumberOfPoints(), 73
GetPoints(), 73, 86, 92
Header file, 72
Inserting cells, 76
Instantiation, 72, 78, 83, 90
Iterating cell data, 79, 83
Iterating cells, 77
K-Complex, 83, 93
MultiVisitor, 103
New(), 73, 75, 78, 81, 84, 91
PixelType, 78, 83, 90
Pointer, 78, 81, 84, 91
Pointer(), 73
PointIterator, 92
PointsContainer, 86, 92
PointsIterators, 86
PointType, 73, 75, 78, 81, 84, 91
PolyLine, 90
SetBoundaryAssignment(), 88
SetCell(), 76, 78, 82, 84, 91
SetPoint(), 73, 75, 78, 81, 84, 91
Static, 72
traits, 74
itk::MeshSpatialObject, 133
itk::PixelAccessor
performing computation, 199
with parameters, 197, 199
itk::PointSet, 59
data iterator, 67
Dynamic, 59
GetNumberOfPoints(), 60, 63
GetPoint(), 61
GetPointData(), 64, 65, 67, 69
GetPoints(), 62, 63, 67, 69
Instantiation, 59
iterating point data, 67
iterating points, 67
itk::CovariantVector, 70
New(), 60
PixelType, 63
PointDataContainer, 64
PointDataIterator, 71
Pointer, 60
Index
PointIterator, 69
points iterator, 67
PointsContainer, 61
PointType, 60
RGBPixel, 66
SetPoint(), 60, 66, 68, 70
SetPointData(), 64–66, 68, 70
SetPoints(), 62
Static, 59
Vector pixels, 68
itk::ReadWriteSpatialObject, 146
itk::RGBPixel, 54
GetBlue(), 54
GetGreen(), 54
GetRed(), 54
header, 54
Image, 54
Instantiation, 54, 66
itk::SceneSpatialObject, 144
itk::SpatialObjectToImageStatisticsCalculator,
147
itk::SpatialObjectHierarchy, 112
itk::SpatialObjectToImageFilter
Update(), 135
itk::SpatialObjectTransforms, 115
itk::SpatialObjectTreeContainer, 114
itk::SurfaceSpatialObject, 136
itk::TetrahedronCell
header, 83
Instantiation, 84
SetPointId(), 84
itk::TreeContainer, 105
itk::TriangleCell
header, 83
Instantiation, 84, 85
SetPointId(), 85
itk::TubeSpatialObject, 137
itk::Vector, 55
header, 55
Instantiation, 56
itk::Image, 56
itk::PointSet, 68
397
itk::VectorContainer
InsertElement(), 62, 64
itk::VertexCell
header, 83, 90
Instantiation, 84, 90
itk::VesselTubeSpatialObject, 139
LargestPossibleRegion, 206
LineCell
GetNumberOfPoints(), 77
Print(), 77
mapper, 37, 205
mesh region, 206
modified time, 206
module, 221
include, 224
src, 225
test, 225
third-party, 237
top level, 221
wrapping, 228
MultiVisitor, 103
Neighborhood iterators
active neighbors, 186
as stencils, 186
boundary conditions, 173
bounds checking, 173
construction of, 168
examples, 174
inactive neighbors, 186
radius of, 168
shaped, 186
NeighborhoodIterator
examples, 174
GetCenterPixel(), 171
GetImagePointer(), 168
GetIndex(), 172
GetNeighborhood(), 172
GetNeighborhoodIndex(), 173
GetNext(), 171
GetOffset(), 173
GetPixel(), 171
398
Index
GetPrevious(), 172
GetRadius(), 168
GetSlice(), 173
NeedToUseBoundaryConditionOff(),
173
NeedToUseBoundaryConditionOn(),
173
OverrideBoundaryCondition(), 173
ResetBoundaryCondition(), 174
SetCenterPixel(), 171
SetNeighborhood(), 172
SetNext(), 171
SetPixel(), 171, 174
SetPrevious(), 172
Size(), 171
NeighborhoodIterators, 171, 172
numerics, 35
PointIdIterator, 87, 92
PointIdsBegin(), 87, 92, 101
PointIdsEnd(), 87, 92, 101
PointsContainer
Begin(), 62, 73, 86, 92
End(), 63, 74, 86, 92
InsertElement(), 62
Iterator, 62, 63, 73, 74
New(), 61
Pointer, 61, 62
Size(), 63
Print(), 77
process object, 37, 205
ProgressEvent(), 32
Python, 39
object factory, 29
reader, 37
region, 205
RequestedRegion, 206
reverse iteration, 150, 153
pipeline
downstream, 206
execution details, 210
information, 206
modified time, 206
overview of execution, 208
PropagateRequestedRegion, 211
streaming large data, 207
ThreadedFilterExecution, 212
UpdateOutputData, 212
UpdateOutputInformation, 210
upstream, 206
PixelAccessor
RGB blue channel, 196
RGB green channel, 195
RGB red channel, 194
PointDataContainer
Begin(), 65
End(), 65
increment ++, 65
InsertElement(), 64
Iterator, 65
New(), 64
Pointer, 64
Quality Dashboard, 244
scene graph, 38
SetBoundaryAssignment()
itk::Mesh, 88
SetCell()
itk::Mesh, 76
ShapedNeighborhoodIterator, 186
ActivateOffset(), 186
ClearActiveList(), 186
DeactivateOffset(), 186
examples of, 187
GetActiveIndexListSize(), 186
Iterator::Begin(), 186
Iterator::End(), 186
smart pointer, 30
Sobel operator, 174, 177
source, 37, 205
spatial object, 38
streaming, 37
template, 28
TetrahedronCell
Index
GetNumberOfFaces(), 102
VNL, 35
399
wrapping, 39
The ITK Software Guide
Book 2: Design and Functionality
Fourth Edition
Updated for ITK version 5.0.0
Hans J. Johnson, Matthew M. McCormick, Luis Ibáñez,
and the Insight Software Consortium
January 18, 2019
https://itk.org
https://discourse.itk.org/
The purpose of computing is Insight, not numbers.
Richard Hamming
ABSTRACT
The National Library of Medicine Insight Segmentation and Registration Toolkit, shortened as the
Insight Toolkit (ITK), is an open-source software toolkit for performing registration and segmentation. Segmentation is the process of identifying and classifying data found in a digitally sampled
representation. Typically the sampled representation is an image acquired from such medical instrumentation as CT or MRI scanners. Registration is the task of aligning or developing correspondences
between data. For example, in the medical environment, a CT scan may be aligned with a MRI scan
in order to combine the information contained in both.
ITK is a cross-platform software. It uses a build environment known as CMake to manage platformspecific project generation and compilation process in a platform-independent way. ITK is implemented in C++. ITK’s implementation style employs generic programming, which involves the
use of templates to generate, at compile-time, code that can be applied generically to any class or
data-type that supports the operations used by the template. The use of C++ templating means that
the code is highly efficient and many issues are discovered at compile-time, rather than at run-time
during program execution. It also means that many of ITK’s algorithms can be applied to arbitrary
spatial dimensions and pixel types.
An automated wrapping system integrated with ITK generates an interface between C++ and a highlevel programming language Python. This enables rapid prototyping and faster exploration of ideas
by shortening the edit-compile-execute cycle. In addition to automated wrapping, the SimpleITK
project provides a streamlined interface to ITK that is available for C++, Python, Java, CSharp, R,
Tcl and Ruby.
Developers from around the world can use, debug, maintain, and extend the software because ITK
is an open-source project. ITK uses a model of software development known as Extreme Programming. Extreme Programming collapses the usual software development methodology into a simultaneous iterative process of design-implement-test-release. The key features of Extreme Programming
are communication and testing. Communication among the members of the ITK community is what
helps manage the rapid evolution of the software. Testing is what keeps the software stable. An
extensive testing process supported by the system known as CDash measures the quality of ITK
code on a daily basis. The ITK Testing Dashboard is updated continuously, reflecting the quality of
the code at any moment.
The
most
recent
version
of
this
document
is
available
online
at
https://itk.org/ItkSoftwareGuide.pdf.
This book is a guide for developing software
with ITK; it is the second of two companion books. This book covers detailed design and
functionality for reading and writing images, filtering, registration, segmentation, and performing
statistical analysis. The first book covers building and installation, general architecture and design,
as well as the process of contributing in the ITK community.
CONTRIBUTORS
The Insight Toolkit (ITK) has been created by the efforts of many talented individuals and prestigious organizations. It is also due in great part to the vision of the program established by Dr. Terry
Yoo and Dr. Michael Ackerman at the National Library of Medicine.
This book lists a few of these contributors in the following paragraphs. Not all developers of ITK are
credited here, so please visit the Web pages at https://itk.org/ITK/project/parti.html for the names of
additional contributors, as well as checking the GIT source logs for code contributions.
The following is a brief description of the contributors to this software guide and their contributions.
Luis Ibáñez is principal author of this text. He assisted in the design and layout of the text, implemented the bulk of the LATEX and CMake build process, and was responsible for the bulk of the
content. He also developed most of the example code found in the Insight/Examples directory.
Will Schroeder helped design and establish the organization of this text and the Insight/Examples
directory. He is principal content editor, and has authored several chapters.
Lydia Ng authored the description for the registration framework and its components, the section
on the multiresolution framework, and the section on deformable registration methods. She also
edited the section on the resampling image filter and the sections on various level set segmentation
algorithms.
Joshua Cates authored the iterators chapter and the text and examples describing watershed segmentation. He also co-authored the level-set segmentation material.
Jisung Kim authored the chapter on the statistics framework.
Julien Jomier contributed the chapter on spatial objects and examples on model-based registration
using spatial objects.
Karthik Krishnan reconfigured the process for automatically generating images from all the examples. Added a large number of new examples and updated the Filtering and Segmentation chapters
vi
for the second edition.
Stephen Aylward contributed material describing spatial objects and their application.
Tessa Sundaram contributed the section on deformable registration using the finite element method.
Mark Foskey contributed the examples on the itk::AutomaticTopologyMeshSource class.
Mathieu Malaterre contributed the entire section on the description and use of DICOM readers and
writers based on the GDCM library. He also contributed an example on the use of the VTKImageIO
class.
Gavin Baker contributed the section on how to write composite filters. Also known as minipipeline
filters.
Since the software guide is generated in part from the ITK source code itself, many ITK developers
have been involved in updating and extending the ITK documentation. These include David Doria,
Bradley Lowekamp, Mark Foskey, Gaëtan Lehmann, Andreas Schuh, Tom Vercauteren, Cory
Quammen, Daniel Blezek, Paul Hughett, Matthew McCormick, Josh Cates, Arnaud Gelas,
Jim Miller, Brad King, Gabe Hart, Hans Johnson.
Hans Johnson, Kent Williams, Constantine Zakkaroff, Xiaoxiao Liu, Ali Ghayoor, and
Matthew McCormick updated the documentation for the initial ITK Version 4 release.
Luis Ibáñez and Sébastien Barré designed the original Book 1 cover. Matthew McCormick and
Brad King updated the code to produce the Book 1 cover for ITK 4 and VTK 6. Xiaoxiao Liu, Bill
Lorensen, Luis Ibáñez, and Matthew McCormick created the 3D printed anatomical objects that
were photographed by Sébastien Barré for the Book 2 cover. Steve Jordan designed the layout of
the covers.
Lisa Avila, Hans Johnson, Matthew McCormick, Sandy McKenzie, Christopher Mullins,
Katie Osterdahl, and Michka Popoff prepared the book for the 4.7 print release.
CONTENTS
1 Reading and Writing Images
1
1.1
Basic Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2
Pluggable Factories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.3
Using ImageIO Classes Explicitly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.4
Reading and Writing RGB Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
1.5
Reading, Casting and Writing Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
1.6
Extracting Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10
1.7
Extracting Slices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
1.8
Reading and Writing Vector Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
1.8.1
The Minimal Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
1.8.2
Producing and Writing Covariant Images . . . . . . . . . . . . . . . . . . . . . . .
17
1.8.3
Reading Covariant Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
Reading and Writing Complex Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
1.9
1.10 Extracting Components from Vector Images
. . . . . . . . . . . . . . . . . . . . . . . . . .
23
1.11 Reading and Writing Image Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
1.11.1
Reading Image Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
1.11.2
Writing Image Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
1.11.3
Reading and Writing Series of RGB Images . . . . . . . . . . . . . . . . . . . . . .
29
1.12 Reading and Writing DICOM Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
1.12.1
Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
viii
CONTENTS
1.12.2
Reading and Writing a 2D Image . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
1.12.3
Reading a 2D DICOM Series and Writing a Volume . . . . . . . . . . . . . . . . . .
37
1.12.4
Reading a 2D DICOM Series and Writing a 2D DICOM Series . . . . . . . . . . . .
40
1.12.5
Printing DICOM Tags From One Slice . . . . . . . . . . . . . . . . . . . . . . . . .
44
1.12.6
Printing DICOM Tags From a Series . . . . . . . . . . . . . . . . . . . . . . . . . .
48
1.12.7
Changing a DICOM Header . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
2 Filtering
2.1
2.2
2.3
2.4
2.5
2.6
55
Thresholding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
2.1.1
Binary Thresholding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
2.1.2
General Thresholding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58
Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
61
2.2.1
Canny Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
61
Casting and Intensity Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
2.3.1
Linear Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
2.3.2
Non Linear Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
66
Gradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
2.4.1
Gradient Magnitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
2.4.2
Gradient Magnitude With Smoothing . . . . . . . . . . . . . . . . . . . . . . . . .
71
2.4.3
Derivative Without Smoothing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
Second Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
2.5.1
Second Order Recursive Gaussian . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
2.5.2
Laplacian Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
Laplacian Filter Recursive Gaussian . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
Neighborhood Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
84
2.6.1
Mean Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
2.6.2
Median Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
86
2.6.3
Mathematical Morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
Binary Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
Grayscale Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
91
Voting Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
93
Binary Median Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
94
2.6.4
CONTENTS
ix
Hole Filling Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
Iterative Hole Filling Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.7
Smoothing Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.7.1
Blurring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Discrete Gaussian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Binomial Blurring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Recursive Gaussian IIR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.7.2
Local Blurring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Gaussian Blur Image Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
2.7.3
Edge Preserving Smoothing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Introduction to Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Gradient Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Curvature Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Curvature Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
MinMaxCurvature Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Bilateral Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
2.7.4
Edge Preserving Smoothing in Vector Images . . . . . . . . . . . . . . . . . . . . . 125
Vector Gradient Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Vector Curvature Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . 126
2.7.5
Edge Preserving Smoothing in Color Images . . . . . . . . . . . . . . . . . . . . . 129
Gradient Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Curvature Anisotropic Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
2.8
Distance Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
2.9
Geometric Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.9.1
Filters You Should be Afraid to Use . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.9.2
Change Information Image Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.9.3
Flip Image Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
2.9.4
Resample Image Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Importance of Spacing and Origin . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
A Complete Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
x
CONTENTS
Rotating an Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Rotating and Scaling an Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Resampling using a deformation field . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Subsampling and image in the same space . . . . . . . . . . . . . . . . . . . . . . . . 161
Resampling an Anisotropic image to make it Isotropic . . . . . . . . . . . . . . . . . 164
2.10 Frequency Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
2.10.1
Computing a Fast Fourier Transform (FFT) . . . . . . . . . . . . . . . . . . . . . . 170
2.10.2
Filtering on the Frequency Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
2.11 Extracting Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
2.11.1
Surface extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3 Registration
179
3.1
Registration Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3.2
”Hello World” Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
3.3
Features of the Registration Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
3.4
Monitoring Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
3.5
Multi-Modality Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.5.1
3.6
3.7
3.8
3.9
Mattes Mutual Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Center Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
3.6.1
Rigid Registration in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
3.6.2
Initializing with Image Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
3.6.3
Similarity Transform in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
3.6.4
Rigid Transform in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
3.6.5
Centered Initialized Affine Transform . . . . . . . . . . . . . . . . . . . . . . . . . 230
Multi-Resolution Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
3.7.1
Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
3.7.2
Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Multi-Stage Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
3.8.1
Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
3.8.2
Cascaded Multistage Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
3.9.1
Geometrical Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
CONTENTS
xi
3.9.2
Transform General Properties
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
3.9.3
Identity Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
3.9.4
Translation Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
3.9.5
Scale Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
3.9.6
Scale Logarithmic Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
3.9.7
Euler2DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
3.9.8
CenteredRigid2DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
3.9.9
Similarity2DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
3.9.10
QuaternionRigidTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
3.9.11
VersorTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
3.9.12
VersorRigid3DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
3.9.13
Euler3DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
3.9.14
Similarity3DTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
3.9.15
Rigid3DPerspectiveTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
3.9.16
AffineTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
3.9.17
BSplineDeformableTransform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
3.9.18
KernelTransforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
3.10 Interpolators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
3.10.1
Nearest Neighbor Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
3.10.2
Linear Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
3.10.3
B-Spline Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
3.10.4
Windowed Sinc Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
3.11 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
3.11.1
Mean Squares Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Exploring a Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
3.11.2
Normalized Correlation Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
3.11.3
Mutual Information Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Parzen Windowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
Mattes et al. Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
3.11.4
Normalized Mutual Information Metric . . . . . . . . . . . . . . . . . . . . . . . . 289
3.11.5
Demons metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
xii
CONTENTS
3.11.6
ANTS neighborhood correlation metric . . . . . . . . . . . . . . . . . . . . . . . . 290
3.12 Optimizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
3.12.1
Registration using the One plus One Evolutionary Optimizer . . . . . . . . . . . . . 293
3.12.2
Registration using masks constructed with Spatial objects . . . . . . . . . . . . . . . 295
3.12.3
Rigid registrations incorporating prior knowledge . . . . . . . . . . . . . . . . . . . 297
3.13 Deformable Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
3.13.1
FEM-Based Image Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
3.13.2
BSplines Image Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
3.13.3
Level Set Motion for Deformable Registration . . . . . . . . . . . . . . . . . . . . . 305
3.13.4
BSplines Multi-Grid Image Registration . . . . . . . . . . . . . . . . . . . . . . . . 309
3.13.5
BSplines Multi-Grid Image Registration in 3D . . . . . . . . . . . . . . . . . . . . . 312
3.13.6
Image Warping with Kernel Splines . . . . . . . . . . . . . . . . . . . . . . . . . . 313
3.13.7
Image Warping with BSplines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
3.14 Demons Deformable Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
3.14.1
Asymmetrical Demons Deformable Registration . . . . . . . . . . . . . . . . . . . . 320
3.14.2
Symmetrical Demons Deformable Registration . . . . . . . . . . . . . . . . . . . . 323
3.15 Visualizing Deformation fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
3.15.1
Visualizing 2D deformation fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
3.15.2
Visualizing 3D deformation fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
3.16 Model Based Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
3.17 Point Set Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
3.17.1
Point Set Registration in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
3.17.2
Point Set Registration in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
3.17.3
Point Set to Distance Map Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
3.18 Registration Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
3.18.1
Too many samples outside moving image buffer . . . . . . . . . . . . . . . . . . . . 352
3.18.2
General heuristics for parameter fine-tunning . . . . . . . . . . . . . . . . . . . . . 352
4 Segmentation
4.1
355
Region Growing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
4.1.1
Connected Threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
4.1.2
Otsu Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
CONTENTS
xiii
4.1.3
Neighborhood Connected . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
4.1.4
Confidence Connected . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Application of the Confidence Connected filter on the Brain Web Data . . . . . . . . . 370
4.2
4.3
4.4
4.1.5
Isolated Connected . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
4.1.6
Confidence Connected in Vector Images . . . . . . . . . . . . . . . . . . . . . . . . 374
Segmentation Based on Watersheds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
4.2.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
4.2.2
Using the ITK Watershed Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Level Set Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
4.3.1
Fast Marching Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
4.3.2
Shape Detection Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
4.3.3
Geodesic Active Contours Segmentation . . . . . . . . . . . . . . . . . . . . . . . . 403
4.3.4
Threshold Level Set Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
4.3.5
Canny-Edge Level Set Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 411
4.3.6
Laplacian Level Set Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
4.3.7
Geodesic Active Contours Segmentation With Shape Guidance . . . . . . . . . . . . 419
Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
4.4.1
Hough Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
Line Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
Circle Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
5 Statistics
5.1
439
Data Containers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
5.1.1
Sample Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
5.1.2
Sample Adaptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
ImageToListSampleAdaptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
PointSetToListSampleAdaptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
5.1.3
Histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
5.1.4
Subsample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
5.1.5
MembershipSample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
5.1.6
MembershipSampleGenerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
5.1.7
K-d Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
xiv
CONTENTS
5.2
Algorithms and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
5.2.1
Sample Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
Mean and Covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
Weighted Mean and Covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
5.2.2
Sample Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
SampleToHistogramFilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
NeighborhoodSampler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
5.2.3
Sample Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
5.2.4
Probability Density Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
Gaussian Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
5.2.5
Distance Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Euclidean Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
5.2.6
Decision Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Maximum Decision Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Minimum Decision Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
Maximum Ratio Decision Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
5.2.7
Random Variable Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Normal (Gaussian) Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
5.3
Statistics applied to Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
5.3.1
Image Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Scalar Image Histogram with Adaptor . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Scalar Image Histogram with Generator . . . . . . . . . . . . . . . . . . . . . . . . . 485
Color Image Histogram with Generator . . . . . . . . . . . . . . . . . . . . . . . . . 487
Color Image Histogram Writing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
5.3.2
Image Information Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
Computing Image Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
Computing Images Mutual Information . . . . . . . . . . . . . . . . . . . . . . . . . 498
5.4
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
5.4.1
k-d Tree Based k-Means Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . 504
5.4.2
K-Means Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
5.4.3
Bayesian Plug-In Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
CONTENTS
xv
5.4.4
Expectation Maximization Mixture Model Estimation . . . . . . . . . . . . . . . . . 520
5.4.5
Classification using Markov Random Field . . . . . . . . . . . . . . . . . . . . . . 523
LIST OF FIGURES
1.1
Collaboration diagram of the ImageIO classes . . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.2
Use cases of ImageIO factories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.3
Class diagram of ImageIO factories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
2.1
BinaryThresholdImageFilter transfer function . . . . . . . . . . . . . . . . . . . . . . . . . .
56
2.2
BinaryThresholdImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58
2.3
ThresholdImageFilter using the threshold-below mode. . . . . . . . . . . . . . . . . . . . . .
59
2.4
ThresholdImageFilter using the threshold-above mode . . . . . . . . . . . . . . . . . . . . .
59
2.5
ThresholdImageFilter using the threshold-outside mode . . . . . . . . . . . . . . . . . . . . .
59
2.6
Sigmoid Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
2.7
Effect of the Sigmoid filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
2.8
GradientMagnitudeImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
71
2.9
GradientMagnitudeRecursiveGaussianImageFilter output . . . . . . . . . . . . . . . . . . . .
73
2.10 Effect of the Derivative filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
2.11 Output of the LaplacianRecursiveGaussianImageFilter. . . . . . . . . . . . . . . . . . . . . .
83
2.12 Effect of the MedianImageFilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
87
2.13 Effect of the Median filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
2.14 Effect of erosion and dilation in a binary image. . . . . . . . . . . . . . . . . . . . . . . . . .
91
2.15 Effect of erosion and dilation in a grayscale image. . . . . . . . . . . . . . . . . . . . . . . .
94
2.16 Effect of the BinaryMedian filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
xviii
List of Figures
2.17 Effect of many iterations on the BinaryMedian filter. . . . . . . . . . . . . . . . . . . . . . .
97
2.18 Effect of the VotingBinaryHoleFilling filter. . . . . . . . . . . . . . . . . . . . . . . . . . . .
99
2.19 Effect of the VotingBinaryIterativeHoleFilling filter. . . . . . . . . . . . . . . . . . . . . . . . 102
2.20 DiscreteGaussianImageFilter Gaussian diagram. . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.21 DiscreteGaussianImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2.22 BinomialBlurImageFilter output. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
2.23 Output of the SmoothingRecursiveGaussianImageFilter.
. . . . . . . . . . . . . . . . . . . . 109
2.24 GradientAnisotropicDiffusionImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . 114
2.25 CurvatureAnisotropicDiffusionImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . 116
2.26 CurvatureFlowImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
2.27 MinMaxCurvatureFlow computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
2.28 MinMaxCurvatureFlowImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
2.29 BilateralImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
2.30 VectorGradientAnisotropicDiffusionImageFilter output . . . . . . . . . . . . . . . . . . . . . 127
2.31 VectorCurvatureAnisotropicDiffusionImageFilter output . . . . . . . . . . . . . . . . . . . . 128
2.32 VectorGradientAnisotropicDiffusionImageFilter on RGB . . . . . . . . . . . . . . . . . . . . 131
2.33 VectorCurvatureAnisotropicDiffusionImageFilter output on RGB . . . . . . . . . . . . . . . . 133
2.34 Various Anisotropic Diffusion compared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
2.35 DanielssonDistanceMapImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
2.36 SignedDanielssonDistanceMapImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . 137
2.37 Effect of the FlipImageFilter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
2.38 Effect of the Resample filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.39 Analysis of resampling in common coordinate system . . . . . . . . . . . . . . . . . . . . . . 142
2.40 ResampleImageFilter with a translation by (−30, −50) . . . . . . . . . . . . . . . . . . . . . 143
2.41 ResampleImageFilter. Analysis of a translation by (−30, −50) . . . . . . . . . . . . . . . . . 144
2.42 ResampleImageFilter highlighting image borders . . . . . . . . . . . . . . . . . . . . . . . . 145
2.43 ResampleImageFilter selecting the origin of the output image . . . . . . . . . . . . . . . . . . 147
2.44 ResampleImageFilter origin in the output image . . . . . . . . . . . . . . . . . . . . . . . . . 147
2.45 ResampleImageFilter selecting the origin of the input image . . . . . . . . . . . . . . . . . . 148
2.46 ResampleImageFilter use of naive viewers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
2.47 ResampleImageFilter and output image spacing . . . . . . . . . . . . . . . . . . . . . . . . . 150
List of Figures
xix
2.48 ResampleImageFilter naive viewers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.49 ResampleImageFilter with non-unit spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
2.50 Effect of a rotation on the resampling filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
2.51 Input and output image placed in a common reference system . . . . . . . . . . . . . . . . . . 153
2.52 Effect of the Resample filter rotating an image . . . . . . . . . . . . . . . . . . . . . . . . . . 157
2.53 Effect of the Resample filter rotating and scaling an image . . . . . . . . . . . . . . . . . . . 159
3.1
Image Registration Concept
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3.2
A Typical Registration Framework Components . . . . . . . . . . . . . . . . . . . . . . . . . 180
3.3
Registration Framework Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
3.4
Fixed and Moving images in registration framework . . . . . . . . . . . . . . . . . . . . . . . 187
3.5
HelloWorld registration output images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.6
Pipeline structure of the registration example . . . . . . . . . . . . . . . . . . . . . . . . . . 190
3.7
Trace of translations and metrics during registration . . . . . . . . . . . . . . . . . . . . . . . 191
3.8
Registration Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
3.9
Command/Observer and the Registration Framework . . . . . . . . . . . . . . . . . . . . . . 198
3.10 Multi-Modality Registration Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
3.11 MattesMutualInformationImageToImageMetricv4 output images . . . . . . . . . . . . . . . . 204
3.12 MattesMutualInformationImageToImageMetricv4 output plots . . . . . . . . . . . . . . . . . 205
3.13 MattesMutualInformationImageToImageMetricv4 number of bins . . . . . . . . . . . . . . . 206
3.14 Rigid2D Registration input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
3.15 Rigid2D Registration output images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
3.16 Rigid2D Registration output plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
3.17 Rigid2D Registration input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
3.18 Rigid2D Registration output images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
3.19 Rigid2D Registration output plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
3.20 Effect of changing the center of rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.21 CenteredTransformInitializer input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.22 CenteredTransformInitializer output images . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3.23 CenteredTransformInitializer output plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
3.24 Fixed and Moving image registered with Simularity2DTransform . . . . . . . . . . . . . . . . 224
3.25 Output of the Simularity2DTransform registration . . . . . . . . . . . . . . . . . . . . . . . . 224
xx
List of Figures
3.26 Simularity2DTransform registration plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
3.27 CenteredTransformInitializer input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
3.28 CenteredTransformInitializer output images . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
3.29 CenteredTransformInitializer output plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
3.30 AffineTransform registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
3.31 AffineTransform output images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.32 AffineTransform output plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
3.33 Conceptual representation of Multi-Resolution registration . . . . . . . . . . . . . . . . . . . 237
3.34 Multi-Resolution registration input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
3.35 Multi-Resolution registration output images . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
3.36 AffineTransform registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
3.37 Multistage registration input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
3.38 Multistage registration input images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
3.39 Geometrical representation objects in ITK . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
3.40 Mapping moving image to fixed image in Registration . . . . . . . . . . . . . . . . . . . . . 277
3.41 Need for interpolation in Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
3.42 BSpline Interpolation Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
3.43 Parzen Windowing in Mutual Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
3.44 Mean Squares Metric Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
3.45 Class diagram of the Optimizers hierarchy in ITKv4 . . . . . . . . . . . . . . . . . . . . . . . 292
3.46 FEM-based deformable registration results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
3.47 Demon’s deformable registration output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
3.48 Demon’s deformable registration output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
3.49 Demon’s deformable registration output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
3.50 Deformation field magnitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
3.51 Calculator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
3.52 Visualized Def field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
3.53 Visualized Def field4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
3.54 Deformation field output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
3.55 Difference image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
3.56 Model to Image Registration Framework Components . . . . . . . . . . . . . . . . . . . . . . 334
List of Figures
xxi
3.57 Model to Image Registration Framework Concept . . . . . . . . . . . . . . . . . . . . . . . . 335
3.58 SpatialObject to Image Registration results
. . . . . . . . . . . . . . . . . . . . . . . . . . . 345
4.1
ConnectedThreshold segmentation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
4.2
OtsuThresholdImageFilter output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
4.3
NeighborhoodConnected segmentation results
4.4
ConfidenceConnected segmentation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
4.5
Whitematter Confidence Connected segmentation. . . . . . . . . . . . . . . . . . . . . . . . . 371
4.6
Axial, sagittal, and coronal slice of Confidence Connected segmentation. . . . . . . . . . . . . 371
4.7
IsolatedConnected segmentation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
4.8
VectorConfidenceConnected segmentation results . . . . . . . . . . . . . . . . . . . . . . . . 377
4.9
Watershed Catchment Basins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
. . . . . . . . . . . . . . . . . . . . . . . . . 366
4.10 Watersheds Hierarchy of Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
4.11 Watersheds filter composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
4.12 Watershed segmentation output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
4.13 Zero Set Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
4.14 Grid position of the embedded level-set surface. . . . . . . . . . . . . . . . . . . . . . . . . . 385
4.15 FastMarchingImageFilter collaboration diagram . . . . . . . . . . . . . . . . . . . . . . . . . 386
4.16 FastMarchingImageFilter intermediate output . . . . . . . . . . . . . . . . . . . . . . . . . . 393
4.17 FastMarchingImageFilter segmentations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
4.18 ShapeDetectionLevelSetImageFilter collaboration diagram . . . . . . . . . . . . . . . . . . . 395
4.19 ShapeDetectionLevelSetImageFilter intermediate output . . . . . . . . . . . . . . . . . . . . 402
4.20 ShapeDetectionLevelSetImageFilter segmentations . . . . . . . . . . . . . . . . . . . . . . . 403
4.21 GeodesicActiveContourLevelSetImageFilter collaboration diagram . . . . . . . . . . . . . . . 404
4.22 GeodesicActiveContourLevelSetImageFilter intermediate output . . . . . . . . . . . . . . . . 407
4.23 GeodesicActiveContourImageFilter segmentations . . . . . . . . . . . . . . . . . . . . . . . 408
4.24 ThresholdSegmentationLevelSetImageFilter collaboration diagram . . . . . . . . . . . . . . . 409
4.25 Propagation term for threshold-based level set segmentation . . . . . . . . . . . . . . . . . . 409
4.26 ThresholdSegmentationLevelSet segmentations . . . . . . . . . . . . . . . . . . . . . . . . . 412
4.27 CannySegmentationLevelSetImageFilter collaboration diagram . . . . . . . . . . . . . . . . . 413
4.28 Segmentation results of CannyLevelSetImageFilter . . . . . . . . . . . . . . . . . . . . . . . 415
4.29 LaplacianSegmentationLevelSetImageFilter collaboration diagram . . . . . . . . . . . . . . . 417
xxii
List of Figures
4.30 Segmentation results of LaplacianLevelSetImageFilter . . . . . . . . . . . . . . . . . . . . . 419
4.31 GeodesicActiveContourShapePriorLevelSetImageFilter collaboration diagram . . . . . . . . . 421
4.32 GeodesicActiveContourShapePriorImageFilter input image and initial model . . . . . . . . . 428
4.33 Corpus callosum PCA modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
4.34 GeodesicActiveContourShapePriorImageFilter segmentations . . . . . . . . . . . . . . . . . . 429
5.1
Sample class inheritance tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
5.2
Histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
5.3
Simple conceptual classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
5.4
Statistical classification framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
5.5
Two normal distributions plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
5.6
Output of the KMeans classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
5.7
Bayesian plug-in classifier for two Gaussian classes . . . . . . . . . . . . . . . . . . . . . . . 515
5.8
Output of the ScalarImageMarkovRandomField . . . . . . . . . . . . . . . . . . . . . . . . . 529
LIST OF TABLES
3.1
Geometrical Elementary Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
3.2
Identity Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
3.3
Translation Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
3.4
Scale Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
3.5
Scale Logarithmic Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 264
3.6
Euler2D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
3.7
CenteredRigid2D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
3.8
Similarity2D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
3.9
QuaternionRigid Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
3.10 Versor Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
3.11 Versor Rigid3D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
3.12 Euler3D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
3.13 Similarity3D Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
3.14 Rigid3DPerspective Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 273
3.15 Affine Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
3.16 BSpline Deformable Transform Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 275
3.17 LBFGS Optimizer trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
4.1
ConnectedThreshold example parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
4.2
IsolatedConnectedImageFilter example parameters . . . . . . . . . . . . . . . . . . . . . . . 374
xxiv
List of Tables
4.3
FastMarching segmentation example parameters . . . . . . . . . . . . . . . . . . . . . . . . . 392
4.4
ShapeDetection example parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
4.5
GeodesicActiveContour segmentation example parameters . . . . . . . . . . . . . . . . . . . 406
4.6
ThresholdSegmentationLevelSet segmentation parameters . . . . . . . . . . . . . . . . . . . 411
CHAPTER
ONE
READING AND WRITING IMAGES
This chapter describes the toolkit architecture supporting reading and writing of images to files. ITK
does not enforce any particular file format, instead, it provides a structure supporting a variety of
formats that can be easily extended by the user as new formats become available.
We begin the chapter with some simple examples of file I/O.
1.1 Basic Example
The source code for this section can be found in the file
ImageReadWrite.cxx.
The classes responsible for reading and writing images are located at the beginning and end of the
data processing pipeline. These classes are known as data sources (readers) and data sinks (writers).
Generally speaking they are referred to as filters, although readers have no pipeline input and writers
have no pipeline output.
The reading of images is managed by the class itk::ImageFileReader while writing is performed
by the class itk::ImageFileWriter. These two classes are independent of any particular file
format. The actual low level task of reading and writing specific file formats is done behind the
注意：由其来解析各种图像格式
scenes by a family of classes of type itk::ImageIO.
The first step for performing reading and writing is to include the following headers.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
Then, as usual, a decision must be made about the type of pixel used to represent the image processed
by the pipeline. Note that when reading and writing images, the pixel type of the image is not
necessarily the same as the pixel type stored in the file. Your choice of the pixel type (and hence
注意：图像的像素类型不一定必要与
template parameter) should be driven mainly by two considerations:
文件中的像素类型相同
2
Chapter 1. Reading and Writing Images
• It should be possible to cast the pixel type in the file to the pixel type you select. This casting
will be performed using the standard C-language rules, so you will have to make sure that the
conversion does not result in information being lost.
• The pixel type in memory should be appropriate to the type of processing you intend to apply
on the images.
A typical selection for medical images is illustrated in the following lines.
using PixelType = short;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
Note that the dimension of the image in memory should match that of the image in the file. There
are a couple of special cases in which this condition may be relaxed, but in general it is better to
ensure that both dimensions match.
We can now instantiate the types of the reader and writer. These two classes are parameterized over
the image type.
using ReaderType = itk::ImageFileReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
Then, we create one object of each type using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
The name of the file to be read or written is passed to the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
We can now connect these readers and writers to filters to create a pipeline. For example, we can
create a short pipeline by passing the output of the reader directly to the input of the writer.
writer->SetInput( reader->GetOutput() );
At first glance this may look like a quite useless program, but it is actually implementing a powerful
file format conversion tool! The execution of the pipeline is triggered by the invocation of the
Update() methods in one of the final objects. In this case, the final data pipeline object is the writer.
It is a wise practice of defensive programming to insert any Update() call inside a try/catch block
in case exceptions are thrown during the execution of the pipeline.
注意：
1.1. Basic Example
3
1
1
ImageIO
ImageFileReader
ImageFileWriter
CanReadFile():bool
CanWriteFile():bool
1
PNGImageIO
MetaImageIO
VTKImageIO
1
DicomImageIO
GiplImageIO
RawImageIO
VOLImageIO
Figure 1.1: Collaboration diagram of the ImageIO classes.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
Note that exceptions should only be caught by pieces of code that know what to do with them. In
a typical application this catch block should probably reside in the GUI code. The action on the
注意：可以不用显示指定文件格式
catch block could inform the user about the failure of the IO operation.
The IO architecture of the toolkit makes it possible to avoid explicit specification of the file format
used to read or write images.1 The object factory mechanism enables the ImageFileReader and
ImageFileWriter to determine (at run-time) which file format it is working with. Typically, file
formats are chosen based on the filename extension, but the architecture supports arbitrarily complex
processes to determine whether a file can be read or written. Alternatively, the user can specify the
data file format by explicit instantiation and assignment of the appropriate itk::ImageIO subclass.
For historical reasons and as a convenience to the user, the itk::ImageFileWriter also has a
Write() method that is aliased to the Update() method. You can in principle use either of them
but Update() is recommended since Write() may be deprecated in the future.
To better understand the IO architecture, please refer to Figures 1.1, 1.2, and 1.3.
The following section describes the internals of the IO architecture provided in the toolkit.
1 In
this example no file format is specified; this program can be used as a general file conversion utility.
注意：实验
中必须指定
IO格式？是
编译时没设
置好？？？
4
Chapter 1. Reading and Writing Images
Pluggable Factories
Pluggable Factories
e
nam
file
CreateImageIO
for Reading
Register
PNGImageIOFactory
CanRead ?
filename
e
am
en
fil
ImageFileReader
CreateImageIO
for Writing
ImageIOFactory
filena
me
CanWrite ?
ImageFileWriter
MetaImageIOFactory
Figure 1.2: Use cases of ImageIO factories.
注意：创建IO--选择合适的IO
factory--执行
ImageIOFactory
ObjectFactoryBase
1
RegisterFactory(factory:ObjectFactoryBase)
PNGImageIOFactory
VTKImageIOFactory
MetaImageIOFactory
GiplImageIOFactory
DicomImageIOFactory
VOLImageIOFactory
GDCMImageIOFactory
*
RegisterBuiltInFactories()
CreateImageIO(string)
RawImageIOFactory
AnalyzeImageIOFactory
Ge4xImageIOFactory
BMPImageIOFactory
StimulateImageIOFactory
MetaImageIOFactory
TIFFImageIOFactory
Figure 1.3: Class diagram of the ImageIO factories.
JPEGImageIOFactory
NrrdImageIOFactory
SiemensVisionIOFactory
1.2. Pluggable Factories
5
1.2 Pluggable Factories
The principle behind the input/output mechanism used in ITK is known as pluggable-factories
[20]. This concept is illustrated in the UML diagram in Figure 1.1. From the user’s point of
view the objects responsible for reading and writing files are the itk::ImageFileReader and
itk::ImageFileWriter classes. These two classes, however, are not aware of the details involved
in reading or writing particular file formats like PNG or DICOM. What they do is dispatch the user’s
requests to a set of specific classes that are aware of the details of image file formats. These classes
are the itk::ImageIO classes. The ITK delegation mechanism enables users to extend the number
注意：可以扩展ITK对其他格式的支持
of supported file formats by just adding new classes to the ImageIO hierarchy.
Each instance of ImageFileReader and ImageFileWriter has a pointer to an ImageIO object. If this
pointer is empty, it will be impossible to read or write an image and the image file reader/writer
must determine which ImageIO class to use to perform IO operations. This is done basically by
passing the filename to a centralized class, the itk::ImageIOFactory and asking it to identify any
subclass of ImageIO capable of reading or writing the user-specified file. This is illustrated by the
use cases on the right side of Figure 1.2. The ImageIOFactory acts here as a dispatcher that helps
注意：工厂模式？？？
locate the actual IO factory classes corresponding to each file format.
Each class derived from ImageIO must provide an associated factory class capable of producing an
instance of the ImageIO class. For example, for PNG files, there is a itk::PNGImageIO object
that knows how to read this image files and there is a itk::PNGImageIOFactory class capable
of constructing a PNGImageIO object and returning a pointer to it. Each time a new file format is
added (i.e., a new ImageIO subclass is created), a factory must be implemented as a derived class of
the ObjectFactoryBase class as illustrated in Figure 1.3.
For example, in order to read PNG files, a PNGImageIOFactory is created and registered with the
central ImageIOFactory singleton2 class as illustrated in the left side of Figure 1.2. When the ImageFileReader asks the ImageIOFactory for an ImageIO capable of reading the file identified with
filename the ImageIOFactory will iterate over the list of registered factories and will ask each one of
them if they know how to read the file. The factory that responds affirmatively will be used to create
the specific ImageIO instance that will be returned to the ImageFileReader and used to perform the
read operations.
In most cases the mechanism is transparent to the user who only interacts with the ImageFileReader
and ImageFileWriter. It is possible, however, to explicitly select the type of ImageIO object to use.
注意：显示选择ImageIO对象类型
This is illustrated by the following example.
1.3 Using ImageIO Classes Explicitly
The source code for this section can be found in the file
ImageReadExportVTK.cxx.
2 Singleton
means that there is only one instance of this class in a particular application
6
Chapter 1. Reading and Writing Images
In cases where the user knows what file format to use and wants to indicate this explicitly, a specific
itk::ImageIO class can be instantiated and assigned to the image file reader or writer. This circumvents the itk::ImageIOFactory mechanism which tries to find the appropriate ImageIO class
for performing the IO operations. Explicit selection of the ImageIO also allows the user to invoke
specialized features of a particular class which may not be available from the general API provided
by ImageIO.
The following example illustrates explicit instantiation of an IO class (in this case a VTK file format),
setting its parameters and then connecting it to the itk::ImageFileWriter.
The example begins by including the appropriate headers.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkVTKImageIO.h"
Then, as usual, we select the pixel types and the image dimension. Remember, if the file format
represents pixels with a particular type, C-style casting will be performed to convert the data.
using PixelType = unsigned short;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
We can now instantiate the reader and writer. These two classes are parameterized over the image
type. We instantiate the itk::VTKImageIO class as well. Note that the ImageIO objects are not
templated.
注意：是处理特定类型的类
using ReaderType = itk::ImageFileReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
using ImageIOType = itk::VTKImageIO;
Then, we create one object of each type using the New() method and assigning the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
ImageIOType::Pointer vtkIO = ImageIOType::New();
The name of the file to be read or written is passed with the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
We can now connect these readers and writers to filters in a pipeline. For example, we can create a
short pipeline by passing the output of the reader directly to the input of the writer.
1.4. Reading and Writing RGB Images
7
writer->SetInput( reader->GetOutput() );
Explicitly declaring the specific VTKImageIO allow users to invoke methods specific to a particular
IO class. For example, the following line specifies to the writer to use ASCII format when writing
the pixel data.
vtkIO->SetFileTypeToASCII();
The VTKImageIO object is then connected to the ImageFileWriter. This will short-circuit the action
of the ImageIOFactory mechanism. The ImageFileWriter will not attempt to look for other ImageIO
objects capable of performing the writing tasks. It will simply invoke the one provided by the user.
writer->SetImageIO( vtkIO );
Finally we invoke Update() on the ImageFileWriter and place this call inside a try/catch block in
case any errors occur during the writing process.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
Although this example only illustrates how to use an explicit ImageIO class with the ImageFileWriter, the same can be done with the ImageFileReader. The typical case in which this is
done is when reading raw image files with the itk::RawImageIO object. The drawback of this
approach is that the parameters of the image have to be explicitly written in the code. The direct use
of raw files is strongly discouraged in medical imaging. It is always better to create a header for
a raw file by using any of the file formats that combine a text header file and a raw binary file, like
itk::MetaImageIO, itk::GiplImageIO and itk::VTKImageIO.
注意：在医学图像中，强烈不推荐直
接使用raw文件
1.4 Reading and Writing RGB Images
The source code for this section can be found in the file
RGBImageReadWrite.cxx.
8
Chapter 1. Reading and Writing Images
RGB images are commonly used for representing data acquired from cryogenic sections, optical
microscopy and endoscopy. This example illustrates how to read and write RGB color images to
and from a file. This requires the following headers as shown.
#include
#include
#include
#include
"itkRGBPixel.h"
"itkImage.h"
"itkImageFileReader.h"
"itkImageFileWriter.h"
The itk::RGBPixel class is templated over the type used to represent each one of the red, green
and blue components. A typical instantiation of the RGB image class might be as follows.
using PixelType = itk::RGBPixel< unsigned char >;
using ImageType = itk::Image< PixelType, 2 >;
The image type is used as a template parameter to instantiate the reader and writer.
using ReaderType = itk::ImageFileReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
The filenames of the input and output files must be provided to the reader and writer respectively.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Finally, execution of the pipeline can be triggered by invoking the Update() method in the writer.
writer->Update();
You may have noticed that apart from the declaration of the PixelType there is nothing in this code
specific to RGB images. All the actions required to support color images are implemented internally
in the itk::ImageIO objects.
1.5 Reading, Casting and Writing Images
The source code for this section can be found in the file
ImageReadCastWrite.cxx.
Given that ITK is based on the Generic Programming paradigm, most of the types are defined at
compilation time. It is sometimes important to anticipate conversion between different types of
1.5. Reading, Casting and Writing Images
9
images. The following example illustrates the common case of reading an image of one pixel type
and writing it as a different pixel type. This process not only involves casting but also rescaling the
image intensity since the dynamic range of the input and output pixel types can be quite different.
The itk::RescaleIntensityImageFilter is used here to linearly rescale the image values.
The first step in this example is to include the appropriate headers.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkRescaleIntensityImageFilter.h"
Then, as usual, a decision should be made about the pixel type that should be used to represent the
images. Note that when reading an image, this pixel type is not necessarily the pixel type of the
image stored in the file. Instead, it is the type that will be used to store the image as soon as it is read
into memory.
using InputPixelType = float;
using OutputPixelType = unsigned char;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
Note that the dimension of the image in memory should match the one of the image in the file. There
are a couple of special cases in which this condition may be relaxed, but in general it is better to
ensure that both dimensions match.
We can now instantiate the types of the reader and writer. These two classes are parameterized over
the image type.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
Below we instantiate the RescaleIntensityImageFilter class that will linearly scale the image intensities.
using FilterType = itk::RescaleIntensityImageFilter<
InputImageType,
OutputImageType >;
A filter object is constructed and the minimum and maximum values of the output are selected using
the SetOutputMinimum() and SetOutputMaximum() methods.
10
Chapter 1. Reading and Writing Images
FilterType::Pointer filter = FilterType::New();
filter->SetOutputMinimum( 0 );
filter->SetOutputMaximum( 255 );
Then, we create the reader and writer and connect the pipeline.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
The name of the files to be read and written are passed with the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Finally we trigger the execution of the pipeline with the Update() method on the writer. The output
image will then be the scaled and cast version of the input image.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.6 Extracting Regions
The source code for this section can be found in the file
ImageReadRegionOfInterestWrite.cxx.
This example should arguably be placed in the previous filtering chapter. However its usefulness for
typical IO operations makes it interesting to mention here. The purpose of this example is to read an
image, extract a subregion and write this subregion to a file. This is a common task when we want
to apply a computationally intensive method to the region of interest of an image.
As usual with ITK IO, we begin by including the appropriate header files.
1.6. Extracting Regions
11
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
The itk::RegionOfInterestImageFilter is the filter used to extract a region from an image. Its
header is included below.
#include "itkRegionOfInterestImageFilter.h"
注意：
Image types are defined below.
using InputPixelType = signed short;
using OutputPixelType = signed short;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
The types for the itk::ImageFileReader and itk::ImageFileWriter are instantiated using the
image types.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The RegionOfInterestImageFilter type is instantiated using the input and output image types. A
filter object is created with the New() method and assigned to a itk::SmartPointer.
using FilterType = itk::RegionOfInterestImageFilter< InputImageType,
OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The RegionOfInterestImageFilter requires a region to be defined by the user. The region is specified
by an itk::Index indicating the pixel where the region starts and an itk::Size indicating how
many pixels the region has along each dimension. In this example, the specification of the region is
taken from the command line arguments (this example assumes that a 2D image is being processed).
OutputImageType::IndexType start;
start[0] = std::stoi( argv[3] );
start[1] = std::stoi( argv[4] );
12
Chapter 1. Reading and Writing Images
OutputImageType::SizeType size;
size[0] = std::stoi( argv[5] );
size[1] = std::stoi( argv[6] );
An itk::ImageRegion object is created and initialized with start and size obtained from the command line.
OutputImageType::RegionType desiredRegion;
desiredRegion.SetSize( size );
desiredRegion.SetIndex( start );
Then the region is passed to the filter using the SetRegionOfInterest() method.
filter->SetRegionOfInterest( desiredRegion );
Below, we create the reader and writer using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
The name of the file to be read or written is passed with the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Below we connect the reader, filter and writer to form the data processing pipeline.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
Finally we execute the pipeline by invoking Update() on the writer. The call is placed in a
try/catch block in case exceptions are thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.7. Extracting Slices
13
1.7 Extracting Slices
The source code for this section can be found in the file
ImageReadExtractWrite.cxx.
注意：从三维图像中提取一个切片，
比较常用
This example illustrates the common task of extracting a 2D slice from a 3D volume. This is typically used for display purposes and for expediting user feedback in interactive programs. Here we
simply read a 3D volume, extract one of its slices and save it as a 2D image. Note that caution
should be used when working with 2D slices from a 3D dataset, since for most image processing
operations, the application of a filter on an extracted slice is not equivalent to first applying the filter
注意：体上filter + 提取切片 ！=
in the volume and then extracting the slice.
提取切片 + 在切片上运用filter
In this example we start by including the appropriate header files.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
The filter used to extract a region from an image is the itk::ExtractImageFilter. Its header is
included below. This filter is capable of extracting (N − 1)-dimensional images from N-dimensional
注意：
ones.
#include "itkExtractImageFilter.h"
Image types are defined below. Note that the input image type is 3D and the output image type is
2D.
using InputPixelType = signed short;
using OutputPixelType = signed short;
using InputImageType = itk::Image< InputPixelType, 3 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The types for the itk::ImageFileReader and itk::ImageFileWriter are instantiated using the
image types.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
Below, we create the reader and writer using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
14
Chapter 1. Reading and Writing Images
The name of the file to be read or written is passed with the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
The ExtractImageFilter type is instantiated using the input and output image types. A filter object is
created with the New() method and assigned to a itk::SmartPointer.
using FilterType = itk::ExtractImageFilter< InputImageType,
OutputImageType >;
FilterType::Pointer filter = FilterType::New();
filter->InPlaceOn();
filter->SetDirectionCollapseToSubmatrix();
The ExtractImageFilter requires a region to be defined by the user. The region is specified by an
itk::Index indicating the pixel where the region starts and an itk::Size indicating how many
pixels the region has along each dimension. In order to extract a 2D image from a 3D data set, it is
enough to set the size of the region to 0 in one dimension. This will indicate to ExtractImageFilter
that a dimensional reduction has been specified. Here we take the region from the largest possible
region of the input image. Note that UpdateOutputInformation() is being called first on the
reader. This method updates the metadata in the output image without actually reading in the bulkdata.
reader->UpdateOutputInformation();
InputImageType::RegionType inputRegion =
reader->GetOutput()->GetLargestPossibleRegion();
We take the size from the region and collapse the size in the Z component by setting its value to 0.
This will indicate to the ExtractImageFilter that the output image should have a dimension less than
the input image.
InputImageType::SizeType size = inputRegion.GetSize();
size[2] = 0;
Note that in this case we are extracting a Z slice, and for that reason, the dimension to be collapsed
is the one with index 2. You may keep in mind the association of index components {X = 0,Y =
1, Z = 2}. If we were interested in extracting a slice perpendicular to the Y axis we would have set
size[1]=0;.
Then, we take the index from the region and set its Z value to the slice number we want to extract.
In this example we obtain the slice number from the command line arguments.
InputImageType::IndexType start = inputRegion.GetIndex();
const unsigned int sliceNumber = std::stoi( argv[3] );
start[2] = sliceNumber;
1.8. Reading and Writing Vector Images
15
Finally, an itk::ImageRegion object is created and initialized with the start and size we just
prepared using the slice information.
InputImageType::RegionType desiredRegion;
desiredRegion.SetSize( size );
desiredRegion.SetIndex( start );
Then the region is passed to the filter using the SetExtractionRegion() method.
filter->SetExtractionRegion( desiredRegion );
Below we connect the reader, filter and writer to form the data processing pipeline.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
Finally we execute the pipeline by invoking Update() on the writer. The call is placed in a
try/catch block in case exceptions are thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.8 Reading and Writing Vector Images
Images whose pixel type is a Vector, a CovariantVector, an Array, or a Complex are quite common
in image processing. It is convenient then to describe rapidly how those images can be saved into
files and how they can be read from those files later on.
1.8.1 The Minimal Example
The source code for this section can be found in the file
VectorImageReadWrite.cxx.
16
Chapter 1. Reading and Writing Images
This example illustrates how to read and write an image of pixel type itk::Vector.
We should include the header files for the Image, the ImageFileReader and the ImageFileWriter.
#include "itkImage.h"
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
Then we define the specific type of vector to be used as pixel type.
constexpr unsigned int VectorDimension = 3;
using PixelType = itk::Vector< float, VectorDimension >;
We define the image dimension, and along with the pixel type we use it for fully instantiating the
image type.
constexpr unsigned int ImageDimension = 2;
using ImageType = itk::Image< PixelType, ImageDimension >;
Having the image type at hand, we can instantiate the reader and writer types, and use them for
creating one object of each type.
using ReaderType = itk::ImageFileReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
A filename must be provided to both the reader and the writer. In this particular case we take those
filenames from the command line arguments.
reader->SetFileName( argv[1] );
writer->SetFileName( argv[2] );
This being a minimal example, we create a short pipeline where we simply connect the output of the
reader to the input of the writer.
writer->SetInput( reader->GetOutput() );
The execution of this short pipeline is triggered by invoking the writer’s Update() method. This
invocation must be placed inside a try/catch block since its execution may result in exceptions
being thrown.
1.8. Reading and Writing Vector Images
17
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
Of course, you could envision the addition of filters in between the reader and the writer. Those
filters could perform operations on the vector image.
1.8.2 Producing and Writing Covariant Images
The source code for this section can be found in the file
CovariantVectorImageWrite.cxx.
This example illustrates how to write an image whose pixel type is CovariantVector. For practical purposes all the content in this example is applicable to images of pixel type itk::Vector,
itk::Point and itk::FixedArray. These pixel types are similar in that they are all arrays of
fixed size in which the components have the same representational type.
In order to make this example a bit more interesting we setup a pipeline to read an image, compute its gradient and write the gradient to a file. Gradients are represented with
itk::CovariantVectors as opposed to Vectors. In this way, gradients are transformed correctly
under itk::AffineTransforms or in general, any transform having anisotropic scaling.
Let’s start by including the relevant header files.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
注意：求梯度的滤波器
We use the itk::GradientRecursiveGaussianImageFilter in order to compute the image gradient. The output of this filter is an image whose pixels are CovariantVectors.
#include "itkGradientRecursiveGaussianImageFilter.h"
We read an image of signed short pixels and compute the gradient to produce an image of CovariantVectors where each component is of type float.
using InputPixelType = signed short;
using ComponentType = float;
constexpr unsigned int Dimension = 2;
18
Chapter 1. Reading and Writing Images
using OutputPixelType = itk::CovariantVector< ComponentType,
Dimension >;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
The itk::ImageFileReader and itk::ImageFileWriter are instantiated using the image types.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The GradientRecursiveGaussianImageFilter class is instantiated using the input and output image
types. A filter object is created with the New() method and assigned to a itk::SmartPointer.
using FilterType = itk::GradientRecursiveGaussianImageFilter<
InputImageType,
OutputImageType
>;
FilterType::Pointer filter = FilterType::New();
We select a value for the σ parameter of the GradientRecursiveGaussianImageFilter. Note that σ for
this filter is specified in millimeters.
filter->SetSigma( 1.5 );
// Sigma in millimeters
Below, we create the reader and writer using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
The name of the file to be read or written is passed to the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Below we connect the reader, filter and writer to form the data processing pipeline.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
Finally we execute the pipeline by invoking Update() on the writer. The call is placed in a
try/catch block in case exceptions are thrown.
1.8. Reading and Writing Vector Images
19
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.8.3 Reading Covariant Images
Let’s now take the image that we just created and read it into another program.
The source code for this section can be found in the file
CovariantVectorImageRead.cxx.
This example illustrates how to read an image whose pixel type is CovariantVector. For practical purposes this example is applicable to images of pixel type itk::Vector, itk::Point and
itk::FixedArray. These pixel types are similar in that they are all arrays of fixed size in which
the components have the same representation type.
In this example we are reading a gradient image from a file (written in the previous example) and
computing its magnitude using the itk::VectorMagnitudeImageFilter. Note that this filter is
different from the itk::GradientMagnitudeImageFilter which actually takes a scalar image as
input and computes the magnitude of its gradient. The VectorMagnitudeImageFilter class takes an
image of vector pixel type as input and computes pixel-wise the magnitude of each vector.
Let’s start by including the relevant header files.
#include
#include
#include
#include
"itkImageFileReader.h"
"itkImageFileWriter.h"
"itkVectorMagnitudeImageFilter.h"
"itkRescaleIntensityImageFilter.h"
We read an image of itk::CovariantVector pixels and compute pixel magnitude to produce an image where each pixel is of type unsigned short. The components of the CovariantVector are selected to be float here. Notice that a renormalization is required in order to
map the dynamic range of the magnitude values into the range of the output pixel type. The
itk::RescaleIntensityImageFilter is used to achieve this.
using ComponentType = float;
constexpr unsigned int Dimension = 2;
20
Chapter 1. Reading and Writing Images
using InputPixelType = itk::CovariantVector< ComponentType,
Dimension >;
using MagnitudePixelType = float;
using OutputPixelType = unsigned short;
using InputImageType = itk::Image< InputPixelType,
Dimension >;
using MagnitudeImageType = itk::Image< MagnitudePixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType,
Dimension >;
The itk::ImageFileReader and itk::ImageFileWriter are instantiated using the image types.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The VectorMagnitudeImageFilter is instantiated using the input and output image types. A filter
object is created with the New() method and assigned to a itk::SmartPointer.
using FilterType = itk::VectorMagnitudeImageFilter<
InputImageType,
MagnitudeImageType
>;
FilterType::Pointer filter = FilterType::New();
The RescaleIntensityImageFilter class is instantiated next.
using RescaleFilterType = itk::RescaleIntensityImageFilter<
MagnitudeImageType,
OutputImageType >;
RescaleFilterType::Pointer rescaler = RescaleFilterType::New();
In the following the minimum and maximum values for the output image are specified. Note the use
of the itk::NumericTraits class which allows us to define a number of type-related constants in
a generic way. The use of traits is a fundamental characteristic of generic programming [5, 1].
rescaler->SetOutputMinimum( itk::NumericTraits< OutputPixelType >::min() );
rescaler->SetOutputMaximum( itk::NumericTraits< OutputPixelType >::max() );
Below, we create the reader and writer using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
1.9. Reading and Writing Complex Images
21
The name of the file to be read or written is passed with the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Below we connect the reader, filter and writer to form the data processing pipeline.
filter->SetInput( reader->GetOutput() );
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
Finally we execute the pipeline by invoking Update() on the writer. The call is placed in a
try/catch block in case exceptions are thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.9 Reading and Writing Complex Images
The source code for this section can be found in the file
ComplexImageReadWrite.cxx.
This example illustrates how to read and write an image of pixel type std::complex. The complex
type is defined as an integral part of the C++ language. The characteristics of the type are specified
in the C++ standard document in Chapter 26 ”Numerics Library”, page 565, in particular in section
26.2 [4].
We start by including the headers of the complex class, the image, and the reader and writer classes.
#include
#include
#include
#include
"itkImage.h"
"itkImageFileReader.h"
"itkImageFileWriter.h"
The image dimension and pixel type must be declared. In this case we use the std::complex<> as
the pixel type. Using the dimension and pixel type we proceed to instantiate the image type.
22
Chapter 1. Reading and Writing Images
constexpr unsigned int Dimension = 2;
using PixelType = std::complex< float >;
using ImageType = itk::Image< PixelType, Dimension >;
The image file reader and writer types are instantiated using the image type. We can then create
objects for both of them.
using ReaderType = itk::ImageFileReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
File names should be provided for both the reader and the writer. In this particular example we take
those file names from the command line arguments.
reader->SetFileName( argv[1] );
writer->SetFileName( argv[2] );
Here we simply connect the output of the reader as input to the writer. This simple program could
be used for converting complex images from one file format to another.
writer->SetInput( reader->GetOutput() );
The execution of this short pipeline is triggered by invoking the Update() method of the writer.
This invocation must be placed inside a try/catch block since its execution may result in exceptions
being thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
For a more interesting use of this code, you may want to add a filter in between the reader and the
writer and perform any complex image to complex image operation. A practical application of this
code is presented in section 2.10 in the context of Fourier analysis.
1.10. Extracting Components from Vector Images
23
1.10 Extracting Components from Vector Images
The source code for this section can be found in the file
CovariantVectorImageExtractComponent.cxx.
This example illustrates how to read an image whose pixel type is CovariantVector, extract one
of its components to form a scalar image and finally save this image into a file.
The itk::VectorIndexSelectionCastImageFilter is used to extract a scalar from the vector
image. It is also possible to cast the component type when using this filter. It is the user’s responsibility to make sure that the cast will not result in any information loss.
Let’s start by including the relevant header files.
#include
#include
#include
#include
"itkImageFileReader.h"
"itkImageFileWriter.h"
"itkVectorIndexSelectionCastImageFilter.h"
"itkRescaleIntensityImageFilter.h"
We read an image of itk::CovariantVector pixels and extract one of its components to generate
a scalar image of a consistent pixel type. Then, we rescale the intensities of this scalar image and
write it as an image of unsigned short pixels.
using ComponentType = float;
constexpr unsigned int Dimension = 2;
using InputPixelType = itk::CovariantVector< ComponentType,
Dimension >;
using OutputPixelType = unsigned short;
using InputImageType = itk::Image< InputPixelType,
using ComponentImageType = itk::Image< ComponentType,
using OutputImageType = itk::Image< OutputPixelType,
Dimension >;
Dimension >;
Dimension >;
The itk::ImageFileReader and itk::ImageFileWriter are instantiated using the image types.
using ReaderType = itk::ImageFileReader< InputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The VectorIndexSelectionCastImageFilter is instantiated using the input and output image types. A
filter object is created with the New() method and assigned to a itk::SmartPointer.
using FilterType = itk::VectorIndexSelectionCastImageFilter<
InputImageType,
ComponentImageType
>;
FilterType::Pointer componentExtractor = FilterType::New();
24
Chapter 1. Reading and Writing Images
The VectorIndexSelectionCastImageFilter class requires us to specify which of the vector components is to be extracted from the vector image. This is done with the SetIndex() method. In this
example we obtain this value from the command line arguments.
componentExtractor->SetIndex( indexOfComponentToExtract );
The itk::RescaleIntensityImageFilter filter is instantiated here.
using RescaleFilterType = itk::RescaleIntensityImageFilter<
ComponentImageType,
OutputImageType >;
RescaleFilterType::Pointer rescaler = RescaleFilterType::New();
注意：使用值特化进行类型最小最大
值指定
The minimum and maximum values for the output image are specified in the following. Note the
use of the itk::NumericTraits class which allows us to define a number of type-related constants
in a generic way. The use of traits is a fundamental characteristic of generic programming [5, 1].
rescaler->SetOutputMinimum( itk::NumericTraits< OutputPixelType >::min() );
rescaler->SetOutputMaximum( itk::NumericTraits< OutputPixelType >::max() );
Below, we create the reader and writer using the New() method and assign the result to a
itk::SmartPointer.
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
The name of the file to be read or written is passed to the SetFileName() method.
reader->SetFileName( inputFilename );
writer->SetFileName( outputFilename );
Below we connect the reader, filter and writer to form the data processing pipeline.
componentExtractor->SetInput( reader->GetOutput() );
rescaler->SetInput( componentExtractor->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
Finally we execute the pipeline by invoking Update() on the writer. The call is placed in a
try/catch block in case exceptions are thrown.
1.11. Reading and Writing Image Series
25
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.11 Reading and Writing Image Series
It is still quite common to store 3D medical images in sets of files each one containing a single slice
of a volume dataset. Those 2D files can be read as individual 2D images, or can be grouped together
in order to reconstruct a 3D dataset. The same practice can be extended to higher dimensions, for
example, for managing 4D datasets by using sets of files each one containing a 3D image. This
practice is common in the domain of cardiac imaging, perfusion, functional MRI and PET. This
section illustrates the functionalities available in ITK for dealing with reading and writing series of
images.
1.11.1 Reading Image Series
The source code for this section can be found in the file
ImageSeriesReadWrite.cxx.
注意：
This example illustrates how to read a series of 2D slices from independent files in order to compose
a volume. The class itk::ImageSeriesReader is used for this purpose. This class works in
combination with a generator of filenames that will provide a list of files to be read. In this particular
example we use the itk::NumericSeriesFileNames class as a filename generator. This generator
uses a printf style of string format with a “%d” field that will be successively replaced by a number
specified by the user. Here we will use a format like “file%03d.png” for reading PNG files named
file001.png, file002.png, file003.png... and so on.
注意：以这种形式存储图像数据
This requires the following headers as shown.
#include
#include
#include
#include
#include
"itkImage.h"
"itkImageSeriesReader.h"
"itkImageFileWriter.h"
"itkNumericSeriesFileNames.h"
"itkPNGImageIO.h"
26
Chapter 1. Reading and Writing Images
We start by defining the PixelType and ImageType.
using PixelType = unsigned char;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
The image type is used as a template parameter to instantiate the reader and writer.
using ReaderType = itk::ImageSeriesReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
WriterType::Pointer writer = WriterType::New();
Then, we declare the filename generator type and create one instance of it.
using NameGeneratorType = itk::NumericSeriesFileNames;
NameGeneratorType::Pointer nameGenerator = NameGeneratorType::New();
The filename generator requires us to provide a pattern of text for the filenames, and numbers for
the initial value, last value and increment to be used for generating the names of the files.
nameGenerator->SetSeriesFormat( "vwe%03d.png" );
nameGenerator->SetStartIndex( first );
nameGenerator->SetEndIndex( last );
nameGenerator->SetIncrementIndex( 1 );
The ImageIO object that actually performs the read process is now connected to the ImageSeriesReader. This is the safest way of making sure that we use an ImageIO object that is appropriate for
the type of files that we want to read.
reader->SetImageIO( itk::PNGImageIO::New() );
The filenames of the input files must be provided to the reader, while the writer is instructed to write
the same volume dataset in a single file.
reader->SetFileNames( nameGenerator->GetFileNames() );
writer->SetFileName( outputFilename );
We connect the output of the reader to the input of the writer.
1.11. Reading and Writing Image Series
27
writer->SetInput( reader->GetOutput() );
Finally, execution of the pipeline can be triggered by invoking the Update() method in the writer.
This call must be placed in a try/catch block since exceptions be potentially be thrown in the
process of reading or writing the images.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
1.11.2 Writing Image Series
The source code for this section can be found in the file
ImageReadImageSeriesWrite.cxx.
This example illustrates how to save an image using the itk::ImageSeriesWriter. This class
enables the saving of a 3D volume as a set of files containing one 2D slice per file.
The type of the input image is declared here and it is used for declaring the type of the reader. This
will be a conventional 3D image reader.
using ImageType = itk::Image< unsigned char, 3 >;
using ReaderType = itk::ImageFileReader< ImageType >;
The reader object is constructed using the New() operator and assigning the result to a
SmartPointer. The filename of the 3D volume to be read is taken from the command line arguments and passed to the reader using the SetFileName() method.
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
The type of the series writer must be instantiated taking into account that the input file is a 3D
volume and the output files are 2D images. Additionally, the output of the reader is connected as
input to the writer.
28
Chapter 1. Reading and Writing Images
using Image2DType = itk::Image< unsigned char, 2 >;
using WriterType = itk::ImageSeriesWriter< ImageType, Image2DType >;
WriterType::Pointer writer = WriterType::New();
writer->SetInput( reader->GetOutput() );
The writer requires a list of filenames to be generated. This list can be produced with the help of the
itk::NumericSeriesFileNames class.
using NameGeneratorType = itk::NumericSeriesFileNames;
NameGeneratorType::Pointer nameGenerator = NameGeneratorType::New();
The NumericSeriesFileNames class requires an input string in order to have a template for generating the filenames of all the output slices. Here we compose this string using a prefix taken from
the command line arguments and adding the extension for PNG files.
std::string format = argv[2];
format += "%03d.";
format += argv[3];
// filename extension
nameGenerator->SetSeriesFormat( format.c_str() );
The input string is going to be used for generating filenames by setting the values of the first and last
slice. This can be done by collecting information from the input image. Note that before attempting
to take any image information from the reader, its execution must be triggered with the invocation
of the Update() method, and since this invocation can potentially throw exceptions, it must be put
inside a try/catch block.
注意：使用reader之前必须调用
update()进行数据更新；否则读取的
数据未知
try
{
reader->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception thrown while reading the image" << std::endl;
std::cerr << excp << std::endl;
}
Now that the image has been read we can query its largest possible region and recover information
about the number of pixels along every dimension.
1.11. Reading and Writing Image Series
ImageType::ConstPointer
ImageType::RegionType
ImageType::IndexType
ImageType::SizeType
inputImage
region
start
size
=
=
=
=
29
reader->GetOutput();
inputImage->GetLargestPossibleRegion();
region.GetIndex();
region.GetSize();
With this information we can find the number that will identify the first and last slices of the 3D data
set. These numerical values are then passed to the filename generator object that will compose the
names of the files where the slices are going to be stored.
const unsigned int firstSlice = start[2];
const unsigned int lastSlice = start[2] + size[2] - 1;
nameGenerator->SetStartIndex( firstSlice );
nameGenerator->SetEndIndex( lastSlice );
nameGenerator->SetIncrementIndex( 1 );
The list of filenames is taken from the names generator and it is passed to the series writer.
writer->SetFileNames( nameGenerator->GetFileNames() );
Finally we trigger the execution of the pipeline with the Update() method on the writer. At this
point the slices of the image will be saved in individual files containing a single slice per file. The
filenames used for these slices are those produced by the filename generator.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception thrown while reading the image" << std::endl;
std::cerr << excp << std::endl;
}
Note that by saving data into isolated slices we are losing information that may be significant for
medical applications, such as the interslice spacing in millimeters.
1.11.3 Reading and Writing Series of RGB Images
The source code for this section can be found in the file
RGBImageSeriesReadWrite.cxx.
RGB images are commonly used for representing data acquired from cryogenic sections, optical
microscopy and endoscopy. This example illustrates how to read RGB color images from a set of
30
Chapter 1. Reading and Writing Images
files containing individual 2D slices in order to compose a 3D color dataset. Then we will save it
into a single 3D file, and finally save it again as a set of 2D slices with other names.
This requires the following headers as shown.
#include
#include
#include
#include
#include
#include
"itkRGBPixel.h"
"itkImage.h"
"itkImageSeriesReader.h"
"itkImageSeriesWriter.h"
"itkNumericSeriesFileNames.h"
"itkPNGImageIO.h"
The itk::RGBPixel class is templated over the type used to represent each one of the Red, Green
and Blue components. A typical instantiation of the RGB image class might be as follows.
using PixelType = itk::RGBPixel< unsigned char >;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
The image type is used as a template parameter to instantiate the series reader and the volumetric
writer.
using SeriesReaderType = itk::ImageSeriesReader< ImageType >;
using WriterType = itk::ImageFileWriter< ImageType >;
SeriesReaderType::Pointer seriesReader = SeriesReaderType::New();
WriterType::Pointer
writer
= WriterType::New();
We use a NumericSeriesFileNames class in order to generate the filenames of the slices to be read.
Later on in this example we will reuse this object in order to generate the filenames of the slices to
be written.
using NameGeneratorType = itk::NumericSeriesFileNames;
NameGeneratorType::Pointer nameGenerator = NameGeneratorType::New();
nameGenerator->SetStartIndex( first );
nameGenerator->SetEndIndex( last );
nameGenerator->SetIncrementIndex( 1 );
nameGenerator->SetSeriesFormat( "vwe%03d.png" );
The ImageIO object that actually performs the read process is now connected to the ImageSeriesReader.
1.11. Reading and Writing Image Series
31
seriesReader->SetImageIO( itk::PNGImageIO::New() );
The filenames of the input slices are taken from the names generator and passed to the series reader.
seriesReader->SetFileNames( nameGenerator->GetFileNames() );
The name of the volumetric output image is passed to the image writer, and we connect the output
of the series reader to the input of the volumetric writer.
writer->SetFileName( outputFilename );
writer->SetInput( seriesReader->GetOutput() );
Finally, execution of the pipeline can be triggered by invoking the Update() method in the volumetric writer. This, of course, is done from inside a try/catch block.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Error reading the series " << std::endl;
std::cerr << excp << std::endl;
}
We now proceed to save the same volumetric dataset as a set of slices. This is done only to illustrate
the process for saving a volume as a series of 2D individual datasets. The type of the series writer
must be instantiated taking into account that the input file is a 3D volume and the output files are 2D
images. Additionally, the output of the series reader is connected as input to the series writer.
using Image2DType = itk::Image< PixelType, 2 >;
using SeriesWriterType = itk::ImageSeriesWriter< ImageType, Image2DType >;
SeriesWriterType::Pointer seriesWriter = SeriesWriterType::New();
seriesWriter->SetInput( seriesReader->GetOutput() );
We now reuse the filename generator in order to produce the list of filenames for the output series.
In this case we just need to modify the format of the filename generator. Then, we pass the list of
output filenames to the series writer.
32
Chapter 1. Reading and Writing Images
nameGenerator->SetSeriesFormat( "output%03d.png" );
seriesWriter->SetFileNames( nameGenerator->GetFileNames() );
Finally we trigger the execution of the series writer from inside a try/catch block.
try
{
seriesWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Error reading the series " << std::endl;
std::cerr << excp << std::endl;
}
You may have noticed that apart from the declaration of the PixelType there is nothing in this code
that is specific to RGB images. All the actions required to support color images are implemented
注意：由ImageIO负责具体图像格式
internally in the itk::ImageIO objects.
的读取
1.12 Reading and Writing DICOM Images
1.12.1 Foreword
With the introduction of computed tomography (CT) followed by other digital diagnostic imaging
modalities such as MRI in the 1970’s, and the increasing use of computers in clinical applications,
the American College of Radiology (ACR)3 and the National Electrical Manufacturers Association
(NEMA)4 recognized the need for a standard method for transferring images as well as associated
information between devices manufactured from various vendors.
ACR and NEMA formed a joint committee to develop a standard for Digital Imaging and Communications in Medicine (DICOM). This standard was developed in liaison with other Standardization
Organizations such as CEN TC251, JIRA including IEEE, HL7 and ANSI USA as reviewers.
DICOM is a comprehensive set of standards for handling, storing and transmitting information in
medical imaging. The DICOM standard was developed based on the previous NEMA specification.
The standard specifies a file format definition as well as a network communication protocol. DICOM
was developed to enable integration of scanners, servers, workstations and network hardware from
multiple vendors into an image archiving and communication system.
DICOM files consist of a header and a body of image data. The header contains standardized as well
as free-form fields. The set of standardized fields is called the public DICOM dictionary, an instance
3 http://www.acr.org
4 http://www.nema.org
1.12. Reading and Writing DICOM Images
33
of this dictionary is available in ITK in the file Insight/Utilities/gdcm/Dict/dicomV3.dic.
The list of free-form fields is also called the shadow dictionary.
A single DICOM file can contain multiples frames, allowing storage of volumes or animations.
Image data can be compressed using a large variety of standards, including JPEG (both lossy and
lossless), LZW (Lempel Ziv Welch), and RLE (Run-length encoding).
The DICOM Standard is an evolving standard and it is maintained in accordance with the Procedures of the DICOM Standards Committee. Proposals for enhancements are forthcoming from the
DICOM Committee member organizations based on input from users of the Standard. These proposals are considered for inclusion in future editions of the Standard. A requirement in updating the
Standard is to maintain effective compatibility with previous editions.
For a more detailed description of the DICOM standard see [43].
The following sections illustrate how to use the functionalities that ITK provides for reading and
writing DICOM files. This is extremely important in the domain of medical imaging since most
of the images that are acquired in a clinical setting are stored and transported using the DICOM
standard.
注意：
DICOM functionalities in ITK are provided by the GDCM library. This open source library was developed by the CREATIS Team 5 at INSA-Lyon [7]. Although originally this library was distributed
under a LGPL License6 , the CREATIS Team was lucid enough to understand the limitations of that
license and agreed to adopt the more open BSD-like License7 . This change in their licensing made
possible to distribute GDCM along with ITK.
GDCM is now maintained by Mathieu Malaterre and the GDCM community. The version distributed
with ITK gets updated with major releases of the GDCM library.
1.12.2 Reading and Writing a 2D Image
The source code for this section can be found in the file
DicomImageReadWrite.cxx.
注意：
This example illustrates how to read a single DICOM slice and write it back as another DICOM
slice. In the process an intensity rescaling is also applied.
注意：
In order to read and write the slice we use the itk::GDCMImageIO class which encapsulates a
connection to the underlying GDCM library. In this way we gain access from ITK to the DICOM
functionalities offered by GDCM. The GDCMImageIO object is connected as the ImageIO object
to be used by the itk::ImageFileWriter.
We should first include the following header files.
5 http://www.creatis.insa-lyon.fr
6
7
http://www.gnu.org/copyleft/lesser.html
http://www.opensource.org/licenses/bsd-license.php
34
Chapter 1. Reading and Writing Images
#include
#include
#include
#include
"itkImageFileReader.h"
"itkImageFileWriter.h"
"itkRescaleIntensityImageFilter.h"
"itkGDCMImageIO.h"
Then we declare the pixel type and image dimension, and use them for instantiating the image type
to be read.
using InputPixelType = signed short;
constexpr unsigned int InputDimension = 2;
using InputImageType = itk::Image< InputPixelType, InputDimension >;
With the image type we can instantiate the type of the reader, create one, and set the filename of the
image to be read.
using ReaderType = itk::ImageFileReader< InputImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
GDCMImageIO is an ImageIO class for reading and writing DICOM v3 and ACR/NEMA images.
The GDCMImageIO object is constructed here and connected to the ImageFileReader.
using ImageIOType = itk::GDCMImageIO;
ImageIOType::Pointer gdcmImageIO = ImageIOType::New();
reader->SetImageIO( gdcmImageIO );
At this point we can trigger the reading process by invoking the Update() method. Since this
reading process may eventually throw an exception, we place the invocation inside a try/catch
block.
try
{
reader->Update();
}
catch (itk::ExceptionObject & e)
{
std::cerr << "exception in file reader " << std::endl;
std::cerr << e << std::endl;
return EXIT_FAILURE;
}
We now have the image in memory and can get access to it using the GetOutput() method of the
1.12. Reading and Writing DICOM Images
35
reader. In the remainder of this current example, we focus on showing how to save this image again
in DICOM format in a new file.
First, we must instantiate an ImageFileWriter type. Then, we construct one, set the filename to be
used for writing, and connect the input image to be written. Since in this example we write the
image in different ways, and in each case use a different writer, we enumerated the variable names
of the writer objects as well as their types.
using Writer1Type = itk::ImageFileWriter< InputImageType >;
Writer1Type::Pointer writer1 = Writer1Type::New();
writer1->SetFileName( argv[2] );
writer1->SetInput( reader->GetOutput() );
We need to explicitly set the proper image IO (GDCMImageIO) to the writer filter since the input
DICOM dictionary is being passed along the writing process. The dictionary contains all necessary
information that a valid DICOM file should contain, like Patient Name, Patient ID, Institution Name,
etc.
writer1->SetImageIO( gdcmImageIO );
The writing process is triggered by invoking the Update() method. Since this execution may result
in exceptions being thrown we place the Update() call inside a try/catch block.
try
{
writer1->Update();
}
catch (itk::ExceptionObject & e)
{
std::cerr << "exception in file writer " << std::endl;
std::cerr << e << std::endl;
return EXIT_FAILURE;
}
We will now rescale the image using the RescaleIntensityImageFilter. For this purpose we use a
better suited pixel type: unsigned char instead of signed short. The minimum and maximum
values of the output image are explicitly defined in the rescaling filter.
using WritePixelType = unsigned char;
using WriteImageType = itk::Image< WritePixelType, 2 >;
using RescaleFilterType = itk::RescaleIntensityImageFilter<
InputImageType, WriteImageType >;
36
Chapter 1. Reading and Writing Images
RescaleFilterType::Pointer rescaler = RescaleFilterType::New();
rescaler->SetOutputMinimum( 0 );
rescaler->SetOutputMaximum( 255 );
We create a second writer object that will save the rescaled image into a new file, which is not in
DICOM format. This is done only for the sake of verifying the image against the one that will be
saved in DICOM format later in this example.
using Writer2Type = itk::ImageFileWriter< WriteImageType >;
Writer2Type::Pointer writer2 = Writer2Type::New();
writer2->SetFileName( argv[3] );
rescaler->SetInput( reader->GetOutput() );
writer2->SetInput( rescaler->GetOutput() );
The writer can be executed by invoking the Update() method from inside a try/catch block.
We proceed now to save the same rescaled image into a file in DICOM format. For this purpose we
just need to set up a itk::ImageFileWriter and pass to it the rescaled image as input.
using Writer3Type = itk::ImageFileWriter< WriteImageType >;
Writer3Type::Pointer writer3 = Writer3Type::New();
writer3->SetFileName( argv[4] );
writer3->SetInput( rescaler->GetOutput() );
We now need to explicitly set the proper image IO (GDCMImageIO), but also we must tell the
ImageFileWriter to not use the MetaDataDictionary from the input but from the GDCMImageIO
注意：
since this is the one that contains the DICOM specific information
The GDCMImageIO object will automatically detect the pixel type, in this case unsigned char
and it will update the DICOM header information accordingly.
writer3->UseInputMetaDataDictionaryOff ();
writer3->SetImageIO( gdcmImageIO );
Finally we trigger the execution of the DICOM writer by invoking the Update() method from inside
a try/catch block.
try
{
writer3->Update();
1.12. Reading and Writing DICOM Images
37
}
catch (itk::ExceptionObject & e)
{
std::cerr << "Exception in file writer " << std::endl;
std::cerr << e << std::endl;
return EXIT_FAILURE;
}
1.12.3 Reading a 2D DICOM Series and Writing a Volume
The source code for this section can be found in the file
DicomSeriesReadImageWrite2.cxx.
Probably the most common representation of datasets in clinical applications is the one that uses
sets of DICOM slices in order to compose 3-dimensional images. This is the case for CT, MRI and
PET scanners. It is very common therefore for image analysts to have to process volumetric images
stored in a set of DICOM files belonging to a common DICOM series.
The following example illustrates how to use ITK functionalities in order to read a DICOM series
into a volume and then save this volume in another file format.
The example begins by including the appropriate headers. In particular we will need the
itk::GDCMImageIO object in order to have access to the capabilities of the GDCM library for reading DICOM files, and the itk::GDCMSeriesFileNames object for generating the lists of filenames
identifying the slices of a common volumetric dataset.
#include
#include
#include
#include
#include
"itkImage.h"
"itkGDCMImageIO.h"
"itkGDCMSeriesFileNames.h"
"itkImageSeriesReader.h"
"itkImageFileWriter.h"
We define the pixel type and dimension of the image to be read. In this particular case, the dimensionality of the image is 3, and we assume a signed short pixel type that is commonly used for
X-Rays CT scanners.
The image orientation information contained in the direction cosines of the DICOM header are read
in and passed correctly down the image processing pipeline.
using PixelType = signed short;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
We use the image type for instantiating the type of the series reader and for constructing one object
38
Chapter 1. Reading and Writing Images
of its type.
using ReaderType = itk::ImageSeriesReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
A GDCMImageIO object is created and connected to the reader. This object is the one that is aware
of the internal intricacies of the DICOM format.
using ImageIOType = itk::GDCMImageIO;
ImageIOType::Pointer dicomIO = ImageIOType::New();
reader->SetImageIO( dicomIO );
注意：
Now we face one of the main challenges of the process of reading a DICOM series: to identify
from a given directory the set of filenames that belong together to the same volumetric image.
Fortunately for us, GDCM offers functionalities for solving this problem and we just need to invoke those functionalities through an ITK class that encapsulates a communication with GDCM
classes. This ITK object is the GDCMSeriesFileNames. Conveniently, we only need to pass to
this class the name of the directory where the DICOM slices are stored. This is done with the
SetDirectory() method. The GDCMSeriesFileNames object will explore the directory and will
generate a sequence of filenames for DICOM files for one study/series. In this example, we also call
the SetUseSeriesDetails(true) function that tells the GDCMSeriesFileNames object to use additional DICOM information to distinguish unique volumes within the directory. This is useful, for
example, if a DICOM device assigns the same SeriesID to a scout scan and its 3D volume; by using
additional DICOM information the scout scan will not be included as part of the 3D volume. Note
注意：
that SetUseSeriesDetails(true) must be called prior to calling SetDirectory(). By default
SetUseSeriesDetails(true) will use the following DICOM tags to sub-refine a set of files into
注意：分期的依
multiple series:
据；将一个机会的
文件分成多个序列
0020 0011 Series Number
0018 0024 Sequence Name
0018 0050 Slice Thickness
0028 0010 Rows
0028 0011 Columns
If this is not enough for your specific case you can always add some more restrictions using the
AddSeriesRestriction() method. In this example we will use the DICOM Tag: 0008 0021 DA
1 Series Date, to sub-refine each series. The format for passing the argument is a string containing
first the group then the element of the DICOM tag, separated by a pipe (|) sign.
1.12. Reading and Writing DICOM Images
39
using NamesGeneratorType = itk::GDCMSeriesFileNames;
NamesGeneratorType::Pointer nameGenerator = NamesGeneratorType::New();
nameGenerator->SetUseSeriesDetails( true );
nameGenerator->AddSeriesRestriction("0008|0021" );
nameGenerator->SetDirectory( argv[1] );
The GDCMSeriesFileNames object first identifies the list of DICOM series present in the given
directory. We receive that list in a reference to a container of strings and then we can do things like
print out all the series identifiers that the generator had found. Since the process of finding the series
identifiers can potentially throw exceptions, it is wise to put this code inside a try/catch block.
using SeriesIdContainer = std::vector< std::string >;
const SeriesIdContainer & seriesUID = nameGenerator->GetSeriesUIDs();
auto seriesItr = seriesUID.begin();
auto seriesEnd = seriesUID.end();
while( seriesItr != seriesEnd )
{
std::cout << seriesItr->c_str() << std::endl;
++seriesItr;
}
Given that it is common to find multiple DICOM series in the same directory, we must tell the
GDCM classes what specific series we want to read. In this example we do this by checking first if
the user has provided a series identifier in the command line arguments. If no series identifier has
been passed, then we simply use the first series found during the exploration of the directory.
std::string seriesIdentifier;
if( argc > 3 ) // If no optional series identifier
{
seriesIdentifier = argv[3];
}
else
{
seriesIdentifier = seriesUID.begin()->c_str();
}
We pass the series identifier to the name generator and ask for all the filenames associated to that
series. This list is returned in a container of strings by the GetFileNames() method.
using FileNamesContainer = std::vector< std::string >;
FileNamesContainer fileNames;
fileNames = nameGenerator->GetFileNames( seriesIdentifier );
40
Chapter 1. Reading and Writing Images
The list of filenames can now be passed to the
SetFileNames() method.
itk::ImageSeriesReader using the
reader->SetFileNames( fileNames );
Finally we can trigger the reading process by invoking the Update() method in the reader. This call
as usual is placed inside a try/catch block.
try
{
reader->Update();
}
catch (itk::ExceptionObject &ex)
{
std::cout << ex << std::endl;
return EXIT_FAILURE;
}
At this point, we have a volumetric image in memory that we can access by invoking the
GetOutput() method of the reader.
We proceed now to save the volumetric image in another file, as specified by the user in the command line arguments of this program. Thanks to the ImageIO factory mechanism, only the filename
extension is needed to identify the file format in this case.
using WriterType = itk::ImageFileWriter< ImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[2] );
writer->SetInput( reader->GetOutput() );
The process of writing the image is initiated by invoking the Update() method of the writer.
writer->Update();
Note that in addition to writing the volumetric image to a file we could have used it as the input
for any 3D processing pipeline. Keep in mind that DICOM is simply a file format and a network
protocol. Once the image data has been loaded into memory, it behaves as any other volumetric
dataset that you could have loaded from any other file format.
1.12.4 Reading a 2D DICOM Series and Writing a 2D DICOM Series
The source code for this section can be found in the file
DicomSeriesReadSeriesWrite.cxx.
注意：
1.12. Reading and Writing DICOM Images
41
This example illustrates how to read a DICOM series into a volume and then save this volume into
another DICOM series using the exact same header information. It makes use of the GDCM library.
The main purpose of this example is to show how to properly propagate the DICOM specific information along the pipeline to be able to correctly write back the image using the information from
the input DICOM files.
Please note that writing DICOM files is quite a delicate operation since we are dealing with a significant amount of patient specific data. It is your responsibility to verify that the DICOM headers
generated from this code are not introducing risks in the diagnosis or treatment of patients. It is as
well your responsibility to make sure that the privacy of the patient is respected when you process
data sets that contain personal information. Privacy issues are regulated in the United States by the
HIPAA norms8. You would probably find similar legislation in every country.
When saving datasets in DICOM format it must be made clear whether these datasets have been
processed in any way, and if so, you should inform the recipients of the data about the purpose and
potential consequences of the processing. This is fundamental if the datasets are intended to be used
for diagnosis, treatment or follow-up of patients. For example, the simple reduction of a dataset from
a 16-bits/pixel to a 8-bits/pixel representation may make it impossible to detect certain pathologies
and as a result will expose the patient to the risk of remaining untreated for a long period of time
while her/his pathology progresses.
You are strongly encouraged to get familiar with the report on medical errors “To Err is Human”,
produced by the U.S. Institute of Medicine [31]. Raising awareness about the high frequency of
注意：医学图像处理中会存在的错误
medical errors is a first step in reducing their occurrence.
After all these warnings, let us now go back to the code and get familiar with the use of ITK and
GDCM for writing DICOM Series. The first step that we must take is to include the header files of
the relevant classes. We include the GDCMImageIO class, the GDCM filenames generator, as well
as the series reader and writer.
#include
#include
#include
#include
"itkGDCMImageIO.h"
"itkGDCMSeriesFileNames.h"
"itkImageSeriesReader.h"
"itkImageSeriesWriter.h"
As a second step, we define the image type to be used in this example. This is done by explicitly
selecting a pixel type and a dimension. Using the image type we can define the type of the series
reader.
using PixelType = signed short;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
using ReaderType = itk::ImageSeriesReader< ImageType >;
8 The
Health Insurance Portability and Accountability Act of 1996. http://www.cms.hhs.gov/hipaa/
42
Chapter 1. Reading and Writing Images
We also declare types for the itk::GDCMImageIO object that will actually read and write the DICOM images, and the itk::GDCMSeriesFileNames object that will generate and order all the
filenames for the slices composing the volume dataset. Once we have the types, we proceed to
create instances of both objects.
using ImageIOType = itk::GDCMImageIO;
using NamesGeneratorType = itk::GDCMSeriesFileNames;
ImageIOType::Pointer gdcmIO = ImageIOType::New();
NamesGeneratorType::Pointer namesGenerator = NamesGeneratorType::New();
Just as the previous example, we get the DICOM filenames from the directory. Note however, that in
this case we use the SetInputDirectory() method instead of the SetDirectory(). This is done
because in the present case we will use the filenames generator for producing both the filenames for
reading and the filenames for writing. Then, we invoke the GetInputFileNames() method in order
to get the list of filenames to read.
namesGenerator->SetInputDirectory( argv[1] );
const ReaderType::FileNamesContainer & filenames =
namesGenerator->GetInputFileNames();
We construct one instance of the series reader object. Set the DICOM image IO object to be used
with it, and set the list of filenames to read.
ReaderType::Pointer reader = ReaderType::New();
reader->SetImageIO( gdcmIO );
reader->SetFileNames( filenames );
We can trigger the reading process by calling the Update() method on the series reader. It is wise
to put this invocation inside a try/catch block since the process may eventually throw exceptions.
reader->Update();
At this point we have the volumetric data loaded in memory and we can access it by invoking the
GetOutput() method in the reader.
Now we can prepare the process for writing the dataset. First, we take the name of the output
directory from the command line arguments.
const char * outputDirectory = argv[2];
Second, we make sure the output directory exists, using the cross-platform tools:
itksys::SystemTools. In this case we choose to create the directory if it does not exist yet.
1.12. Reading and Writing DICOM Images
43
itksys::SystemTools::MakeDirectory( outputDirectory );
We explicitly instantiate the image type to be used for writing, and use the image type for instantiating the type of the series writer.
using OutputPixelType = signed short;
constexpr unsigned int OutputDimension = 2;
using Image2DType = itk::Image< OutputPixelType, OutputDimension >;
using SeriesWriterType = itk::ImageSeriesWriter<
ImageType, Image2DType >;
We construct a series writer and connect to its input the output from the reader. Then we pass the
GDCM image IO object in order to be able to write the images in DICOM format.
SeriesWriterType::Pointer seriesWriter = SeriesWriterType::New();
seriesWriter->SetInput( reader->GetOutput() );
seriesWriter->SetImageIO( gdcmIO );
It is time now to setup the GDCMSeriesFileNames to generate new filenames using another output
directory. Then simply pass those newly generated files to the series writer.
namesGenerator->SetOutputDirectory( outputDirectory );
seriesWriter->SetFileNames( namesGenerator->GetOutputFileNames() );
The following line of code is extremely important for this process to work correctly. The line is
taking the MetaDataDictionary from the input reader and passing it to the output writer. This step is
important because the MetaDataDictionary contains all the entries of the input DICOM header.
seriesWriter->SetMetaDataDictionaryArray(
reader->GetMetaDataDictionaryArray() );
Finally we trigger the writing process by invoking the Update() method in the series writer. We
place this call inside a try/catch block, in case any exception is thrown during the writing process.
try
{
seriesWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception thrown while writing the series " << std::endl;
44
Chapter 1. Reading and Writing Images
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
Please keep in mind that you should avoid generating DICOM files which have the appearance of
being produced by a scanner. It should be clear from the directory or filenames that these data were
the result of the execution of some sort of algorithm. This will prevent your dataset from being used
as scanner data by accident.
1.12.5 Printing DICOM Tags From One Slice
The source code for this section can be found in the file
DicomImageReadPrintTags.cxx.
注意：
It is often valuable to be able to query the entries from the header of a DICOM file. This can be
used for consistency checking, or simply for verifying that we have the correct dataset in our hands.
This example illustrates how to read a DICOM file and then print out most of the DICOM header
注意：
information. The binary fields of the DICOM header are skipped.
The headers of the main classes involved in this example are specified below. They include the
image file reader, the GDCMImageIO object, the MetaDataDictionary and its entry element, the
注意：存储在这里面
MetaDataObject.
#include "itkImageFileReader.h"
#include "itkGDCMImageIO.h"
#include "itkMetaDataObject.h"
We instantiate the type to be used for storing the image once it is read into memory.
using PixelType = signed short;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
Using the image type as a template parameter we instantiate the type of the image file reader and
construct one instance of it.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
The GDCM image IO type is declared and used for constructing one image IO object.
1.12. Reading and Writing DICOM Images
45
using ImageIOType = itk::GDCMImageIO;
ImageIOType::Pointer dicomIO = ImageIOType::New();
We pass to the reader the filename of the image to be read and connect the ImageIO object to it too.
reader->SetFileName( argv[1] );
reader->SetImageIO( dicomIO );
The reading process is triggered with a call to the Update() method. This call should be placed
inside a try/catch block because its execution may result in exceptions being thrown.
reader->Update();
Now that the image has been read, we obtain the MetaDataDictionary from the ImageIO object
using the GetMetaDataDictionary() method.
注意：
using DictionaryType = itk::MetaDataDictionary;
const
DictionaryType & dictionary = dicomIO->GetMetaDataDictionary();
Since we are interested only in the DICOM tags that can be expressed in strings, we declare a
MetaDataObject suitable for managing strings.
using MetaDataStringType = itk::MetaDataObject< std::string >;
We instantiate the iterators that will make possible to walk through all the entries of the MetaDataDictionary.
auto itr = dictionary.Begin();
auto end = dictionary.End();
For each one of the entries in the dictionary, we check first if its element can be converted to a string,
a dynamic cast is used for this purpose.
while( itr != end )
{
itk::MetaDataObjectBase::Pointer entry = itr->second;
MetaDataStringType::Pointer entryvalue =
dynamic_cast( entry.GetPointer() );
For those entries that can be converted, we take their DICOM tag and pass it to the
GetLabelFromTag() method of the GDCMImageIO class. This method checks the DICOM dictionary and returns the string label associated with the tag that we are providing in the tagkey variable.
46
Chapter 1. Reading and Writing Images
If the label is found, it is returned in labelId variable. The method itself returns false if the tagkey
is not found in the dictionary. For example ”0010|0010” in tagkey becomes ”Patient’s Name” in
labelId.
注意：以前做过
if( entryvalue )
{
std::string tagkey
= itr->first;
std::string labelId;
bool found = itk::GDCMImageIO::GetLabelFromTag( tagkey, labelId );
The actual value of the dictionary
GetMetaDataObjectValue() method.
entry
is
obtained
as
a
string
with
the
std::string tagvalue = entryvalue->GetMetaDataObjectValue();
At this point we can print out an entry by concatenating the DICOM Name or label, the numeric tag
and its actual value.
if( found )
{
std::cout << "(" << tagkey << ") " << labelId;
std::cout << " = " << tagvalue.c_str() << std::endl;
}
Finally we just close the loop that will walk through all the Dictionary entries.
++itr;
}
注意：
It is also possible to read a specific tag. In that case the string of the entry can be used for querying
the MetaDataDictionary.
std::string entryId = "0010|0010";
auto tagItr = dictionary.Find( entryId );
If the entry is actually found in the Dictionary, then we can attempt to convert it to a string entry by
using a dynamic cast.
if( tagItr != end )
{
MetaDataStringType::ConstPointer entryvalue =
dynamic_cast(
tagItr->second.GetPointer() );
1.12. Reading and Writing DICOM Images
47
If the dynamic cast succeeds, then we can print out the values of the label, the tag and the actual
value.
if( entryvalue )
{
std::string tagvalue = entryvalue->GetMetaDataObjectValue();
std::cout << "Patient's Name (" << entryId << ") ";
std::cout << " is: " << tagvalue.c_str() << std::endl;
}
Another way to read a specific tag is to use the encapsulation above MetaDataDictionary. Note that
this is stricly equivalent to the above code.
std::string tagkey = "0008|1050";
std::string labelId;
if( itk::GDCMImageIO::GetLabelFromTag( tagkey, labelId ) )
{
std::string value;
std::cout << labelId << " (" << tagkey << "): ";
if( dicomIO->GetValueFromTag(tagkey, value) )
{
std::cout << value;
}
else
{
std::cout << "(No Value Found in File)";
}
std::cout << std::endl;
}
else
{
std::cerr << "Trying to access inexistant DICOM tag." << std::endl;
}
For a full description of the DICOM dictionary please look at the file.
Insight/Utilities/gdcm/Dicts/dicomV3.dic
The following piece of code will print out the proper pixel type / component for instantiating an
itk::ImageFileReader that can properly import the printed DICOM file.
itk::ImageIOBase::IOPixelType pixelType
= reader->GetImageIO()->GetPixelType();
itk::ImageIOBase::IOComponentType componentType
= reader->GetImageIO()->GetComponentType();
std::cout << "PixelType: " << reader->GetImageIO()
->GetPixelTypeAsString(pixelType) << std::endl;
std::cout << "Component Type: " << reader->GetImageIO()
->GetComponentTypeAsString(componentType) << std::endl;
48
Chapter 1. Reading and Writing Images
1.12.6 Printing DICOM Tags From a Series
The source code for this section can be found in the file
DicomSeriesReadPrintTags.cxx.
This example illustrates how to read a DICOM series into a volume and then print most of the
DICOM header information. The binary fields are skipped.
The header files for the series reader and the GDCM classes for image IO and name generation
should be included first.
#include "itkImageSeriesReader.h"
#include "itkGDCMImageIO.h"
#include "itkGDCMSeriesFileNames.h"
Next, we instantiate the type to be used for storing the image once it is read into memory.
using PixelType = signed short;
constexpr unsigned int Dimension = 3;
using ImageType = itk::Image< PixelType, Dimension >;
We use the image type for instantiating the series reader type and then we construct one object of
this class.
using ReaderType = itk::ImageSeriesReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
A GDCMImageIO object is created and assigned to the reader.
using ImageIOType = itk::GDCMImageIO;
ImageIOType::Pointer dicomIO = ImageIOType::New();
reader->SetImageIO( dicomIO );
注意：
A GDCMSeriesFileNames is declared in order to generate the names of DICOM slices. We specify
the directory with the SetInputDirectory() method and, in this case, take the directory name
from the command line arguments. You could have obtained the directory name from a file dialog
in a GUI.
using NamesGeneratorType = itk::GDCMSeriesFileNames;
NamesGeneratorType::Pointer nameGenerator = NamesGeneratorType::New();
nameGenerator->SetInputDirectory( argv[1] );
1.12. Reading and Writing DICOM Images
49
The list of files to read is obtained from the name generator by invoking the GetInputFileNames()
method and receiving the results in a container of strings. The list of filenames is passed to the
reader using the SetFileNames() method.
注意：获取的是目录下dicom文件的
路径名（保存在vector中）
using FileNamesContainer = std::vector;
FileNamesContainer fileNames = nameGenerator->GetInputFileNames();
reader->SetFileNames( fileNames );
We trigger the reader by invoking the Update() method. This invocation should normally be done
inside a try/catch block given that it may eventually throw exceptions.
reader->Update();
ITK internally queries GDCM and obtains all the DICOM tags from the file headers. The tag
values are stored in the itk::MetaDataDictionary which is a general-purpose container for
{key,value} pairs. The Metadata dictionary can be recovered from any ImageIO class by invoking the GetMetaDataDictionary() method.
using DictionaryType = itk::MetaDataDictionary;
const
注意：获取的是整个dicom体的tags
DictionaryType & dictionary = dicomIO->GetMetaDataDictionary();
In this example, we are only interested in the DICOM tags that can be represented as strings. Therefore, we declare a itk::MetaDataObject of string type in order to receive those particular values.
using MetaDataStringType = itk::MetaDataObject< std::string >;
The metadata dictionary is organized as a container with its corresponding iterators. We can therefore visit all its entries by first getting access to its Begin() and End() methods.
auto itr = dictionary.Begin();
auto end = dictionary.End();
We are now ready for walking through the list of DICOM tags. For this purpose we use the iterators that we just declared. At every entry we attempt to convert it into a string entry by using the
dynamic cast based on RTTI information9. The dictionary is organized like a std::map structure, so we should use the first and second members of every entry in order to get access to the
{key,value} pairs.
9 Run
Time Type Information
50
Chapter 1. Reading and Writing Images
while( itr != end )
{
itk::MetaDataObjectBase::Pointer entry = itr->second;
MetaDataStringType::Pointer entryvalue =
dynamic_cast( entry.GetPointer() );
if( entryvalue )
{
std::string tagkey
= itr->first;
std::string tagvalue = entryvalue->GetMetaDataObjectValue();
std::cout << tagkey << " = " << tagvalue << std::endl;
}
++itr;
}
It is also possible to query for specific entries instead of reading all of them as we did above. In this
case, the user must provide the tag identifier using the standard DICOM encoding. The identifier is
stored in a string and used as key in the dictionary.
std::string entryId = "0010|0010";
auto tagItr = dictionary.Find( entryId );
if( tagItr == end )
{
std::cerr << "Tag " << entryId;
std::cerr << " not found in the DICOM header" << std::endl;
return EXIT_FAILURE;
}
Since the entry may or may not be of string type we must again use a dynamic cast in order to
attempt to convert it to a string dictionary entry. If the conversion is successful, we can then print
out its content.
MetaDataStringType::ConstPointer entryvalue =
dynamic_cast( tagItr->second.GetPointer() );
if( entryvalue )
{
std::string tagvalue = entryvalue->GetMetaDataObjectValue();
std::cout << "Patient's Name (" << entryId << ") ";
std::cout << " is: " << tagvalue << std::endl;
}
else
{
std::cerr << "Entry was not of string type" << std::endl;
return EXIT_FAILURE;
}
1.12. Reading and Writing DICOM Images
51
This type of functionality will probably be more useful when provided through a graphical user
interface. For a full description of the DICOM dictionary please look at the following file.
Insight/Utilities/gdcm/Dicts/dicomV3.dic
注意：
1.12.7 Changing a DICOM Header
The source code for this section can be found in the file
DicomImageReadChangeHeaderWrite.cxx.
This example illustrates how to read a single DICOM slice and write it back with some changed
header information as another DICOM slice. Header Key/Value pairs can be specified on the command line. The keys are defined in the file
Insight/Utilities/gdcm/Dicts/dicomV3.dic.
Please note that modifying the content of a DICOM header is a very risky operation. The header
contains fundamental information about the patient and therefore its consistency must be protected
from any data corruption. Before attempting to modify the DICOM headers of your files, you
must make sure that you have a very good reason for doing so, and that you can ensure that this
information change will not result in a lower quality of health care being delivered to the patient.
We must start by including the relevant header files. Here we include the image reader, image writer,
the image, the metadata dictionary and its entries, the metadata objects and the GDCMImageIO. The
metadata dictionary is the data container that stores all the entries from the DICOM header once the
DICOM image file is read into an ITK image.
#include
#include
#include
#include
#include
"itkImageFileReader.h"
"itkImageFileWriter.h"
"itkImage.h"
"itkMetaDataObject.h"
"itkGDCMImageIO.h"
We declare the image type by selecting a particular pixel type and image dimension.
using InputPixelType = signed short;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
We instantiate the reader type by using the image type as template parameter. An instance of the
reader is created and the file name to be read is taken from the command line arguments.
using ReaderType = itk::ImageFileReader< InputImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
52
Chapter 1. Reading and Writing Images
The GDCMImageIO object is created in order to provide the services for reading and writing DICOM files. The newly created image IO class is connected to the reader.
using ImageIOType = itk::GDCMImageIO;
ImageIOType::Pointer gdcmImageIO = ImageIOType::New();
reader->SetImageIO( gdcmImageIO );
The reading of the image is triggered by invoking Update() in the reader.
reader->Update();
We take the metadata dictionary from the image that the reader had loaded in memory.
InputImageType::Pointer inputImage = reader->GetOutput();
using DictionaryType = itk::MetaDataDictionary;
DictionaryType & dictionary = inputImage->GetMetaDataDictionary();
Now we access the entries in the metadata dictionary, and for particular key values we assign a
new content to the entry. This is done here by taking {key,value} pairs from the command line
arguments. The relevant method is EncapsulateMetaData that takes the dictionary and for a given
key provided by entryId, replaces the current value with the content of the value variable. This is
repeated for every potential pair present in the command line arguments.
for (int i = 3; i < argc; i+=2)
{
std::string entryId( argv[i] );
注意：
std::string value( argv[i+1] );
itk::EncapsulateMetaData( dictionary, entryId, value );
}
Now that the dictionary has been updated, we proceed to save the image. This output image will
have the modified data associated with its DICOM header.
Using the image type, we instantiate a writer type and construct a writer. A short pipeline between
the reader and the writer is connected. The filename to write is taken from the command line
arguments. The image IO object is connected to the writer.
using Writer1Type = itk::ImageFileWriter< InputImageType >;
Writer1Type::Pointer writer1 = Writer1Type::New();
writer1->SetInput( reader->GetOutput() );
writer1->SetFileName( argv[2] );
writer1->SetImageIO( gdcmImageIO );
Execution of the writer is triggered by invoking the Update() method.
1.12. Reading and Writing DICOM Images
53
writer1->Update();
Remember again, that modifying the header entries of a DICOM file involves very serious risks for
patients and therefore must be done with extreme caution.
CHAPTER
TWO
FILTERING
This chapter introduces the most commonly used filters found in the toolkit. Most of these filters are
intended to process images. They will accept one or more images as input and will produce one or
more images as output. ITK is based on a data pipeline architecture in which the output of one filter
is passed as input to another filter. (See the Data Processing Pipeline section in Book 1 for more
information.)
2.1 Thresholding
The thresholding operation is used to change or identify pixel values based on specifying one or more
values (called the threshold value). The following sections describe how to perform thresholding
operations using ITK.
2.1.1 Binary Thresholding
The source code for this section can be found in the file
BinaryThresholdImageFilter.cxx.
This example illustrates the use of the binary threshold image filter. This filter is used to transform an image into a binary image by changing the pixel values according to the rule illustrated in
Figure 2.1. The user defines two thresholds—Upper and Lower—and two intensity values—Inside
and Outside. For each pixel in the input image, the value of the pixel is compared with the lower
and upper thresholds. If the pixel value is inside the range defined by [Lower, Upper] the output
pixel is assigned the InsideValue. Otherwise the output pixels are assigned to the OutsideValue.
Thresholding is commonly applied as the last operation of a segmentation pipeline.
The first step required to use the itk::BinaryThresholdImageFilter is to include its header file.
56
Chapter 2. Filtering
Output
Intensity
Inside
Value
Outside
Value
Lower
Threshold
Upper
Threshold
Input
Intensity
Figure 2.1: Transfer function of the BinaryThresholdImageFilter.
#include "itkBinaryThresholdImageFilter.h"
The next step is to decide which pixel types to use for the input and output images.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
The input and output image types are now defined using their respective pixel types and dimensions.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type can be instantiated using the input and output image types defined above.
using FilterType = itk::BinaryThresholdImageFilter<
InputImageType, OutputImageType >;
An itk::ImageFileReader class is also instantiated in order to read image data from a file. (See
Section 1 on page 1 for more information about reading and writing data.)
using ReaderType = itk::ImageFileReader< InputImageType >;
An itk::ImageFileWriter is instantiated in order to write the output image to a file.
2.1. Thresholding
57
using WriterType = itk::ImageFileWriter< OutputImageType >;
Both the filter and the reader are created by invoking their New() methods and assigning the result
to itk::SmartPointers.
ReaderType::Pointer reader = ReaderType::New();
FilterType::Pointer filter = FilterType::New();
The image obtained with the reader is passed as input to the BinaryThresholdImageFilter.
filter->SetInput( reader->GetOutput() );
The method SetOutsideValue() defines the intensity value to be assigned to those pixels
whose intensities are outside the range defined by the lower and upper thresholds. The method
SetInsideValue() defines the intensity value to be assigned to pixels with intensities falling inside the threshold range.
filter->SetOutsideValue( outsideValue );
filter->SetInsideValue( insideValue );
The methods SetLowerThreshold() and SetUpperThreshold() define the range of the input
image intensities that will be transformed into the InsideValue. Note that the lower and upper
thresholds are values of the type of the input image pixels, while the inside and outside values are of
the type of the output image pixels.
filter->SetLowerThreshold( lowerThreshold );
filter->SetUpperThreshold( upperThreshold );
The execution of the filter is triggered by invoking the Update() method. If the filter’s output
has been passed as input to subsequent filters, the Update() call on any downstream filters in the
pipeline will indirectly trigger the update of this filter.
filter->Update();
Figure 2.2 illustrates the effect of this filter on a MRI proton density image of the brain. This
figure shows the limitations of the filter for performing segmentation by itself. These limitations are
particularly noticeable in noisy images and in images lacking spatial uniformity as is the case with
MRI due to field bias.
The following classes provide similar functionality:
• itk::ThresholdImageFilter
58
Chapter 2. Filtering
Figure 2.2: Effect of the BinaryThresholdImageFilter on a slice from a MRI proton density image of the brain.
2.1.2 General Thresholding
The source code for this section can be found in the file
ThresholdImageFilter.cxx.
This example illustrates the use of the itk::ThresholdImageFilter. This filter can be used to
transform the intensity levels of an image in three different ways.
• First, the user can define a single threshold. Any pixels with values below this threshold will
be replaced by a user defined value, called here the OutsideValue. Pixels with values above
the threshold remain unchanged. This type of thresholding is illustrated in Figure 2.3.
• Second, the user can define a particular threshold such that all the pixels with values above
the threshold will be replaced by the OutsideValue. Pixels with values below the threshold
remain unchanged. This is illustrated in Figure 2.4.
• Third, the user can provide two thresholds. All the pixels with intensity values inside the
range defined by the two thresholds will remain unchanged. Pixels with values outside this
range will be assigned to the OutsideValue. This is illustrated in Figure 2.5.
The following methods choose among the three operating modes of the filter.
• ThresholdBelow()
2.1. Thresholding
59
Output
Intensity
Unchanged
Intensities
Outside
Value
Input
Intensity
Threshold
Below
Figure 2.3: ThresholdImageFilter using the threshold-below mode.
Output
Intensity
Outside
Value
Unchanged
Intensities
Input
Intensity
Threshold
Above
Figure 2.4: ThresholdImageFilter using the threshold-above mode.
Output
Intensity
Outside
Value
Unchanged
Intensities
Lower
Threshold
Upper
Threshold
Input
Intensity
Figure 2.5: ThresholdImageFilter using the threshold-outside mode.
60
Chapter 2. Filtering
• ThresholdAbove()
• ThresholdOutside()
The first step required to use this filter is to include its header file.
#include "itkThresholdImageFilter.h"
Then we must decide what pixel type to use for the image. This filter is templated over a single
image type because the algorithm only modifies pixel values outside the specified range, passing the
rest through unchanged.
using PixelType = unsigned char;
The image is defined using the pixel type and the dimension.
using ImageType = itk::Image< PixelType, 2 >;
The filter can be instantiated using the image type defined above.
using FilterType = itk::ThresholdImageFilter< ImageType >;
An itk::ImageFileReader class is also instantiated in order to read image data from a file.
using ReaderType = itk::ImageFileReader< ImageType >;
An itk::ImageFileWriter is instantiated in order to write the output image to a file.
using WriterType = itk::ImageFileWriter< ImageType >;
Both the filter and the reader are created by invoking their New() methods and assigning the result
to SmartPointers.
ReaderType::Pointer reader = ReaderType::New();
FilterType::Pointer filter = FilterType::New();
The image obtained with the reader is passed as input to the itk::ThresholdImageFilter.
filter->SetInput( reader->GetOutput() );
2.2. Edge Detection
61
The method SetOutsideValue() defines the intensity value to be assigned to those pixels whose
intensities are outside the range defined by the lower and upper thresholds.
filter->SetOutsideValue( 0 );
The method ThresholdBelow() defines the intensity value below which pixels of the input image
will be changed to the OutsideValue.
filter->ThresholdBelow( 180 );
The filter is executed by invoking the Update() method. If the filter is part of a larger image
processing pipeline, calling Update() on a downstream filter will also trigger update of this filter.
filter->Update();
The output of this example is shown in Figure 2.3. The second operating mode of the filter is now
enabled by calling the method ThresholdAbove().
filter->ThresholdAbove( 180 );
filter->Update();
Updating the filter with this new setting produces the output shown in Figure 2.4. The third operating
mode of the filter is enabled by calling ThresholdOutside().
filter->ThresholdOutside( 170,190 );
filter->Update();
The output of this third, “band-pass” thresholding mode is shown in Figure 2.5.
The examples in this section also illustrate the limitations of the thresholding filter for performing
segmentation by itself. These limitations are particularly noticeable in noisy images and in images
lacking spatial uniformity, as is the case with MRI due to field bias.
The following classes provide similar functionality:
• itk::BinaryThresholdImageFilter
2.2 Edge Detection
2.2.1 Canny Edge Detection
The source code for this section can be found in the file
CannyEdgeDetectionImageFilter.cxx.
62
Chapter 2. Filtering
This example introduces the use of the itk::CannyEdgeDetectionImageFilter. Canny edge
detection is widely used for edge detection since it is the optimal solution satisfying the constraints
of good sensitivity, localization and noise robustness. To achieve this end, Canny edge detection is
implemented internally as a multi-stage algorithm, which involves Gaussian smoothing to remove
noise, calculation of gradient magnitudes to localize edge features, non-maximum suppression to
remove suprious features, and finally thresholding to yield a binary image. Though the specifics of
this internal pipeline are largely abstracted from the user of the class, it is nonetheless beneficial to
have a general understanding of these components so that parameters can be appropriately adjusted.
The first step required for using this filter is to include its header file.
#include "itkCannyEdgeDetectionImageFilter.h"
In this example, images are read and written with unsigned char pixel type. However, Canny edge
detection requires floating point pixel types in order to avoid numerical errors. For this reason, a
separate internal image type with pixel type double is defined for edge detection.
constexpr unsigned int Dimension = 2;
using CharPixelType = unsigned char; // IO
using RealPixelType = double; // Operations
using CharImageType = itk::Image< CharPixelType, Dimension >;
using RealImageType = itk::Image< RealPixelType, Dimension >;
The CharImageType image is cast to and from RealImageType using itk::CastImageFilter
and RescaleIntensityImageFilter, respectively; both the input and output of
CannyEdgeDetectionImageFilter are RealImageType.
using CastToRealFilterType =
itk::CastImageFilter< CharImageType, RealImageType >;
using CannyFilterType =
itk::CannyEdgeDetectionImageFilter< RealImageType, RealImageType >;
using RescaleFilterType =
itk::RescaleIntensityImageFilter< RealImageType, CharImageType >;
In this example, three parameters of the Canny edge detection filter may be set via the
SetVariance(), SetUpperThreshold(), and SetLowerThreshold() methods. Based on the previous discussion of the steps in the internal pipeline, we understand that variance adjusts the
amount of Gaussian smoothing and upperThreshold and lowerThreshold control which edges
are selected in the final step.
cannyFilter->SetVariance( variance );
cannyFilter->SetUpperThreshold( upperThreshold );
cannyFilter->SetLowerThreshold( lowerThreshold );
2.3. Casting and Intensity Mapping
63
Finally, Update() is called on writer to trigger excecution of the pipeline. As usual, the call is
wrapped in a try/catch block.
try
{
writer->Update();
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
2.3 Casting and Intensity Mapping
The filters discussed in this section perform pixel-wise intensity mappings. Casting is used to convert
one pixel type to another, while intensity mappings also take into account the different intensity
ranges of the pixel types.
2.3.1 Linear Mappings
The source code for this section can be found in the file
CastingImageFilters.cxx.
Due to the use of Generic Programming in the toolkit, most types are resolved at compile-time. Few
decisions regarding type conversion are left to run-time. It is up to the user to anticipate the pixel
type-conversions required in the data pipeline. In medical imaging applications it is usually not
desirable to use a general pixel type since this may result in the loss of valuable information.
This section introduces the mechanisms for explicit casting of images that flow through the
pipeline. The following four filters are treated in this section:
itk::CastImageFilter,
itk::RescaleIntensityImageFilter,
itk::ShiftScaleImageFilter
and
itk::NormalizeImageFilter. These filters are not directly related to each other except
that they all modify pixel values. They are presented together here for the purpose of comparing
their individual features.
The CastImageFilter is a very simple filter that acts pixel-wise on an input image, casting every pixel
to the type of the output image. Note that this filter does not perform any arithmetic operation on the
intensities. Applying CastImageFilter is equivalent to performing a C-Style cast on every pixel.
outputPixel = static cast( inputPixel )
64
Chapter 2. Filtering
The RescaleIntensityImageFilter linearly scales the pixel values in such a way that the minimum and
maximum values of the input are mapped to minimum and maximum values provided by the user.
This is a typical process for forcing the dynamic range of the image to fit within a particular scale
and is common for image display. The linear transformation applied by this filter can be expressed
as
out putPixel = (inputPixel − inpMin) ×
.
(outMax − outMin)
+ outMin
(inpMax − inpMin)
The ShiftScaleImageFilter also applies a linear transformation to the intensities of the input image,
but the transformation is specified by the user in the form of a multiplying factor and a value to be
added. This can be expressed as
out putPixel = (inputPixel + Shi f t) × Scale
.
The parameters of the linear transformation applied by the NormalizeImageFilter are computed
internally such that the statistical distribution of gray levels in the output image have zero mean and
a variance of one. This intensity correction is particularly useful in registration applications as a
preprocessing step to the evaluation of mutual information metrics. The linear transformation of
NormalizeImageFilter is given as
out putPixel =
.
(inputPixel − mean)
√
variance
As usual, the first step required to use these filters is to include their header files.
#include "itkCastImageFilter.h"
#include "itkRescaleIntensityImageFilter.h"
#include "itkNormalizeImageFilter.h"
Let’s define pixel types for the input and output images.
using InputPixelType = unsigned char;
using OutputPixelType = float;
Then, the input and output image types are defined.
using InputImageType = itk::Image< InputPixelType, 3 >;
using OutputImageType = itk::Image< OutputPixelType, 3 >;
The filters are instantiated using the defined image types.
2.3. Casting and Intensity Mapping
65
using CastFilterType = itk::CastImageFilter<
InputImageType, OutputImageType >;
using RescaleFilterType = itk::RescaleIntensityImageFilter<
InputImageType, OutputImageType >;
using ShiftScaleFilterType = itk::ShiftScaleImageFilter<
InputImageType, OutputImageType >;
using NormalizeFilterType = itk::NormalizeImageFilter<
InputImageType, OutputImageType >;
Object filters are created by invoking the New() method and assigning the result to
itk::SmartPointers.
CastFilterType::Pointer
RescaleFilterType::Pointer
ShiftScaleFilterType::Pointer
NormalizeFilterType::Pointer
castFilter
rescaleFilter
shiftFilter
normalizeFilter
= CastFilterType::New();
= RescaleFilterType::New();
= ShiftScaleFilterType::New();
= NormalizeFilterType::New();
The output of a reader filter (whose creation is not shown here) is now connected as input to the
various casting filters.
castFilter->SetInput(
shiftFilter->SetInput(
rescaleFilter->SetInput(
normalizeFilter->SetInput(
reader->GetOutput() );
reader->GetOutput() );
reader->GetOutput() );
reader->GetOutput() );
Next we proceed to setup the parameters required by each filter. The CastImageFilter and the NormalizeImageFilter do not require any parameters. The RescaleIntensityImageFilter, on the other
hand, requires the user to provide the desired minimum and maximum pixel values of the output
image. This is done by using the SetOutputMinimum() and SetOutputMaximum() methods as
illustrated below.
rescaleFilter->SetOutputMinimum( 10 );
rescaleFilter->SetOutputMaximum( 250 );
The ShiftScaleImageFilter requires a multiplication factor (scale) and a post-scaling additive value
(shift). The methods SetScale() and SetShift() are used, respectively, to set these values.
shiftFilter->SetScale( 1.2 );
shiftFilter->SetShift( 25 );
Finally, the filters are executed by invoking the Update() method.
66
Chapter 2. Filtering
castFilter->Update();
shiftFilter->Update();
rescaleFilter->Update();
normalizeFilter->Update();
2.3.2 Non Linear Mappings
The following filter can be seen as a variant of the casting filters. Its main difference is the use of a
smooth and continuous transition function of non-linear form.
The source code for this section can be found in the file
SigmoidImageFilter.cxx.
The itk::SigmoidImageFilter is commonly used as an intensity transform. It maps a specific
range of intensity values into a new intensity range by making a very smooth and continuous transition in the borders of the range. Sigmoids are widely used as a mechanism for focusing attention
on a particular set of values and progressively attenuating the values outside that range. In order to
extend the flexibility of the Sigmoid filter, its implementation in ITK includes four parameters that
can be tuned to select its input and output intensity ranges. The following equation represents the
Sigmoid intensity transformation, applied pixel-wise.
1
+ Min
I ′ = (Max − Min) ·
− I−β
1+e α
(2.1)
In the equation above, I is the intensity of the input pixel, I ′ the intensity of the output pixel,
Min, Max are the minimum and maximum values of the output image, α defines the width of the
input intensity range, and β defines the intensity around which the range is centered. Figure 2.6
illustrates the significance of each parameter.
This filter will work on images of any dimension and will take advantage of multiple processors
when available.
The header file corresponding to this filter should be included first.
#include "itkSigmoidImageFilter.h"
Then pixel and image types for the filter input and output must be defined.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
2.3. Casting and Intensity Mapping
67
OutputMaximum
1
1
Alpha=−1
0.8
0.8
0.6
0.4
Beta = −4
Beta = −2
Beta = 0
0.6
Alpha=4
Alpha=2
0.4
Alpha=0.25
Alpha=1
Alpha=0.5
0.2
0
−10
Beta = 2
Beta = 4
0.2
OutputMinimum
−5
0
5
10
0
−10
−5
0
5
10
Figure 2.6: Effects of the various parameters in the SigmoidImageFilter. The alpha parameter defines the width
of the intensity window. The beta parameter defines the center of the intensity window.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
Using the image types, we instantiate the filter type and create the filter object.
using SigmoidFilterType = itk::SigmoidImageFilter<
InputImageType, OutputImageType >;
SigmoidFilterType::Pointer sigmoidFilter = SigmoidFilterType::New();
The minimum and maximum values desired in the output are defined using the methods
SetOutputMinimum() and SetOutputMaximum().
sigmoidFilter->SetOutputMinimum(
sigmoidFilter->SetOutputMaximum(
outputMinimum );
outputMaximum );
The coefficients α and β are set with the methods SetAlpha() and SetBeta(). Note that α is
proportional to the width of the input intensity window. As rule of thumb, we may say that the
window is the interval [−3α, 3α]. The boundaries of the intensity window are not sharp. The α
curve approaches its extrema smoothly, as shown in Figure 2.6. You may want to think about this
in the same terms as when taking a range in a population of measures by defining an interval of
[−3σ, +3σ] around the population mean.
sigmoidFilter->SetAlpha( alpha
sigmoidFilter->SetBeta( beta
);
);
The input to the SigmoidImageFilter can be taken from any other filter, such as an image file reader,
for example. The output can be passed down the pipeline to other filters, like an image file writer.
An Update() call on any downstream filter will trigger the execution of the Sigmoid filter.
68
Chapter 2. Filtering
Figure 2.7: Effect of the Sigmoid filter on a slice from a MRI proton density brain image.
sigmoidFilter->SetInput( reader->GetOutput() );
writer->SetInput( sigmoidFilter->GetOutput() );
writer->Update();
Figure 2.7 illustrates the effect of this filter on a slice of MRI brain image using the following
parameters.
• Minimum = 10
• Maximum = 240
• α = 10
• β = 170
As can be seen from the figure, the intensities of the white matter were expanded in their dynamic
range, while intensity values lower than β−3α and higher than β+3α became progressively mapped
to the minimum and maximum output values. This is the way in which a Sigmoid can be used for
performing smooth intensity windowing.
Note that both α and β can be positive and negative. A negative α will have the effect of negating
the image. This is illustrated on the left side of Figure 2.6. An application of the Sigmoid filter as
preprocessing for segmentation is presented in Section 4.3.1.
2.4. Gradients
69
Sigmoid curves are common in the natural world. They represent the plot of sensitivity to a stimulus.
They are also the integral curve of the Gaussian and, therefore, appear naturally as the response to
signals whose distribution is Gaussian.
2.4 Gradients
Computation of gradients is a fairly common operation in image processing. The term “gradient”
may refer in some contexts to the gradient vectors and in others to the magnitude of the gradient vectors. ITK filters attempt to reduce this ambiguity by including the magnitude term when appropriate.
ITK provides filters for computing both the image of gradient vectors and the image of magnitudes.
2.4.1 Gradient Magnitude
The source code for this section can be found in the file
GradientMagnitudeImageFilter.cxx.
The magnitude of the image gradient is extensively used in image analysis, mainly to help
in the determination of object contours and the separation of homogeneous regions. The
itk::GradientMagnitudeImageFilter computes the magnitude of the image gradient at each
pixel location using a simple finite differences approach. For example, in the case of 2D the computation is equivalent to convolving the image with masks of type
-1
-1
0
1
0
1
then adding the sum of their squares and computing the square root of the sum.
This filter will work on images of any dimension thanks to the internal use of
itk::NeighborhoodIterator and itk::NeighborhoodOperator.
The first step required to use this filter is to include its header file.
#include "itkGradientMagnitudeImageFilter.h"
Types should be chosen for the pixels of the input and output images.
using InputPixelType = float;
using OutputPixelType = float;
The input and output image types can be defined using the pixel types.
70
Chapter 2. Filtering
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The type of the gradient magnitude filter is defined by the input image and the output image types.
using FilterType = itk::GradientMagnitudeImageFilter<
InputImageType, OutputImageType >;
A filter object is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, the source is an image
reader.
filter->SetInput( reader->GetOutput() );
Finally, the filter is executed by invoking the Update() method.
filter->Update();
If the output of this filter has been connected to other filters in a pipeline, updating any of the
downstream filters will also trigger an update of this filter. For example, the gradient magnitude
filter may be connected to an image writer.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.8 illustrates the effect of the gradient magnitude filter on a MRI proton density image of
the brain. The figure shows the sensitivity of this filter to noisy data.
Attention should be paid to the image type chosen to represent the output image since the dynamic
range of the gradient magnitude image is usually smaller than the dynamic range of the input image.
As always, there are exceptions to this rule, for example, synthetic images that contain high contrast
objects.
This filter does not apply any smoothing to the image before computing the gradients. The results
can therefore be very sensitive to noise and may not be the best choice for scale-space analysis.
2.4. Gradients
71
Figure 2.8: Effect of the GradientMagnitudeImageFilter on a slice from a MRI proton density image of the brain.
2.4.2 Gradient Magnitude With Smoothing
The source code for this section can be found in the file
GradientMagnitudeRecursiveGaussianImageFilter.cxx.
Differentiation is an ill-defined operation over digital data. In practice it is convenient to define a
scale in which the differentiation should be performed. This is usually done by preprocessing the
data with a smoothing filter. It has been shown that a Gaussian kernel is the most convenient choice
for performing such smoothing. By choosing a particular value for the standard deviation (σ) of the
Gaussian, an associated scale is selected that ignores high frequency content, commonly considered
image noise.
The itk::GradientMagnitudeRecursiveGaussianImageFilter computes the magnitude of the
image gradient at each pixel location. The computational process is equivalent to first smoothing the
image by convolving it with a Gaussian kernel and then applying a differential operator. The user
selects the value of σ.
Internally this is done by applying an IIR 1 filter that approximates a convolution with the derivative
of the Gaussian kernel. Traditional convolution will produce a more accurate result, but the IIR
approach is much faster, especially using large σs [15, 16].
GradientMagnitudeRecursiveGaussianImageFilter will work on images of any dimension by taking
advantage of the natural separability of the Gaussian kernel and its derivatives.
1 Infinite
Impulse Response
72
Chapter 2. Filtering
The first step required to use this filter is to include its header file.
#include "itkGradientMagnitudeRecursiveGaussianImageFilter.h"
Types should be instantiated based on the pixels of the input and output images.
using InputPixelType = float;
using OutputPixelType = float;
With them, the input and output image types can be instantiated.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types.
using FilterType = itk::GradientMagnitudeRecursiveGaussianImageFilter<
InputImageType, OutputImageType >;
A filter object is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
filter->SetInput( reader->GetOutput() );
The standard deviation of the Gaussian smoothing kernel is now set.
filter->SetSigma( sigma );
Finally the filter is executed by invoking the Update() method.
filter->Update();
If connected to other filters in a pipeline, this filter will automatically update when any downstream
filters are updated. For example, we may connect this gradient magnitude filter to an image file
writer and then update the writer.
2.4. Gradients
73
Figure 2.9: Effect of the GradientMagnitudeRecursiveGaussianImageFilter on a slice from a MRI proton density
image of the brain.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.9 illustrates the effect of this filter on a MRI proton density image of the brain using σ
values of 3 (left) and 5 (right). The figure shows how the sensitivity to noise can be regulated by
selecting an appropriate σ. This type of scale-tunable filter is suitable for performing scale-space
analysis.
Attention should be paid to the image type chosen to represent the output image since the dynamic
range of the gradient magnitude image is usually smaller than the dynamic range of the input image.
2.4.3 Derivative Without Smoothing
The source code for this section can be found in the file
DerivativeImageFilter.cxx.
The itk::DerivativeImageFilter is used for computing the partial derivative of an image, the
derivative of an image along a particular axial direction.
The header file corresponding to this filter should be included first.
74
Chapter 2. Filtering
#include "itkDerivativeImageFilter.h"
Next, the pixel types for the input and output images must be defined and, with them, the image
types can be instantiated. Note that it is important to select a signed type for the image, since the
values of the derivatives will be positive as well as negative.
using InputPixelType = float;
using OutputPixelType = float;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
Using the image types, it is now possible to define the filter type and create the filter object.
using FilterType = itk::DerivativeImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The order of the derivative is selected with the SetOrder() method. The direction along which the
derivative will be computed is selected with the SetDirection() method.
filter->SetOrder(
std::stoi( argv[4] ) );
filter->SetDirection( std::stoi( argv[5] ) );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An Update() call on any downstream
filter will trigger the execution of the derivative filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.10 illustrates the effect of the DerivativeImageFilter on a slice of MRI brain image. The
derivative is taken along the x direction. The sensitivity to noise in the image is evident from this
result.
2.5. Second Order Derivatives
75
Figure 2.10: Effect of the Derivative filter on a slice from a MRI proton density brain image.
2.5 Second Order Derivatives
2.5.1 Second Order Recursive Gaussian
The source code for this section can be found in the file
SecondDerivativeRecursiveGaussianImageFilter.cxx.
This example illustrates how to compute second derivatives of a 3D image using the
itk::RecursiveGaussianImageFilter.
It’s good to be able to compute the raw derivative without any smoothing, but this can be problematic in a medical imaging scenario, when images will often have a certain amount of noise. It’s
almost always more desirable to include a smoothing step first, where an image is convolved with
a Gaussian kernel in whichever directions the user desires a derivative. The nature of the Gaussian
kernel makes it easy to combine these two steps into one, using an infinite impulse response (IIR)
filter. In this example, all the second derivatives are computed independently in the same way, as
if they were intended to be used for building the Hessian matrix of the image (a square matrix of
second-order derivatives of an image, which is useful in many image processing techniques).
First, we will include the relevant header files: the itkRecursiveGaussianImageFilter, the image
reader, writer, and duplicator.
76
Chapter 2. Filtering
#include
#include
#include
#include
#include
"itkRecursiveGaussianImageFilter.h"
"itkImageFileReader.h"
"itkImageFileWriter.h"
"itkImageDuplicator.h"
Next, we declare our pixel type and output pixel type to be floats, and our image dimension to be 3.
using PixelType = float;
using OutputPixelType = float;
constexpr unsigned int Dimension = 3;
Using these definitions, define the image types, reader and writer types, and duplicator types, which
are templated over the pixel types and dimension. Then, instantiate the reader, writer, and duplicator
with the New() method.
using ImageType = itk::Image< PixelType,
Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
using ReaderType = itk::ImageFileReader< ImageType
>;
using WriterType = itk::ImageFileWriter< OutputImageType >;
using DuplicatorType = itk::ImageDuplicator< OutputImageType >;
using FilterType = itk::RecursiveGaussianImageFilter<
ImageType,
ImageType >;
ReaderType::Pointer reader
WriterType::Pointer writer
= ReaderType::New();
= WriterType::New();
DuplicatorType::Pointer duplicator = DuplicatorType::New();
Here we create three new filters. For each derivative we take, we will want to smooth in that
direction first. So after the filters are created, each is given a dimension, and set to (in this
example) the same sigma. Note that here, σ represents the standard deviation, whereas the
itk::DiscreteGaussianImageFilter exposes the SetVariance method.
FilterType::Pointer ga = FilterType::New();
FilterType::Pointer gb = FilterType::New();
FilterType::Pointer gc = FilterType::New();
ga->SetDirection( 0 );
gb->SetDirection( 1 );
gc->SetDirection( 2 );
if( argc > 3 )
2.5. Second Order Derivatives
{
const float
ga->SetSigma(
gb->SetSigma(
gc->SetSigma(
}
sigma
sigma
sigma
sigma
77
= std::stod( argv[3] );
);
);
);
First we will compute the second derivative of the z-direction. In order to do this, we smooth in the
x- and y- directions, and finally smooth and compute the derivative in the z-direction. Taking the
zero-order derivative is equivalent to simply smoothing in that direction. This result is commonly
notated Izz .
ga->SetZeroOrder();
gb->SetZeroOrder();
gc->SetSecondOrder();
ImageType::Pointer inputImage = reader->GetOutput();
ga->SetInput( inputImage );
gb->SetInput( ga->GetOutput() );
gc->SetInput( gb->GetOutput() );
duplicator->SetInputImage( gc->GetOutput() );
gc->Update();
duplicator->Update();
ImageType::Pointer Izz = duplicator->GetOutput();
Recall that gc is the filter responsible for taking the second derivative. We can now take advantage
of the pipeline architecture and, without much hassle, switch the direction of gc and gb, so that gc
now takes the derivatives in the y-direction. Now we only need to call Update() on gc to re-run
the entire pipeline from ga to gc, obtaining the second-order derivative in the y-direction, which is
commonly notated Iyy .
gc->SetDirection( 1 );
gb->SetDirection( 2 );
// gc now works along Y
// gb now works along Z
gc->Update();
duplicator->Update();
ImageType::Pointer Iyy = duplicator->GetOutput();
Now we switch the directions of gc with that of ga in order to take the derivatives in the x-direction.
This will give us Ixx .
78
Chapter 2. Filtering
gc->SetDirection( 0 );
ga->SetDirection( 1 );
// gc now works along X
// ga now works along Y
gc->Update();
duplicator->Update();
ImageType::Pointer Ixx = duplicator->GetOutput();
Now we can reset the directions to their original values, and compute first derivatives in different
directions. Since we set both gb and gc to compute first derivatives, and ga to zero-order (which is
only smoothing) we will obtain Iyz .
ga->SetDirection( 0 );
gb->SetDirection( 1 );
gc->SetDirection( 2 );
ga->SetZeroOrder();
gb->SetFirstOrder();
gc->SetFirstOrder();
gc->Update();
duplicator->Update();
ImageType::Pointer Iyz = duplicator->GetOutput();
Here is how you may easily obtain Ixz .
ga->SetDirection( 1 );
gb->SetDirection( 0 );
gc->SetDirection( 2 );
ga->SetZeroOrder();
gb->SetFirstOrder();
gc->SetFirstOrder();
gc->Update();
duplicator->Update();
ImageType::Pointer Ixz = duplicator->GetOutput();
For the sake of completeness, here is how you may compute Ixz and Ixy .
writer->SetInput( Ixz );
outputFileName = outputPrefix + "-Ixz.mhd";
writer->SetFileName( outputFileName.c_str() );
writer->Update();
ga->SetDirection( 2 );
gb->SetDirection( 0 );
2.5. Second Order Derivatives
79
gc->SetDirection( 1 );
ga->SetZeroOrder();
gb->SetFirstOrder();
gc->SetFirstOrder();
gc->Update();
duplicator->Update();
ImageType::Pointer Ixy = duplicator->GetOutput();
writer->SetInput( Ixy );
outputFileName = outputPrefix + "-Ixy.mhd";
writer->SetFileName( outputFileName.c_str() );
writer->Update();
2.5.2 Laplacian Filters
Laplacian Filter Recursive Gaussian
The source code for this section can be found in the file
LaplacianRecursiveGaussianImageFilter1.cxx.
This example illustrates how to use the itk::RecursiveGaussianImageFilter for computing
the Laplacian of a 2D image.
The first step required to use this filter is to include its header file.
#include "itkRecursiveGaussianImageFilter.h"
Types should be selected on the desired input and output pixel types.
using InputPixelType = float;
using OutputPixelType = float;
The input and output image types are instantiated using the pixel types.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types.
80
Chapter 2. Filtering
using FilterType = itk::RecursiveGaussianImageFilter<
InputImageType, OutputImageType >;
This filter applies the approximation of the convolution along a single dimension. It is therefore
necessary to concatenate several of these filters to produce smoothing in all directions. In this
example, we create a pair of filters since we are processing a 2D image. The filters are created by
invoking the New() method and assigning the result to a itk::SmartPointer.
We need two filters for computing the X component of the Laplacian and two other filters for computing the Y component.
FilterType::Pointer filterX1 = FilterType::New();
FilterType::Pointer filterY1 = FilterType::New();
FilterType::Pointer filterX2 = FilterType::New();
FilterType::Pointer filterY2 = FilterType::New();
Since each one of the newly created filters has the potential to perform filtering along any dimension,
we have to restrict each one to a particular direction. This is done with the SetDirection() method.
filterX1->SetDirection( 0 );
filterY1->SetDirection( 1 );
// 0 --> X direction
// 1 --> Y direction
filterX2->SetDirection( 0 );
filterY2->SetDirection( 1 );
// 0 --> X direction
// 1 --> Y direction
The itk::RecursiveGaussianImageFilter can approximate the convolution with the Gaussian
or with its first and second derivatives. We select one of these options by using the SetOrder()
method. Note that the argument is an enum whose values can be ZeroOrder, FirstOrder and
SecondOrder. For example, to compute the x partial derivative we should select FirstOrder for
x and ZeroOrder for y. Here we want only to smooth in x and y, so we select ZeroOrder in both
directions.
filterX1->SetOrder( FilterType::ZeroOrder );
filterY1->SetOrder( FilterType::SecondOrder );
filterX2->SetOrder( FilterType::SecondOrder );
filterY2->SetOrder( FilterType::ZeroOrder );
There are two typical ways of normalizing Gaussians depending on their application. For scalespace analysis it is desirable to use a normalization that will preserve the maximum value of the
input. This normalization is represented by the following equation.
1
√
σ 2π
(2.2)
2.5. Second Order Derivatives
81
In applications that use the Gaussian as a solution of the diffusion equation it is desirable to use a
normalization that preserves the integral of the signal. This last approach can be seen as a conservation of mass principle. This is represented by the following equation.
σ2
1
√
2π
(2.3)
The itk::RecursiveGaussianImageFilter has a boolean flag that allows users to select between
these two normalization options. Selection is done with the method SetNormalizeAcrossScale().
Enable this flag when analyzing an image across scale-space. In the current example, this setting
has no impact because we are actually renormalizing the output to the dynamic range of the reader,
so we simply disable the flag.
const bool normalizeAcrossScale = false;
filterX1->SetNormalizeAcrossScale( normalizeAcrossScale
filterY1->SetNormalizeAcrossScale( normalizeAcrossScale
filterX2->SetNormalizeAcrossScale( normalizeAcrossScale
filterY2->SetNormalizeAcrossScale( normalizeAcrossScale
);
);
);
);
The input image can be obtained from the output of another filter. Here, an image reader is used as
the source. The image is passed to the x filter and then to the y filter. The reason for keeping these
two filters separate is that it is usual in scale-space applications to compute not only the smoothing
but also combinations of derivatives at different orders and smoothing. Some factorization is possible when separate filters are used to generate the intermediate results. Here this capability is less
interesting, though, since we only want to smooth the image in all directions.
filterX1->SetInput( reader->GetOutput() );
filterY1->SetInput( filterX1->GetOutput() );
filterY2->SetInput( reader->GetOutput() );
filterX2->SetInput( filterY2->GetOutput() );
It is now time to select the σ of the Gaussian used to smooth the data. Note that σ must be passed to
both filters and that sigma is considered to be in millimeters. That is, at the moment of applying the
smoothing process, the filter will take into account the spacing values defined in the image.
filterX1->SetSigma(
filterY1->SetSigma(
filterX2->SetSigma(
filterY2->SetSigma(
sigma
sigma
sigma
sigma
);
);
);
);
Finally the two components of the Laplacian should be added together. The itk::AddImageFilter
is used for this purpose.
82
Chapter 2. Filtering
using AddFilterType = itk::AddImageFilter<
OutputImageType,
OutputImageType,
OutputImageType >;
AddFilterType::Pointer addFilter = AddFilterType::New();
addFilter->SetInput1( filterY1->GetOutput() );
addFilter->SetInput2( filterX2->GetOutput() );
The filters are triggered by invoking Update() on the Add filter at the end of the pipeline.
try
{
addFilter->Update();
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
The resulting image could be saved to a file using the itk::ImageFileWriter class.
using WritePixelType = float;
using WriteImageType = itk::Image< WritePixelType, 2 >;
using WriterType = itk::ImageFileWriter< WriteImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetInput( addFilter->GetOutput() );
writer->SetFileName( argv[2] );
writer->Update();
The source code for this section can be found in the file
LaplacianRecursiveGaussianImageFilter2.cxx.
The previous example showed how to use the itk::RecursiveGaussianImageFilter
for computing the equivalent of a Laplacian of an image after smoothing with a Gaussian. The elements used in this previous example have been packaged together in the
itk::LaplacianRecursiveGaussianImageFilter in order to simplify its usage. This current
example shows how to use this convenience filter for achieving the same results as the previous
example.
2.5. Second Order Derivatives
83
Figure 2.11: Effect of the LaplacianRecursiveGaussianImageFilter on a slice from a MRI proton density image
of the brain.
The first step required to use this filter is to include its header file.
#include "itkLaplacianRecursiveGaussianImageFilter.h"
Types should be selected on the desired input and output pixel types.
using InputPixelType = float;
using OutputPixelType = float;
The input and output image types are instantiated using the pixel types.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types.
using FilterType = itk::LaplacianRecursiveGaussianImageFilter<
InputImageType, OutputImageType >;
This filter packages all the components illustrated in the previous example. The filter is created by
invoking the New() method and assigning the result to a itk::SmartPointer.
84
Chapter 2. Filtering
FilterType::Pointer laplacian = FilterType::New();
The option for normalizing across scale space can also be selected in this filter.
laplacian->SetNormalizeAcrossScale( false );
The input image can be obtained from the output of another filter. Here, an image reader is used as
the source.
laplacian->SetInput( reader->GetOutput() );
It is now time to select the σ of the Gaussian used to smooth the data. Note that σ must be passed to
both filters and that sigma is considered to be in millimeters. That is, at the moment of applying the
smoothing process, the filter will take into account the spacing values defined in the image.
laplacian->SetSigma( sigma );
Finally the pipeline is executed by invoking the Update() method.
try
{
laplacian->Update();
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
2.6 Neighborhood Filters
The concept of locality is frequently encountered in image processing in the form of filters that
compute every output pixel using information from a small region in the neighborhood of the input
pixel. The classical form of these filters are the 3 × 3 filters in 2D images. Convolution masks
based on these neighborhoods can perform diverse tasks ranging from noise reduction, to differential
operations, to mathematical morphology.
The Insight Toolkit implements an elegant approach to neighborhood-based image filtering. The
input image is processed using a special iterator called the itk::NeighborhoodIterator. This
2.6. Neighborhood Filters
85
iterator is capable of moving over all the pixels in an image and, for each position, it can address
the pixels in a local neighborhood. Operators are defined that apply an algorithmic operation in
the neighborhood of the input pixel to produce a value for the output pixel. The following section
describes some of the more commonly used filters that take advantage of this construction. (See the
Iterators chapter in Book 1 for more information.)
2.6.1 Mean Filter
The source code for this section can be found in the file
MeanImageFilter.cxx.
The itk::MeanImageFilter is commonly used for noise reduction. The filter computes the value
of each output pixel by finding the statistical mean of the neighborhood of the corresponding input
pixel. The following figure illustrates the local effect of the MeanImageFilter in a 2D case. The
statistical mean of the neighborhood on the left is passed as the output value associated with the
pixel at the center of the neighborhood.
28
26
27
25
25
30
50
29
✲
30.22
✲
30
32
Note that this algorithm is sensitive to the presence of outliers in the neighborhood. This filter will work on images of any dimension thanks to the internal use of
itk::SmartNeighborhoodIterator and itk::NeighborhoodOperator. The size of the neighborhood over which the mean is computed can be set by the user.
The header file corresponding to this filter should be included first.
#include "itkMeanImageFilter.h"
Then the pixel types for input and output image must be defined and, with them, the image types
can be instantiated.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
Using the image types it is now possible to instantiate the filter type and create the filter object.
86
Chapter 2. Filtering
using FilterType = itk::MeanImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The size of the neighborhood is defined along every dimension by passing a SizeType object with
the corresponding values. The value on each dimension is used as the semi-size of a rectangular
box. For example, in 2D a size of 1, 2 will result in a 3 × 5 neighborhood.
InputImageType::SizeType indexRadius;
indexRadius[0] = 1; // radius along x
indexRadius[1] = 1; // radius along y
filter->SetRadius( indexRadius );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An update call on any downstream
filter will trigger the execution of the mean filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.12 illustrates the effect of this filter on a slice of MRI brain image using neighborhood radii
of 1, 1 which corresponds to a 3 × 3 classical neighborhood. It can be seen from this picture that
edges are rapidly degraded by the diffusion of intensity values among neighbors.
2.6.2 Median Filter
The source code for this section can be found in the file
MedianImageFilter.cxx.
The itk::MedianImageFilter is commonly used as a robust approach for noise reduction. This
filter is particularly efficient against salt-and-pepper noise. In other words, it is robust to the presence
of gray-level outliers. MedianImageFilter computes the value of each output pixel as the statistical
median of the neighborhood of values around the corresponding input pixel. The following figure
illustrates the local effect of this filter in a 2D case. The statistical median of the neighborhood on
the left is passed as the output value associated with the pixel at the center of the neighborhood.
28
26
50
27
25
25
30
29
32
✲
28
2.6. Neighborhood Filters
87
Figure 2.12: Effect of the MeanImageFilter on a slice from a MRI proton density brain image.
This filter will work on images of any dimension thanks to the internal use of
itk::NeighborhoodIterator and itk::NeighborhoodOperator. The size of the neighborhood
over which the median is computed can be set by the user.
The header file corresponding to this filter should be included first.
#include "itkMedianImageFilter.h"
Then the pixel and image types of the input and output must be defined.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
Using the image types, it is now possible to define the filter type and create the filter object.
using FilterType = itk::MedianImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The size of the neighborhood is defined along every dimension by passing a SizeType object with
88
Chapter 2. Filtering
Figure 2.13: Effect of the MedianImageFilter on a slice from a MRI proton density brain image.
the corresponding values. The value on each dimension is used as the semi-size of a rectangular
box. For example, in 2D a size of 1, 2 will result in a 3 × 5 neighborhood.
InputImageType::SizeType indexRadius;
indexRadius[0] = 1; // radius along x
indexRadius[1] = 1; // radius along y
filter->SetRadius( indexRadius );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An update call on any downstream
filter will trigger the execution of the median filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.13 illustrates the effect of the MedianImageFilter filter on a slice of MRI brain image using
a neighborhood radius of 1, 1, which corresponds to a 3 × 3 classical neighborhood. The filtered
image demonstrates the moderate tendency of the median filter to preserve edges.
2.6. Neighborhood Filters
89
2.6.3 Mathematical Morphology
Mathematical morphology has proved to be a powerful resource for image processing and analysis [55]. ITK implements mathematical morphology filters using NeighborhoodIterators and
itk::NeighborhoodOperators. The toolkit contains two types of image morphology algorithms:
filters that operate on binary images and filters that operate on grayscale images.
Binary Filters
The source code for this section can be found in the file
MathematicalMorphologyBinaryFilters.cxx.
The following section illustrates the use of filters that perform basic mathematical
morphology operations on binary images.
The
itk::BinaryErodeImageFilter and
itk::BinaryDilateImageFilter are described here. The filter names clearly specify the type
of image on which they operate. The header files required to construct a simple example of the use
of the mathematical morphology filters are included below.
#include "itkBinaryErodeImageFilter.h"
#include "itkBinaryDilateImageFilter.h"
#include "itkBinaryBallStructuringElement.h"
The following code defines the input and output pixel types and their associated image types.
constexpr unsigned int Dimension = 2;
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
Mathematical morphology operations are implemented by applying an operator over the neighborhood of each input pixel. The combination of the rule and the neighborhood is known as structuring
element. Although some rules have become de facto standards for image processing, there is a good
deal of freedom as to what kind of algorithmic rule should be applied to the neighborhood. The
implementation in ITK follows the typical rule of minimum for erosion and maximum for dilation.
The structuring element is implemented as a NeighborhoodOperator. In particular, the default structuring element is the itk::BinaryBallStructuringElement class. This class is instantiated
using the pixel type and dimension of the input image.
using StructuringElementType = itk::BinaryBallStructuringElement<
InputPixelType,
Dimension >;
90
Chapter 2. Filtering
The structuring element type is then used along with the input and output image types for instantiating the type of the filters.
using ErodeFilterType = itk::BinaryErodeImageFilter<
InputImageType,
OutputImageType,
StructuringElementType >;
using DilateFilterType = itk::BinaryDilateImageFilter<
InputImageType,
OutputImageType,
StructuringElementType >;
The filters can now be created by invoking the New() method and assigning the result to
itk::SmartPointers.
ErodeFilterType::Pointer binaryErode = ErodeFilterType::New();
DilateFilterType::Pointer binaryDilate = DilateFilterType::New();
The structuring element is not a reference counted class. Thus it is created as a C++ stack object
instead of using New() and SmartPointers. The radius of the neighborhood associated with the structuring element is defined with the SetRadius() method and the CreateStructuringElement()
method is invoked in order to initialize the operator. The resulting structuring element is passed to
the mathematical morphology filter through the SetKernel() method, as illustrated below.
StructuringElementType structuringElement;
structuringElement.SetRadius( 1 );
// 3x3 structuring element
structuringElement.CreateStructuringElement();
binaryErode->SetKernel( structuringElement );
binaryDilate->SetKernel( structuringElement );
A binary image is provided as input to the filters. This image might be, for example, the output of a
binary threshold image filter.
thresholder->SetInput( reader->GetOutput() );
InputPixelType background =
0;
InputPixelType foreground = 255;
thresholder->SetOutsideValue( background );
thresholder->SetInsideValue( foreground );
thresholder->SetLowerThreshold( lowerThreshold );
thresholder->SetUpperThreshold( upperThreshold );
2.6. Neighborhood Filters
91
Figure 2.14: Effect of erosion and dilation in a binary image.
binaryErode->SetInput( thresholder->GetOutput() );
binaryDilate->SetInput( thresholder->GetOutput() );
The values that correspond to “objects” in the binary image are specified with the methods
SetErodeValue() and SetDilateValue(). The value passed to these methods will be considered the value over which the dilation and erosion rules will apply.
binaryErode->SetErodeValue( foreground );
binaryDilate->SetDilateValue( foreground );
The filter is executed by invoking its Update() method, or by updating any downstream filter, such
as an image writer.
writerDilation->SetInput( binaryDilate->GetOutput() );
writerDilation->Update();
Figure 2.14 illustrates the effect of the erosion and dilation filters on a binary image from a MRI
brain slice. The figure shows how these operations can be used to remove spurious details from
segmented images.
Grayscale Filters
The source code for this section can be found in the file
MathematicalMorphologyGrayscaleFilters.cxx.
92
Chapter 2. Filtering
The following section illustrates the use of filters for performing basic mathematical morphology operations on grayscale images.
The itk::GrayscaleErodeImageFilter and
itk::GrayscaleDilateImageFilter are covered in this example. The filter names clearly specify the type of image on which they operate. The header files required for a simple example of the
use of grayscale mathematical morphology filters are presented below.
#include "itkGrayscaleErodeImageFilter.h"
#include "itkGrayscaleDilateImageFilter.h"
#include "itkBinaryBallStructuringElement.h"
The following code defines the input and output pixel types and their associated image types.
constexpr unsigned int Dimension = 2;
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
Mathematical morphology operations are based on the application of an operator over a neighborhood of each input pixel. The combination of the rule and the neighborhood is known as structuring
element. Although some rules have become the de facto standard in image processing there is a
good deal of freedom as to what kind of algorithmic rule should be applied on the neighborhood.
The implementation in ITK follows the typical rule of minimum for erosion and maximum for dilation.
The structuring element is implemented as a itk::NeighborhoodOperator. In particular, the
default structuring element is the itk::BinaryBallStructuringElement class. This class is
instantiated using the pixel type and dimension of the input image.
using StructuringElementType = itk::BinaryBallStructuringElement<
InputPixelType,
Dimension >;
The structuring element type is then used along with the input and output image types for instantiating the type of the filters.
using ErodeFilterType = itk::GrayscaleErodeImageFilter<
InputImageType,
OutputImageType,
StructuringElementType >;
using DilateFilterType = itk::GrayscaleDilateImageFilter<
InputImageType,
OutputImageType,
StructuringElementType >;
2.6. Neighborhood Filters
93
The filters can now be created by invoking the New() method and assigning the result to SmartPointers.
ErodeFilterType::Pointer grayscaleErode = ErodeFilterType::New();
DilateFilterType::Pointer grayscaleDilate = DilateFilterType::New();
The structuring element is not a reference counted class. Thus it is created as a C++ stack object
instead of using New() and SmartPointers. The radius of the neighborhood associated with the structuring element is defined with the SetRadius() method and the CreateStructuringElement()
method is invoked in order to initialize the operator. The resulting structuring element is passed to
the mathematical morphology filter through the SetKernel() method, as illustrated below.
StructuringElementType structuringElement;
structuringElement.SetRadius( 1 );
// 3x3 structuring element
structuringElement.CreateStructuringElement();
grayscaleErode->SetKernel( structuringElement );
grayscaleDilate->SetKernel( structuringElement );
A grayscale image is provided as input to the filters. This image might be, for example, the output
of a reader.
grayscaleErode->SetInput( reader->GetOutput() );
grayscaleDilate->SetInput( reader->GetOutput() );
The filter is executed by invoking its Update() method, or by updating any downstream filter, such
as an image writer.
writerDilation->SetInput( grayscaleDilate->GetOutput() );
writerDilation->Update();
Figure 2.15 illustrates the effect of the erosion and dilation filters on a binary image from a MRI
brain slice. The figure shows how these operations can be used to remove spurious details from
segmented images.
2.6.4 Voting Filters
Voting filters are quite a generic family of filters. In fact, both the Dilate and Erode filters from
Mathematical Morphology are very particular cases of the broader family of voting filters. In a voting filter, the outcome of a pixel is decided by counting the number of pixels in its neighborhood and
applying a rule to the result of that counting. For example, the typical implementation of erosion in
94
Chapter 2. Filtering
Figure 2.15: Effect of erosion and dilation in a grayscale image.
terms of a voting filter will be to label a foreground pixel as background if the number of background
neighbors is greater than or equal to 1. In this context, you could imagine variations of erosion in
which the count could be changed to require at least 3 foreground pixels in its neighborhood.
Binary Median Filter
One case of a voting filter is the BinaryMedianImageFilter. This filter is equivalent to applying a
Median filter over a binary image. Having a binary image as input makes it possible to optimize the
execution of the filter since there is no real need for sorting the pixels according to their frequency
in the neighborhood.
The source code for this section can be found in the file
BinaryMedianImageFilter.cxx.
The itk::BinaryMedianImageFilter is commonly used as a robust approach for noise reduction.
BinaryMedianImageFilter computes the value of each output pixel as the statistical median of the
neighborhood of values around the corresponding input pixel. When the input images are binary,
the implementation can be optimized by simply counting the number of pixels ON/OFF around the
current pixel.
This filter will work on images of any dimension thanks to the internal use of
itk::NeighborhoodIterator and itk::NeighborhoodOperator. The size of the neighborhood
over which the median is computed can be set by the user.
The header file corresponding to this filter should be included first.
#include "itkBinaryMedianImageFilter.h"
2.6. Neighborhood Filters
95
Then the pixel and image types of the input and output must be defined.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
Using the image types, it is now possible to define the filter type and create the filter object.
using FilterType = itk::BinaryMedianImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The size of the neighborhood is defined along every dimension by passing a SizeType object with
the corresponding values. The value on each dimension is used as the semi-size of a rectangular
box. For example, in 2D a size of 1, 2 will result in a 3 × 5 neighborhood.
InputImageType::SizeType indexRadius;
indexRadius[0] = radiusX; // radius along x
indexRadius[1] = radiusY; // radius along y
filter->SetRadius( indexRadius );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An update call on any downstream
filter will trigger the execution of the median filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.16 illustrates the effect of the BinaryMedianImageFilter filter on a slice of MRI brain image
using a neighborhood radius of 2, 2, which corresponds to a 5 × 5 classical neighborhood. The
filtered image demonstrates the capability of this filter for reducing noise both in the background
and foreground of the image, as well as smoothing the contours of the regions.
The typical effect of median filtration on a noisy digital image is a dramatic reduction in impulse
noise spikes. The filter also tends to preserve brightness differences across signal steps, resulting in
reduced blurring of regional boundaries. The filter also tends to preserve the positions of boundaries
in an image.
Figure 2.17 below shows the effect of running the median filter with a 3x3 classical window size
1, 10 and 50 times. There is a tradeoff in noise reduction and the sharpness of the image when the
window size is increased.
96
Chapter 2. Filtering
Figure 2.16: Effect of the BinaryMedianImageFilter on a slice from a MRI proton density brain image that has
been thresholded in order to produce a binary image.
Hole Filling Filter
Another variation of voting filters is the Hole Filling filter. This filter converts background pixels into
foreground only when the number of foreground pixels is a majority of the neighbors. By selecting
the size of the majority, this filter can be tuned to fill in holes of different sizes. To be more precise,
the effect of the filter is actually related to the curvature of the edge in which the pixel is located.
The source code for this section can be found in the file
VotingBinaryHoleFillingImageFilter.cxx.
The itk::VotingBinaryHoleFillingImageFilter applies a voting operation in order to fill in
cavities. This can be used for smoothing contours and for filling holes in binary images.
The header file corresponding to this filter should be included first.
#include "itkVotingBinaryHoleFillingImageFilter.h"
Then the pixel and image types of the input and output must be defined.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
2.6. Neighborhood Filters
Figure 2.17: Effect of 1, 10 and 50 iterations of the BinaryMedianImageFilter using a 3x3 window.
97
98
Chapter 2. Filtering
Using the image types, it is now possible to define the filter type and create the filter object.
using FilterType =
itk::VotingBinaryHoleFillingImageFilter< InputImageType,
OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The size of the neighborhood is defined along every dimension by passing a SizeType object with
the corresponding values. The value on each dimension is used as the semi-size of a rectangular
box. For example, in 2D a size of 1, 2 will result in a 3 × 5 neighborhood.
InputImageType::SizeType indexRadius;
indexRadius[0] = radiusX; // radius along x
indexRadius[1] = radiusY; // radius along y
filter->SetRadius( indexRadius );
Since the filter is expecting a binary image as input, we must specify the levels that are going to
be considered background and foreground. This is done with the SetForegroundValue() and
SetBackgroundValue() methods.
filter->SetBackgroundValue( 0 );
filter->SetForegroundValue( 255 );
We must also specify the majority threshold that is going to be used as the decision criterion for
converting a background pixel into a foreground pixel. The rule of conversion is that a background
pixel will be converted into a foreground pixel if the number of foreground neighbors surpass the
number of background neighbors by the majority value. For example, in a 2D image, with neighborhood of radius 1, the neighborhood will have size 3 × 3. If we set the majority value to 2, then
we are requiring that the number of foreground neighbors should be at least (3x3 -1 )/2 + majority.
This is done with the SetMajorityThreshold() method.
filter->SetMajorityThreshold( 2 );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An update call on any downstream
filter will trigger the execution of the median filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
2.6. Neighborhood Filters
99
Figure 2.18: Effect of the VotingBinaryHoleFillingImageFilter on a slice from a MRI proton density brain image
that has been thresholded in order to produce a binary image. The output images have used radius 1,2 and 3
respectively.
100
Chapter 2. Filtering
Figure 2.18 illustrates the effect of the VotingBinaryHoleFillingImageFilter filter on a thresholded
slice of MRI brain image using neighborhood radii of 1, 1, 2, 2 and 3, 3 that correspond respectively
to neighborhoods of size 3 × 3, 5 × 5, 7 × 7. The filtered image demonstrates the capability of this
filter for reducing noise both in the background and foreground of the image, as well as smoothing
the contours of the regions.
Iterative Hole Filling Filter
The Hole Filling filter can be used in an iterative way, by applying it repeatedly until no pixel
changes. In this context, the filter can be seen as a binary variation of a Level Set filter.
The source code for this section can be found in the file
VotingBinaryIterativeHoleFillingImageFilter.cxx.
The itk::VotingBinaryIterativeHoleFillingImageFilter applies a voting operation in order to fill in cavities. This can be used for smoothing contours and for filling holes in binary images.
This filter runs a itk::VotingBinaryHoleFillingImageFilter internally until no pixels change
or the maximum number of iterations has been reached.
The header file corresponding to this filter should be included first.
#include "itkVotingBinaryIterativeHoleFillingImageFilter.h"
Then the pixel and image types must be defined. Note that this filter requires the input and output
images to be of the same type, therefore a single image type is required for the template instantiation.
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, 2 >;
Using the image types, it is now possible to define the filter type and create the filter object.
using FilterType =
itk::VotingBinaryIterativeHoleFillingImageFilter;
FilterType::Pointer filter = FilterType::New();
The size of the neighborhood is defined along every dimension by passing a SizeType object with
the corresponding values. The value on each dimension is used as the semi-size of a rectangular
box. For example, in 2D a size of 1, 2 will result in a 3 × 5 neighborhood.
ImageType::SizeType indexRadius;
indexRadius[0] = radiusX; // radius along x
indexRadius[1] = radiusY; // radius along y
2.6. Neighborhood Filters
101
filter->SetRadius( indexRadius );
Since the filter is expecting a binary image as input, we must specify the levels that are going to
be considered background and foreground. This is done with the SetForegroundValue() and
SetBackgroundValue() methods.
filter->SetBackgroundValue( 0 );
filter->SetForegroundValue( 255 );
We must also specify the majority threshold that is going to be used as the decision criterion for
converting a background pixel into a foreground pixel. The rule of conversion is that a background
pixel will be converted into a foreground pixel if the number of foreground neighbors surpass the
number of background neighbors by the majority value. For example, in a 2D image, with neighborhood of radius 1, the neighborhood will have size 3 × 3. If we set the majority value to 2, then
we are requiring that the number of foreground neighbors should be at least (3x3 -1 )/2 + majority.
This is done with the SetMajorityThreshold() method.
filter->SetMajorityThreshold( 2 );
Finally we specify the maximum number of iterations for which this filter should run. The number
of iterations will determine the maximum size of holes and cavities that this filter will be able to fill.
The more iterations you run, the larger the cavities that will be filled in.
filter->SetMaximumNumberOfIterations( numberOfIterations );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example, a writer. An update call on any downstream
filter will trigger the execution of the median filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.19 illustrates the effect of the VotingBinaryIterativeHoleFillingImageFilter filter on a
thresholded slice of MRI brain image using neighborhood radii of 1, 1, 2, 2 and 3, 3 that correspond respectively to neighborhoods of size 3 × 3, 5 × 5, 7 × 7. The filtered image demonstrates the
capability of this filter for reducing noise both in the background and foreground of the image, as
well as smoothing the contours of the regions.
102
Chapter 2. Filtering
Figure 2.19: Effect of the VotingBinaryIterativeHoleFillingImageFilter on a slice from a MRI proton density brain
image that has been thresholded in order to produce a binary image. The output images have used radius 1,2
and 3 respectively.
2.7. Smoothing Filters
103
2.7 Smoothing Filters
Real image data has a level of uncertainty which is manifested in the variability of measures assigned
to pixels. This uncertainty is usually interpreted as noise and considered an undesirable component
of the image data. This section describes several methods that can be applied to reduce noise on
images.
2.7.1 Blurring
Blurring is the traditional approach for removing noise from images. It is usually implemented in
the form of a convolution with a kernel. The effect of blurring on the image spectrum is to attenuate
high spatial frequencies. Different kernels attenuate frequencies in different ways. One of the most
commonly used kernels is the Gaussian. Two implementations of Gaussian smoothing are available
in the toolkit. The first one is based on a traditional convolution while the other is based on the
application of IIR filters that approximate the convolution with a Gaussian [15, 16].
Discrete Gaussian
The source code for this section can be found in the file
DiscreteGaussianImageFilter.cxx.
The itk::DiscreteGaussianImageFilter com1
putes the convolution of the input image with a Gaus- 0.9
sian kernel. This is done in ND by taking advantage 0.8
Error
of the separability of the Gaussian kernel. A one- 0.7
dimensional Gaussian function is discretized on a con- 0.6
0.5
volution kernel. The size of the kernel is extended 0.4
until there are enough discrete points in the Gaussian 0.3
to ensure that a user-provided maximum error is not 0.2
exceeded. Since the size of the kernel is unknown a 0.10
−2
−1.5
−1
−0.5
0
0.5
1
1.5
priori, it is necessary to impose a limit to its growth.
KernelWidth
The user can thus provide a value to be the maximum
admissible size of the kernel. Discretization error is
Figure 2.20: Discretized Gaussian.
defined as the difference between the area under the
discrete Gaussian curve (which has finite support) and the area under the continuous Gaussian.
2
Gaussian kernels in ITK are constructed according to the theory of Tony Lindeberg [34] so that
smoothing and derivative operations commute before and after discretization. In other words, finite
difference derivatives on an image I that has been smoothed by convolution with the Gaussian are
equivalent to finite differences computed on I by convolving with a derivative of the Gaussian.
The first step required to use this filter is to include its header file. As with other examples, the
includes here are truncated to those specific for this example.
104
Chapter 2. Filtering
#include "itkDiscreteGaussianImageFilter.h"
Types should be chosen for the pixels of the input and output images. Image types can be instantiated
using the pixel type and dimension.
using InputPixelType = float;
using OutputPixelType = float;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The discrete Gaussian filter type is instantiated using the input and output image types. A corresponding filter object is created.
using FilterType = itk::DiscreteGaussianImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
its input.
filter->SetInput( reader->GetOutput() );
The filter requires the user to provide a value for the variance associated with the Gaussian kernel.
The method SetVariance() is used for this purpose. The discrete Gaussian is constructed as a
convolution kernel. The maximum kernel size can be set by the user. Note that the combination of
variance and kernel-size values may result in a truncated Gaussian kernel.
filter->SetVariance( gaussianVariance );
filter->SetMaximumKernelWidth( maxKernelWidth );
Finally, the filter is executed by invoking the Update() method.
filter->Update();
If the output of this filter has been connected to other filters down the pipeline, updating any of the
downstream filters will trigger the execution of this one. For example, a writer could be used after
the filter.
2.7. Smoothing Filters
105
Figure 2.21: Effect of the DiscreteGaussianImageFilter on a slice from a MRI proton density image of the brain.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.21 illustrates the effect of this filter on a MRI proton density image of the brain.
Note that large Gaussian variances will produce large convolution kernels and correspondingly
longer computation times. Unless a high degree of accuracy is required, it may be more desirable to
use the approximating itk::RecursiveGaussianImageFilter with large variances.
Binomial Blurring
The source code for this section can be found in the file
BinomialBlurImageFilter.cxx.
The itk::BinomialBlurImageFilter computes a nearest neighbor average along each dimension. The process is repeated a number of times, as specified by the user. In principle, after a large
number of iterations the result will approach the convolution with a Gaussian.
The first step required to use this filter is to include its header file.
#include "itkBinomialBlurImageFilter.h"
106
Chapter 2. Filtering
Types should be chosen for the pixels of the input and output images. Image types can be instantiated
using the pixel type and dimension.
using InputPixelType = float;
using OutputPixelType = float;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types. Then a
filter object is created.
using FilterType = itk::BinomialBlurImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used
as the source. The number of repetitions is set with the SetRepetitions() method. Computation
time will increase linearly with the number of repetitions selected. Finally, the filter can be executed
by calling the Update() method.
filter->SetInput( reader->GetOutput() );
filter->SetRepetitions( repetitions );
filter->Update();
Figure 2.22 illustrates the effect of this filter on a MRI proton density image of the brain.
Note that the standard deviation σ of the equivalent Gaussian is fixed. In the spatial spectrum, the
effect of every iteration of this filter is like a multiplication with a sinus cardinal function.
Recursive Gaussian IIR
The source code for this section can be found in the file
SmoothingRecursiveGaussianImageFilter.cxx.
The classical method of smoothing an image by convolution with a Gaussian kernel has the drawback that it is slow when the standard deviation σ of the Gaussian is large. This is due to the larger
size of the kernel, which results in a higher number of computations per pixel.
The itk::RecursiveGaussianImageFilter implements an approximation of convolution with
the Gaussian and its derivatives by using IIR2 filters. In practice this filter requires a constant number
of operations for approximating the convolution, regardless of the σ value [15, 16].
The first step required to use this filter is to include its header file.
2 Infinite
Impulse Response
2.7. Smoothing Filters
107
Figure 2.22: Effect of the BinomialBlurImageFilter on a slice from a MRI proton density image of the brain.
#include "itkRecursiveGaussianImageFilter.h"
Types should be selected on the desired input and output pixel types.
using InputPixelType = float;
using OutputPixelType = float;
The input and output image types are instantiated using the pixel types.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types.
using FilterType = itk::RecursiveGaussianImageFilter<
InputImageType, OutputImageType >;
This filter applies the approximation of the convolution along a single dimension. It is therefore
necessary to concatenate several of these filters to produce smoothing in all directions. In this
example, we create a pair of filters since we are processing a 2D image. The filters are created by
invoking the New() method and assigning the result to a itk::SmartPointer.
108
Chapter 2. Filtering
FilterType::Pointer filterX = FilterType::New();
FilterType::Pointer filterY = FilterType::New();
Since each one of the newly created filters has the potential to perform filtering along any dimension,
we have to restrict each one to a particular direction. This is done with the SetDirection() method.
filterX->SetDirection( 0 );
filterY->SetDirection( 1 );
// 0 --> X direction
// 1 --> Y direction
The itk::RecursiveGaussianImageFilter can approximate the convolution with the Gaussian
or with its first and second derivatives. We select one of these options by using the SetOrder()
method. Note that the argument is an enum whose values can be ZeroOrder, FirstOrder and
SecondOrder. For example, to compute the x partial derivative we should select FirstOrder for
x and ZeroOrder for y. Here we want only to smooth in x and y, so we select ZeroOrder in both
directions.
filterX->SetOrder( FilterType::ZeroOrder );
filterY->SetOrder( FilterType::ZeroOrder );
There are two typical ways of normalizing Gaussians depending on their application. For scalespace analysis it is desirable to use a normalization that will preserve the maximum value of the
input. This normalization is represented by the following equation.
1
√
σ 2π
(2.4)
In applications that use the Gaussian as a solution of the diffusion equation it is desirable to use a
normalization that preserve the integral of the signal. This last approach can be seen as a conservation of mass principle. This is represented by the following equation.
σ2
1
√
2π
(2.5)
The itk::RecursiveGaussianImageFilter has a boolean flag that allows users to select between
these two normalization options. Selection is done with the method SetNormalizeAcrossScale().
Enable this flag to analyzing an image across scale-space. In the current example, this setting has
no impact because we are actually renormalizing the output to the dynamic range of the reader, so
we simply disable the flag.
filterX->SetNormalizeAcrossScale( false );
filterY->SetNormalizeAcrossScale( false );
2.7. Smoothing Filters
109
Figure 2.23: Effect of the SmoothingRecursiveGaussianImageFilter on a slice from a MRI proton density image
of the brain.
The input image can be obtained from the output of another filter. Here, an image reader is used as
the source. The image is passed to the x filter and then to the y filter. The reason for keeping these
two filters separate is that it is usual in scale-space applications to compute not only the smoothing
but also combinations of derivatives at different orders and smoothing. Some factorization is possible when separate filters are used to generate the intermediate results. Here this capability is less
interesting, though, since we only want to smooth the image in all directions.
filterX->SetInput( reader->GetOutput() );
filterY->SetInput( filterX->GetOutput() );
It is now time to select the σ of the Gaussian used to smooth the data. Note that σ must be passed to
both filters and that sigma is considered to be in millimeters. That is, at the moment of applying the
smoothing process, the filter will take into account the spacing values defined in the image.
filterX->SetSigma( sigma );
filterY->SetSigma( sigma );
Finally the pipeline is executed by invoking the Update() method.
filterY->Update();
110
Chapter 2. Filtering
Figure 2.23 illustrates the effect of this filter on a MRI proton density image of the brain using σ
values of 3 (left) and 5 (right). The figure shows how the attenuation of noise can be regulated by selecting the appropriate standard deviation. This type of scale-tunable filter is suitable for performing
scale-space analysis.
The RecursiveGaussianFilters can also be applied on multi-component images. For instance, the
above filter could have applied with RGBPixel as the pixel type. Each component is then independently filtered. However the RescaleIntensityImageFilter will not work on RGBPixels since it does
not mathematically make sense to rescale the output of multi-component images.
2.7.2 Local Blurring
In some cases it is desirable to compute smoothing in restricted regions of the image, or to do it
using different parameters that are computed locally. The following sections describe options for
applying local smoothing in images.
Gaussian Blur Image Function
The source code for this section can be found in the file
GaussianBlurImageFunction.cxx.
2.7.3 Edge Preserving Smoothing
Introduction to Anisotropic Diffusion
The drawback of image denoising (smoothing) is that it tends to blur away the sharp boundaries in
the image that help to distinguish between the larger-scale anatomical structures that one is trying
to characterize (which also limits the size of the smoothing kernels in most applications). Even in
cases where smoothing does not obliterate boundaries, it tends to distort the fine structure of the
image and thereby changes subtle aspects of the anatomical shapes in question.
Perona and Malik [45] introduced an alternative to linear-filtering that they called anisotropic diffusion. Anisotropic diffusion is closely related to the earlier work of Grossberg [22], who used similar
nonlinear diffusion processes to model human vision. The motivation for anisotropic diffusion (also
called nonuniform or variable conductance diffusion) is that a Gaussian smoothed image is a single
time slice of the solution to the heat equation, that has the original image as its initial conditions.
Thus, the solution to
∂g(x, y,t)
= ∇ · ∇g(x, y,t),
(2.6)
∂t
√
where g(x, y, 0) = f (x, y) is the input image, is g(x, y,t) = G( 2t) ⊗ f (x, y), where G(σ) is a Gaussian with standard deviation σ.
2.7. Smoothing Filters
111
Anisotropic diffusion includes a variable conductance term that, in turn, depends on the differential
structure of the image. Thus, the variable conductance can be formulated to limit the smoothing at
“edges” in images, as measured by high gradient magnitude, for example.
gt = ∇ · c(|∇g|)∇g,
(2.7)
where, for notational convenience, we leave off the independent parameters of g and use the subscripts with respect to those parameters to indicate partial derivatives. The function c(|∇g|) is a
fuzzy cutoff that reduces the conductance at areas of large |∇g|, and can be any one of a number of
functions. The literature has shown
c(|∇g|) = e
−
|∇g|2
2k2
(2.8)
to be quite effective. Notice that conductance term introduces a free parameter k, the conductance
parameter, that controls the sensitivity of the process to edge contrast. Thus, anisotropic diffusion entails two free parameters: the conductance parameter, k, and the time parameter, t, that is
analogous to σ, the effective width of the filter when using Gaussian kernels.
Equation 2.7 is a nonlinear partial differential equation that can be solved on a discrete grid using
finite forward differences. Thus, the smoothed image is obtained only by an iterative process, not a
convolution or non-stationary, linear filter. Typically, the number of iterations required for practical
results are small, and large 2D images can be processed in several tens of seconds using carefully
written code running on modern, general purpose, single-processor computers. The technique applies readily and effectively to 3D images, but requires more processing time.
In the early 1990’s several research groups [21, 67] demonstrated the effectiveness of anisotropic
diffusion on medical images. In a series of papers on the subject [71, 69, 70, 67, 68, 65], Whitaker
described a detailed analytical and empirical analysis, introduced a smoothing term in the conductance that made the process more robust, invented a numerical scheme that virtually eliminated
directional artifacts in the original algorithm, and generalized anisotropic diffusion to vector-valued
images, an image processing technique that can be used on vector-valued medical data (such as the
color cryosection data of the Visible Human Project).
For a vector-valued input ~F : U 7→ ℜm the process takes the form
~Ft = ∇ · c(D ~F)~F,
(2.9)
where D ~F is a dissimilarity measure of ~F, a generalization of the gradient magnitude to vectorvalued images, that can incorporate linear and nonlinear coordinate transformations on the range of
~F. In this way, the smoothing of the multiple images associated with vector-valued data is coupled
through the conductance term, that fuses the information in the different images. Thus vectorvalued, nonlinear diffusion can combine low-level image features (e.g. edges) across all “channels”
of a vector-valued image in order to preserve or enhance those features in all of image “channels”.
Vector-valued anisotropic diffusion is useful for denoising data from devices that produce multiple
values such as MRI or color photography. When performing nonlinear diffusion on a color image,
the color channels are diffused separately, but linked through the conductance term. Vector-valued
112
Chapter 2. Filtering
diffusion is also useful for processing registered data from different devices or for denoising higherorder geometric or statistical features from scalar-valued images [65, 72].
The output of anisotropic diffusion is an image or set of images that demonstrates reduced noise and
texture but preserves, and can also enhance, edges. Such images are useful for a variety of processes
including statistical classification, visualization, and geometric feature extraction. Previous work
has shown [68] that anisotropic diffusion, over a wide range of conductance parameters, offers
quantifiable advantages over linear filtering for edge detection in medical images.
Since the effectiveness of nonlinear diffusion was first demonstrated, numerous variations of this approach have surfaced in the literature [60]. These include alternatives for constructing dissimilarity
measures [53], directional (i.e., tensor-valued) conductance terms [64, 3] and level set interpretations
[66].
Gradient Anisotropic Diffusion
The source code for this section can be found in the file
GradientAnisotropicDiffusionImageFilter.cxx.
The itk::GradientAnisotropicDiffusionImageFilter implements an N-dimensional version
of the classic Perona-Malik anisotropic diffusion equation for scalar-valued images [45].
The conductance term for this implementation is chosen as a function of the gradient magnitude of
the image at each point, reducing the strength of diffusion at edge pixels.
C(x) = e−(
k∇U (x)k 2
)
K
(2.10)
The numerical implementation of this equation is similar to that described in the Perona-Malik paper
[45], but uses a more robust technique for gradient magnitude estimation and has been generalized
to N-dimensions.
The first step required to use this filter is to include its header file.
#include "itkGradientAnisotropicDiffusionImageFilter.h"
Types should be selected based on the pixel types required for the input and output images. The
image types are defined using the pixel type and the dimension.
using InputPixelType = float;
using OutputPixelType = float;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types. The filter
object is created by the New() method.
2.7. Smoothing Filters
113
using FilterType = itk::GradientAnisotropicDiffusionImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
filter->SetInput( reader->GetOutput() );
This filter requires three parameters: the number of iterations to be performed, the time
step and the conductance parameter used in the computation of the level set evolution.
These parameters are set using the methods SetNumberOfIterations(), SetTimeStep() and
SetConductanceParameter() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter( conductance );
filter->Update();
Typical values for the time step are 0.25 in 2D images and 0.125 in 3D images. The number of
iterations is typically set to 5; more iterations result in further smoothing and will increase the
computing time linearly.
Figure 2.24 illustrates the effect of this filter on a MRI proton density image of the brain. In this
example the filter was run with a time step of 0.25, and 5 iterations. The figure shows how homogeneous regions are smoothed and edges are preserved.
The following classes provide similar functionality:
• itk::BilateralImageFilter
• itk::CurvatureAnisotropicDiffusionImageFilter
• itk::CurvatureFlowImageFilter
Curvature Anisotropic Diffusion
The source code for this section can be found in the file
CurvatureAnisotropicDiffusionImageFilter.cxx.
The itk::CurvatureAnisotropicDiffusionImageFilter performs anisotropic diffusion on an
image using a modified curvature diffusion equation (MCDE).
114
Chapter 2. Filtering
Figure 2.24: Effect of the GradientAnisotropicDiffusionImageFilter on a slice from a MRI Proton Density image
of the brain.
MCDE does not exhibit the edge enhancing properties of classic anisotropic diffusion, which can under certain conditions undergo a “negative” diffusion, which enhances the contrast of edges. Equations of the form of MCDE always undergo positive diffusion, with the conductance term only
varying the strength of that diffusion.
Qualitatively, MCDE compares well with other non-linear diffusion techniques. It is less sensitive to
contrast than classic Perona-Malik style diffusion, and preserves finer detailed structures in images.
There is a potential speed trade-off for using this function in place of itkGradientNDAnisotropicDiffusionFunction. Each iteration of the solution takes roughly twice as long. Fewer iterations,
however, may be required to reach an acceptable solution.
The MCDE equation is given as:
ft =| ∇ f | ∇ · c(| ∇ f |)
∇f
| ∇f |
(2.11)
where the conductance modified curvature term is
∇·
∇f
| ∇f |
The first step required for using this filter is to include its header file.
(2.12)
2.7. Smoothing Filters
115
#include "itkCurvatureAnisotropicDiffusionImageFilter.h"
Types should be selected based on the pixel types required for the input and output images. The
image types are defined using the pixel type and the dimension.
using InputPixelType = float;
using OutputPixelType = float;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types. The filter
object is created by the New() method.
using FilterType = itk::CurvatureAnisotropicDiffusionImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
filter->SetInput( reader->GetOutput() );
This filter requires three parameters: the number of iterations to be performed, the time step used in
the computation of the level set evolution and the value of conductance. These parameters are set
using the methods SetNumberOfIterations(), SetTimeStep() and SetConductance() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter( conductance );
if (useImageSpacing)
{
filter->UseImageSpacingOn();
}
filter->Update();
Typical values for the time step are 0.125 in 2D images and 0.0625 in 3D images. The number of
iterations can be usually around 5, more iterations will result in further smoothing and will increase
the computing time linearly. The conductance parameter is usually around 3.0.
Figure 2.25 illustrates the effect of this filter on a MRI proton density image of the brain. In this
example the filter was run with a time step of 0.125, 5 iterations and a conductance value of 3.0.
The figure shows how homogeneous regions are smoothed and edges are preserved.
116
Chapter 2. Filtering
Figure 2.25: Effect of the CurvatureAnisotropicDiffusionImageFilter on a slice from a MRI Proton Density image
of the brain.
The following classes provide similar functionality:
• itk::BilateralImageFilter
• itk::CurvatureFlowImageFilter
• itk::GradientAnisotropicDiffusionImageFilter
Curvature Flow
The source code for this section can be found in the file
CurvatureFlowImageFilter.cxx.
The itk::CurvatureFlowImageFilter performs edge-preserving smoothing in a similar fashion
to the classical anisotropic diffusion. The filter uses a level set formulation where the iso-intensity
contours in an image are viewed as level sets, where pixels of a particular intensity form one level
set. The level set function is then evolved under the control of a diffusion equation where the speed
is proportional to the curvature of the contour:
It = κ|∇I|
where κ is the curvature.
(2.13)
2.7. Smoothing Filters
117
Areas of high curvature will diffuse faster than areas of low curvature. Hence, small jagged noise
artifacts will disappear quickly, while large scale interfaces will be slow to evolve, thereby preserving sharp boundaries between objects. However, it should be noted that although the evolution at
the boundary is slow, some diffusion will still occur. Thus, continual application of this curvature
flow scheme will eventually result in the removal of information as each contour shrinks to a point
and disappears.
The first step required to use this filter is to include its header file.
#include "itkCurvatureFlowImageFilter.h"
Types should be selected based on the pixel types required for the input and output images.
using InputPixelType = float;
using OutputPixelType = float;
With them, the input and output image types can be instantiated.
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The CurvatureFlow filter type is now instantiated using both the input image and the output image
types.
using FilterType = itk::CurvatureFlowImageFilter<
InputImageType, OutputImageType >;
A filter object is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
filter->SetInput( reader->GetOutput() );
The CurvatureFlow filter requires two parameters: the number of iterations to be performed and the
time step used in the computation of the level set evolution. These two parameters are set using
the methods SetNumberOfIterations() and SetTimeStep() respectively. Then the filter can be
executed by invoking Update().
118
Chapter 2. Filtering
Figure 2.26: Effect of the CurvatureFlowImageFilter on a slice from a MRI proton density image of the brain.
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->Update();
Typical values for the time step are 0.125 in 2D images and 0.0625 in 3D images. The number
of iterations can be usually around 10, more iterations will result in further smoothing and will
increase the computing time linearly. Edge-preserving behavior is not guaranteed by this filter.
Some degradation will occur on the edges and will increase as the number of iterations is increased.
If the output of this filter has been connected to other filters down the pipeline, updating any of the
downstream filters will trigger the execution of this one. For example, a writer filter could be used
after the curvature flow filter.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.26 illustrates the effect of this filter on a MRI proton density image of the brain. In this
example the filter was run with a time step of 0.25 and 10 iterations. The figure shows how homogeneous regions are smoothed and edges are preserved.
The following classes provide similar functionality:
2.7. Smoothing Filters
119
−c
Iso
s
ve
ur
Gradient
Figure 2.27: Elements involved in the computation of min-max curvature flow.
• itk::GradientAnisotropicDiffusionImageFilter
• itk::CurvatureAnisotropicDiffusionImageFilter
• itk::BilateralImageFilter
MinMaxCurvature Flow
The source code for this section can be found in the file
MinMaxCurvatureFlowImageFilter.cxx.
The MinMax curvature flow filter applies a variant of the curvature flow algorithm where diffusion
is turned on or off depending of the scale of the noise that one wants to remove. The evolution speed
is switched between min(κ, 0) and max(κ, 0) such that:
It = F|∇I|
(2.14)
where F is defined as
F=
max(κ, 0) : Average < T hreshold
min(κ, 0) : Average ≥ T hreshold
(2.15)
The Average is the average intensity computed over a neighborhood of a user-specified radius of
the pixel. The choice of the radius governs the scale of the noise to be removed. The T hreshold is
120
Chapter 2. Filtering
calculated as the average of pixel intensities along the direction perpendicular to the gradient at the
extrema of the local neighborhood.
A speed of F = max(κ, 0) will cause small dark regions in a predominantly light region to shrink.
Conversely, a speed of F = min(κ, 0), will cause light regions in a predominantly dark region to
shrink. Comparison between the neighborhood average and the threshold is used to select the the
right speed function to use. This switching prevents the unwanted diffusion of the simple curvature
flow method.
Figure 2.27 shows the main elements involved in the computation. The set of square pixels represent
the neighborhood over which the average intensity is being computed. The gray pixels are those lying close to the direction perpendicular to the gradient. The pixels which intersect the neighborhood
bounds are used to compute the threshold value in the equation above. The integer radius of the
neighborhood is selected by the user.
The first step required to use the itk::MinMaxCurvatureFlowImageFilter is to include its header
file.
#include "itkMinMaxCurvatureFlowImageFilter.h"
Types should be selected based on the pixel types required for the input and output images. The
input and output image types are instantiated.
using InputPixelType = float;
using OutputPixelType = float;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The itk::MinMaxCurvatureFlowImageFilter type is now instantiated using both the input image and the output image types. The filter is then created using the New() method.
using FilterType = itk::MinMaxCurvatureFlowImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
filter->SetInput( reader->GetOutput() );
The itk::MinMaxCurvatureFlowImageFilter requires the two normal parameters of the CurvatureFlow image, the number of iterations to be performed and the time step used in the computation
of the level set evolution. In addition, the radius of the neighborhood is also required. This last
parameter is passed using the SetStencilRadius() method. Note that the radius is provided as
2.7. Smoothing Filters
121
Figure 2.28: Effect of the MinMaxCurvatureFlowImageFilter on a slice from a MRI proton density image of the
brain.
an integer number since it is referring to a number of pixels from the center to the border of the
neighborhood. Then the filter can be executed by invoking Update().
filter->SetTimeStep( timeStep );
filter->SetNumberOfIterations( numberOfIterations );
filter->SetStencilRadius( radius );
filter->Update();
Typical values for the time step are 0.125 in 2D images and 0.0625 in 3D images. The number of
iterations can be usually around 10, more iterations will result in further smoothing and will increase
the computing time linearly. The radius of the stencil can be typically 1. The edge-preserving
characteristic is not perfect on this filter. Some degradation will occur on the edges and will increase
as the number of iterations is increased.
If the output of this filter has been connected to other filters down the pipeline, updating any of the
downstream filters will trigger the execution of this one. For example, a writer filter can be used
after the curvature flow filter.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.28 illustrates the effect of this filter on a MRI proton density image of the brain. In this
122
Chapter 2. Filtering
example the filter was run with a time step of 0.125, 10 iterations and a radius of 1. The figure shows
how homogeneous regions are smoothed and edges are preserved. Notice also, that the result in the
figure has sharper edges than the same example using simple curvature flow in Figure 2.26.
The following classes provide similar functionality:
• itk::CurvatureFlowImageFilter
Bilateral Filter
The source code for this section can be found in the file
BilateralImageFilter.cxx.
The itk::BilateralImageFilter performs smoothing by using both domain and range neighborhoods. Pixels that are close to a pixel in the image domain and similar to a pixel in the image
range are used to calculate the filtered value. Two Gaussian kernels (one in the image domain and
one in the image range) are used to smooth the image. The result is an image that is smoothed in
homogeneous regions yet has edges preserved. The result is similar to anisotropic diffusion but the
implementation is non-iterative. Another benefit to bilateral filtering is that any distance metric can
be used for kernel smoothing the image range. Bilateral filtering is capable of reducing the noise in
an image by an order of magnitude while maintaining edges. The bilateral operator used here was
described by Tomasi and Manduchi (Bilateral Filtering for Gray and Color Images. IEEE ICCV.
1998.)
The filtering operation can be described by the following equation
h(x) = k(x)−1
Z
ω
f (w)c(x, w)s( f (x), f (w))dw
(2.16)
where x holds the coordinates of a ND point, f (x) is the input image and h(x) is the output image.
The convolution kernels c() and s() are associated with the spatial and intensity domain respectively. The ND integral is computed over ω which is a neighborhood of the pixel located at x. The
normalization factor k(x) is computed as
k(x) =
Z
ω
c(x, w)s( f (x), f (w))dw
(2.17)
The default implementation of this filter uses Gaussian kernels for both c() and s(). The c kernel
can be described as
c(x, w) = e
(
||x−w||2
)
σ2c
(2.18)
2.7. Smoothing Filters
123
where σc is provided by the user and defines how close pixel neighbors should be in order to be
considered for the computation of the output value. The s kernel is given by
s( f (x), f (w)) = e
(
( f (x)− f (w) 2
σ2s
)
(2.19)
where σs is provided by the user and defines how close the neighbor’s intensity be in order to be
considered for the computation of the output value.
The first step required to use this filter is to include its header file.
#include "itkBilateralImageFilter.h"
The image types are instantiated using pixel type and dimension.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, 2 >;
using OutputImageType = itk::Image< OutputPixelType, 2 >;
The bilateral filter type is now instantiated using both the input image and the output image types
and the filter object is created.
using FilterType = itk::BilateralImageFilter<
InputImageType, OutputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
a source.
filter->SetInput( reader->GetOutput() );
The Bilateral filter requires two parameters. First, we must specify the standard deviation σ to be
used for the Gaussian kernel on image intensities. Second, the set of σs to be used along each
dimension in the space domain. This second parameter is supplied as an array of float or double
values. The array dimension matches the image dimension. This mechanism makes it possible to
enforce more coherence along some directions. For example, more smoothing can be done along
the X direction than along the Y direction.
In the following code example, the σ values are taken from the command line. Note the use of
ImageType::ImageDimension to get access to the image dimension at compile time.
124
Chapter 2. Filtering
Figure 2.29: Effect of the BilateralImageFilter on a slice from a MRI proton density image of the brain.
const unsigned int Dimension = InputImageType::ImageDimension;
double domainSigmas[ Dimension ];
for(double & domainSigma : domainSigmas)
{
domainSigma = std::stod( argv[3] );
}
const double rangeSigma = std::stod( argv[4] );
The filter parameters are set with the methods SetRangeSigma() and SetDomainSigma().
filter->SetDomainSigma( domainSigmas );
filter->SetRangeSigma( rangeSigma
);
The output of the filter is connected here to a intensity rescaler filter and then to a writer. Invoking
Update() on the writer triggers the execution of both filters.
rescaler->SetInput( filter->GetOutput() );
writer->SetInput( rescaler->GetOutput() );
writer->Update();
Figure 2.29 illustrates the effect of this filter on a MRI proton density image of the brain. In this
example the filter was run with a range σ of 5.0 and a domain σ of 6.0. The figure shows how
homogeneous regions are smoothed and edges are preserved.
2.7. Smoothing Filters
125
The following classes provide similar functionality:
• itk::GradientAnisotropicDiffusionImageFilter
• itk::CurvatureAnisotropicDiffusionImageFilter
• itk::CurvatureFlowImageFilter
2.7.4 Edge Preserving Smoothing in Vector Images
Anisotropic diffusion can also be applied to images whose pixels are vectors. In this case the diffusion is computed independently for each vector component. The following classes implement
versions of anisotropic diffusion on vector images.
Vector Gradient Anisotropic Diffusion
The source code for this section can be found in the file
VectorGradientAnisotropicDiffusionImageFilter.cxx.
The itk::VectorGradientAnisotropicDiffusionImageFilter implements an N-dimensional
version of the classic Perona-Malik anisotropic diffusion equation for vector-valued images. Typically in vector-valued diffusion, vector components are diffused independently of one another using
a conductance term that is linked across the components. The diffusion equation was illustrated in
2.7.3.
This filter is designed to process images of itk::Vector type. The code relies on various type
alias and overloaded operators defined in itk::Vector. It is perfectly reasonable, however, to
apply this filter to images of other, user-defined types as long as the appropriate type alias and
operator overloads are in place. As a general rule, follow the example of itk::Vector in defining
your data types.
The first step required to use this filter is to include its header file.
#include "itkVectorGradientAnisotropicDiffusionImageFilter.h"
Types should be selected based on required pixel type for the input and output images. The image
types are defined using the pixel type and the dimension.
using
using
using
using
InputPixelType = float;
VectorPixelType = itk::CovariantVector< float, 2 >;
InputImageType = itk::Image< InputPixelType, 2 >;
VectorImageType = itk::Image< VectorPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types. The filter
object is created by the New() method.
126
Chapter 2. Filtering
using FilterType = itk::VectorGradientAnisotropicDiffusionImageFilter<
VectorImageType, VectorImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source and its data is passed through a gradient filter in order to generate an image of vectors.
gradient->SetInput( reader->GetOutput() );
filter->SetInput( gradient->GetOutput() );
This filter requires two parameters: the number of iterations to be performed and the time step
used in the computation of the level set evolution. These parameters are set using the methods
SetNumberOfIterations() and SetTimeStep() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter(1.0);
filter->Update();
Typical values for the time step are 0.125 in 2D images and 0.0625 in 3D images. The number of
iterations can be usually around 5, however more iterations will result in further smoothing and will
linearly increase the computing time.
Figure 2.30 illustrates the effect of this filter on a MRI proton density image of the brain. The images
show the X component of the gradient before (left) and after (right) the application of the filter. In
this example the filter was run with a time step of 0.25, and 5 iterations.
Vector Curvature Anisotropic Diffusion
The source code for this section can be found in the file
VectorCurvatureAnisotropicDiffusionImageFilter.cxx.
The itk::VectorCurvatureAnisotropicDiffusionImageFilter performs anisotropic diffusion on a vector image using a modified curvature diffusion equation (MCDE). The MCDE is the
same described in 2.7.3.
Typically in vector-valued diffusion, vector components are diffused independently of one another
using a conductance term that is linked across the components.
This filter is designed to process images of itk::Vector type. The code relies on various type
alias and overloaded operators defined in itk::Vector. It is perfectly reasonable, however, to
apply this filter to images of other, user-defined types as long as the appropriate type alias and
operator overloads are in place. As a general rule, follow the example of the itk::Vector class in
defining your data types.
2.7. Smoothing Filters
127
Figure 2.30: Effect of the VectorGradientAnisotropicDiffusionImageFilter on the X component of the gradient
from a MRI proton density brain image.
The first step required to use this filter is to include its header file.
#include "itkVectorCurvatureAnisotropicDiffusionImageFilter.h"
Types should be selected based on required pixel type for the input and output images. The image
types are defined using the pixel type and the dimension.
using
using
using
using
InputPixelType = float;
VectorPixelType = itk::CovariantVector< float, 2 >;
InputImageType = itk::Image< InputPixelType, 2 >;
VectorImageType = itk::Image< VectorPixelType, 2 >;
The filter type is now instantiated using both the input image and the output image types. The filter
object is created by the New() method.
using FilterType = itk::VectorCurvatureAnisotropicDiffusionImageFilter<
VectorImageType, VectorImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source and its data is passed through a gradient filter in order to generate an image of vectors.
128
Chapter 2. Filtering
Figure 2.31: Effect of the VectorCurvatureAnisotropicDiffusionImageFilter on the X component of the gradient
from a MRIproton density brain image.
gradient->SetInput( reader->GetOutput() );
filter->SetInput( gradient->GetOutput() );
This filter requires two parameters: the number of iterations to be performed and the time step
used in the computation of the level set evolution. These parameters are set using the methods
SetNumberOfIterations() and SetTimeStep() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter(1.0);
filter->Update();
Typical values for the time step are 0.125 in 2D images and 0.0625 in 3D images. The number of
iterations can be usually around 5, however more iterations will result in further smoothing and will
increase the computing time linearly.
Figure 2.31 illustrates the effect of this filter on a MRI proton density image of the brain. The images
show the X component of the gradient before (left) and after (right) the application of the filter. In
this example the filter was run with a time step of 0.25, and 5 iterations.
2.7. Smoothing Filters
129
2.7.5 Edge Preserving Smoothing in Color Images
Gradient Anisotropic Diffusion
The source code for this section can be found in the file
RGBGradientAnisotropicDiffusionImageFilter.cxx.
The vector anisotropic diffusion approach applies to color images equally well. As in the vector
case, each RGB component is diffused independently. The following example illustrates the use of
the Vector curvature anisotropic diffusion filter on an image with itk::RGBPixel type.
The first step required to use this filter is to include its header file.
#include "itkVectorGradientAnisotropicDiffusionImageFilter.h"
Also the headers for Image and RGBPixel type are required.
#include "itkRGBPixel.h"
#include "itkImage.h"
It is desirable to perform the computation on the RGB image using float representation. However
for input and output purposes unsigned char RGB components are commonly used. It is necessary to cast the type of color components along the pipeline before writing them to a file. The
itk::VectorCastImageFilter is used to achieve this goal.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkVectorCastImageFilter.h"
The image type is defined using the pixel type and the dimension.
using InputPixelType = itk::RGBPixel< float >;
using InputImageType = itk::Image< InputPixelType, 2 >;
The filter type is now instantiated and a filter object is created by the New() method.
using FilterType = itk::VectorGradientAnisotropicDiffusionImageFilter<
InputImageType, InputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
source.
130
Chapter 2. Filtering
using ReaderType = itk::ImageFileReader< InputImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
filter->SetInput( reader->GetOutput() );
This filter requires two parameters: the number of iterations to be performed and the time step
used in the computation of the level set evolution. These parameters are set using the methods
SetNumberOfIterations() and SetTimeStep() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter(1.0);
filter->Update();
The filter output is now cast to unsigned char RGB components by using the
itk::VectorCastImageFilter.
using WritePixelType = itk::RGBPixel< unsigned char >;
using WriteImageType = itk::Image< WritePixelType, 2 >;
using CasterType = itk::VectorCastImageFilter<
InputImageType, WriteImageType >;
CasterType::Pointer caster = CasterType::New();
Finally, the writer type can be instantiated. One writer is created and connected to the output of the
cast filter.
using WriterType = itk::ImageFileWriter< WriteImageType >;
WriterType::Pointer writer = WriterType::New();
caster->SetInput( filter->GetOutput() );
writer->SetInput( caster->GetOutput() );
writer->SetFileName( argv[2] );
writer->Update();
Figure 2.32 illustrates the effect of this filter on a RGB image from a cryogenic section of the Visible
Woman data set. In this example the filter was run with a time step of 0.125, and 20 iterations. The
input image has 570 × 670 pixels and the processing took 4 minutes on a Pentium 4 2GHz.
Curvature Anisotropic Diffusion
The source code for this section can be found in the file
RGBCurvatureAnisotropicDiffusionImageFilter.cxx.
The vector anisotropic diffusion approach can be applied equally well to color images. As in the
vector case, each RGB component is diffused independently. The following example illustrates
2.7. Smoothing Filters
131
Figure 2.32: Effect of the VectorGradientAnisotropicDiffusionImageFilter on a RGB image from a cryogenic
section of the Visible Woman data set.
the use of the itk::VectorCurvatureAnisotropicDiffusionImageFilter on an image with
itk::RGBPixel type.
The first step required to use this filter is to include its header file.
#include "itkVectorCurvatureAnisotropicDiffusionImageFilter.h"
Also the headers for Image and RGBPixel type are required.
#include "itkRGBPixel.h"
#include "itkImage.h"
It is desirable to perform the computation on the RGB image using float representation. However for input and output purposes unsigned char RGB components are commonly used. It is
necessary to cast the type of color components in the pipeline before writing them to a file. The
itk::VectorCastImageFilter is used to achieve this goal.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkVectorCastImageFilter.h"
The image type is defined using the pixel type and the dimension.
132
Chapter 2. Filtering
using InputPixelType = itk::RGBPixel< float >;
using InputImageType = itk::Image< InputPixelType, 2 >;
The filter type is now instantiated and a filter object is created by the New() method.
using FilterType = itk::VectorCurvatureAnisotropicDiffusionImageFilter<
InputImageType, InputImageType >;
FilterType::Pointer filter = FilterType::New();
The input image can be obtained from the output of another filter. Here, an image reader is used as
a source.
using ReaderType = itk::ImageFileReader< InputImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
filter->SetInput( reader->GetOutput() );
This filter requires two parameters: the number of iterations to be performed and the time step
used in the computation of the level set evolution. These parameters are set using the methods
SetNumberOfIterations() and SetTimeStep() respectively. The filter can be executed by invoking Update().
filter->SetNumberOfIterations( numberOfIterations );
filter->SetTimeStep( timeStep );
filter->SetConductanceParameter(1.0);
filter->Update();
The filter output is now cast to unsigned char RGB components by using the
itk::VectorCastImageFilter.
using WritePixelType = itk::RGBPixel< unsigned char >;
using WriteImageType = itk::Image< WritePixelType, 2 >;
using CasterType = itk::VectorCastImageFilter<
InputImageType, WriteImageType >;
CasterType::Pointer caster = CasterType::New();
Finally, the writer type can be instantiated. One writer is created and connected to the output of the
cast filter.
using WriterType = itk::ImageFileWriter< WriteImageType >;
WriterType::Pointer writer = WriterType::New();
caster->SetInput( filter->GetOutput() );
writer->SetInput( caster->GetOutput() );
writer->SetFileName( argv[2] );
writer->Update();
2.8. Distance Map
133
Figure 2.33: Effect of the VectorCurvatureAnisotropicDiffusionImageFilter on a RGB image from a cryogenic
section of the Visible Woman data set.
Figure 2.33 illustrates the effect of this filter on a RGB image from a cryogenic section of the Visible
Woman data set. In this example the filter was run with a time step of 0.125, and 20 iterations. The
input image has 570 × 670 pixels and the processing took 4 minutes on a Pentium 4 at 2GHz.
Figure 2.34 compares the effect of the gradient and curvature anisotropic diffusion filters on a small
region of the same cryogenic slice used in Figure 2.33. The region used in this figure is only 127 ×
162 pixels and took 14 seconds to compute on the same platform.
2.8 Distance Map
The source code for this section can be found in the file
DanielssonDistanceMapImageFilter.cxx.
This example illustrates the use of the itk::DanielssonDistanceMapImageFilter. This filter
generates a distance map from the input image using the algorithm developed by Danielsson [13].
As secondary outputs, a Voronoi partition of the input elements is produced, as well as a vector
image with the components of the distance vector to the closest point. The input to the map is
assumed to be a set of points on the input image. The label of each group of pixels is assigned by
the itk::ConnectedComponentImageFilter.
The first step required to use this filter is to include its header file.
134
Chapter 2. Filtering
Figure 2.34: Comparison between the gradient (center) and curvature (right) Anisotropic Diffusion filters. Original image at left.
#include "itkDanielssonDistanceMapImageFilter.h"
Then we must decide what pixel types to use for the input and output images. Since the output will
contain distances measured in pixels, the pixel type should be able to represent at least the width
of the image, or said in N-dimensional terms, the maximum extension along all the dimensions.
The input, output (distance map), and voronoi partition image types are now defined using their
respective pixel type and dimension.
using
using
using
using
using
using
InputPixelType = unsigned char;
OutputPixelType = unsigned short;
VoronoiPixelType = unsigned char;
InputImageType = itk::Image< InputPixelType, 2 >;
OutputImageType = itk::Image< OutputPixelType, 2 >;
VoronoiImageType = itk::Image< VoronoiPixelType, 2 >;
The filter type can be instantiated using the input and output image types defined above. A filter
object is created with the New() method.
using FilterType = itk::DanielssonDistanceMapImageFilter<
InputImageType, OutputImageType, VoronoiImageType >;
FilterType::Pointer filter = FilterType::New();
The input to the filter is taken from a reader and its output is passed to a
itk::RescaleIntensityImageFilter and then to a writer. The scaler and writer are both
templated over the image type, so we instantiate a separate pipeline for the voronoi partition map
starting at the scaler.
2.8. Distance Map
135
Figure 2.35: DanielssonDistanceMapImageFilter output. Set of pixels, distance map and Voronoi partition.
labeler->SetInput(reader->GetOutput() );
filter->SetInput( labeler->GetOutput() );
scaler->SetInput( filter->GetOutput() );
writer->SetInput( scaler->GetOutput() );
The Voronoi map is obtained with the GetVoronoiMap() method. In the lines below we connect
this output to the intensity rescaler.
voronoiScaler->SetInput( filter->GetVoronoiMap() );
voronoiWriter->SetInput( voronoiScaler->GetOutput() );
Figure 2.35 illustrates the effect of this filter on a binary image with a set of points. The input image
is shown at the left, and the distance map at the center and the Voronoi partition at the right. This
filter computes distance maps in N-dimensions and is therefore capable of producing N-dimensional
Voronoi partitions.
The distance filter also produces an image of itk::Offset pixels representing the vectorial distance
to the closest object in the scene. The type of this output image is defined by the VectorImageType
trait of the filter type.
using OffsetImageType = FilterType::VectorImageType;
We can use this type for instantiating an itk::ImageFileWriter type and creating an object of
this class in the following lines.
using WriterOffsetType = itk::ImageFileWriter< OffsetImageType >;
WriterOffsetType::Pointer offsetWriter = WriterOffsetType::New();
The output of the distance filter can be connected as input to the writer.
136
Chapter 2. Filtering
offsetWriter->SetInput( filter->GetVectorDistanceMap() );
Execution of the writer is triggered by the invocation of the Update() method. Since this method
can potentially throw exceptions it must be placed in a try/catch block.
try
{
offsetWriter->Update();
}
catch( itk::ExceptionObject & exp )
{
std::cerr << "Exception caught !" << std::endl;
exp
<< std::endl;
std::cerr <<
}
Note that only the itk::MetaImageIO class supports reading and writing images of pixel type
itk::Offset.
The source code for this section can be found in the file
SignedDanielssonDistanceMapImageFilter.cxx.
This example illustrates the use of the itk::SignedDanielssonDistanceMapImageFilter. This
filter generates a distance map by running Danielsson distance map twice, once on the input image
and once on the flipped image.
The first step required to use this filter is to include its header file.
#include "itkSignedDanielssonDistanceMapImageFilter.h"
Then we must decide what pixel types to use for the input and output images. Since the output will
contain distances measured in pixels, the pixel type should be able to represent at least the width of
the image, or said in N-dimensional terms, the maximum extension along all the dimensions. The
input and output image types are now defined using their respective pixel type and dimension.
using InputPixelType = unsigned char;
using OutputPixelType = float;
using VoronoiPixelType = unsigned short;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
using VoronoiImageType = itk::Image< VoronoiPixelType, Dimension >;
The only change with respect to the previous example is to replace the DanielssonDistanceMapImageFilter with the SignedDanielssonDistanceMapImageFilter.
2.9. Geometric Transformations
137
Figure 2.36: SignedDanielssonDistanceMapImageFilter applied on a binary circle image. The intensity has
been rescaled for purposes of display.
using FilterType = itk::SignedDanielssonDistanceMapImageFilter<
InputImageType,
OutputImageType,
VoronoiImageType >;
FilterType::Pointer filter = FilterType::New();
The distances inside the circle are defined to be negative, while the distances outside the circle are
positive. To change the convention, use the InsideIsPositive(bool) function.
Figure 2.36 illustrates the effect of this filter. The input image and the distance map are shown.
2.9 Geometric Transformations
2.9.1 Filters You Should be Afraid to Use
2.9.2 Change Information Image Filter
This one is the scariest and most dangerous filter in the entire toolkit. You should not use this filter
unless you are entirely certain that you know what you are doing. In fact if you decide to use this
filter, you should write your code, then go for a long walk, get more coffee and ask yourself if you
really needed to use this filter. If the answer is yes, then you should discuss this issue with someone
you trust and get his/her opinion in writing. In general, if you need to use this filter, it means that
you have a poor image provider that is putting your career at risk along with the life of any potential
patient whose images you may end up processing.
138
Chapter 2. Filtering
2.9.3 Flip Image Filter
The source code for this section can be found in the file
FlipImageFilter.cxx.
The itk::FlipImageFilter is used for flipping the image content in any of the coordinate axes.
This filter must be used with EXTREME caution. You probably don’t want to appear in the newspapers as responsible for a surgery mistake in which a doctor extirpates the left kidney when he
should have extracted the right one3 . If that prospect doesn’t scare you, maybe it is time for you to
reconsider your career in medical image processing. Flipping effects which seem innocuous at first
view may still have dangerous consequences. For example, flipping the cranio-caudal axis of a CT
scan forces an observer to flip the left-right axis in order to make sense of the image.
The header file corresponding to this filter should be included first.
#include "itkFlipImageFilter.h"
Then the pixel types for input and output image must be defined and, with them, the image types
can be instantiated.
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, 2 >;
Using the image types it is now possible to instantiate the filter type and create the filter object.
using FilterType = itk::FlipImageFilter< ImageType >;
FilterType::Pointer filter = FilterType::New();
The axes to flip are specified in the form of an Array. In this case we take them from the command
line arguments.
using FlipAxesArrayType = FilterType::FlipAxesArrayType;
FlipAxesArrayType flipArray;
flipArray[0] = std::stoi( argv[3] );
flipArray[1] = std::stoi( argv[4] );
filter->SetFlipAxes( flipArray );
The input to the filter can be taken from any other filter, for example a reader. The output can
be passed down the pipeline to other filters, for example, a writer. Invoking Update() on any
downstream filter will trigger the execution of the FlipImage filter.
3 Wrong
side surgery accounts for 2% of the reported medical errors in the United States. Trivial... but equally dangerous.
2.9. Geometric Transformations
139
Figure 2.37: Effect of the FlipImageFilter on a slice from a MRI proton density brain image.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
Figure 2.37 illustrates the effect of this filter on a slice of an MRI brain image using a flip array [0, 1]
which means that the Y axis was flipped while the X axis was conserved.
2.9.4 Resample Image Filter
Introduction
The source code for this section can be found in the file
ResampleImageFilter.cxx.
Resampling an image is a very important task in image analysis. It is especially important in the
frame of image registration. The itk::ResampleImageFilter implements image resampling
through the use of itk::Transforms. The inputs expected by this filter are an image, a transform and an interpolator. The space coordinates of the image are mapped through the transform in
order to generate a new image. The extent and spacing of the resulting image are selected by the
user. Resampling is performed in space coordinates, not pixel/grid coordinates. It is quite important
to ensure that image spacing is properly set on the images involved. The interpolator is required
since the mapping from one space to the other will often require evaluation of the intensity of the
140
Chapter 2. Filtering
image at non-grid positions.
The header file corresponding to this filter should be included first.
#include "itkResampleImageFilter.h"
The header files corresponding to the transform and interpolator must also be included.
#include "itkAffineTransform.h"
#include "itkNearestNeighborInterpolateImageFunction.h"
The dimension and pixel types for input and output image must be defined and with them the image
types can be instantiated.
constexpr unsigned int Dimension = 2;
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
Using the image and transform types it is now possible to instantiate the filter type and create the
filter object.
using FilterType = itk::ResampleImageFilter;
FilterType::Pointer filter = FilterType::New();
The transform type is typically defined using the image dimension and the type used for representing
space coordinates.
using TransformType = itk::AffineTransform< double, Dimension >;
An instance of the transform object is instantiated and passed to the resample filter. By default, the
parameters of the transform are set to represent the identity transform.
TransformType::Pointer transform = TransformType::New();
filter->SetTransform( transform );
The interpolator type is defined using the full image type and the type used for representing space
coordinates.
using InterpolatorType = itk::NearestNeighborInterpolateImageFunction<
InputImageType, double >;
2.9. Geometric Transformations
141
An instance of the interpolator object is instantiated and passed to the resample filter.
InterpolatorType::Pointer interpolator = InterpolatorType::New();
filter->SetInterpolator( interpolator );
Given that some pixels of the output image may end up being mapped outside the extent of the
input image it is necessary to decide what values to assign to them. This is done by invoking the
SetDefaultPixelValue() method.
filter->SetDefaultPixelValue( 0 );
The sampling grid of the output space is specified with the spacing along each dimension and the
origin.
// pixel spacing in millimeters along X and Y
const double spacing[ Dimension ] = { 1.0, 1.0 };
filter->SetOutputSpacing( spacing );
// Physical space coordinate of origin for X and Y
const double origin[ Dimension ] = { 0.0, 0.0 };
filter->SetOutputOrigin( origin );
InputImageType::DirectionType direction;
direction.SetIdentity();
filter->SetOutputDirection( direction );
The extent of the sampling grid on the output image is defined by a SizeType and is set using the
SetSize() method.
InputImageType::SizeType
size[0] = 300;
size[1] = 300;
size;
// number of pixels along X
// number of pixels along Y
filter->SetSize( size );
The input to the filter can be taken from any other filter, for example a reader. The output can be
passed down the pipeline to other filters, for example a writer. An update call on any downstream
filter will trigger the execution of the resampling filter.
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
writer->Update();
142
Chapter 2. Filtering
Figure 2.38: Effect of the resample filter.
300 x 300 pixels
300
300
250
250
181 x 217 pixels
200
200
Identity
Transform
150
150
100
100
50
50
0
0
0
50
100
Input Image
150
200
0
50
100
150
200
Resampled Image
250
Figure 2.39: Analysis of the resample image done in a common coordinate system.
300
2.9. Geometric Transformations
143
Figure 2.40: ResampleImageFilter with a translation by (−30, −50).
Figure 2.38 illustrates the effect of this filter on a slice of MRI brain image using an affine transform
containing an identity transform. Note that any analysis of the behavior of this filter must be done
on the space coordinate system in millimeters, not with respect to the sampling grid in pixels. The
figure shows the resulting image in the lower left quarter of the extent. This may seem odd if
analyzed in terms of the image grid but is quite clear when seen with respect to space coordinates.
Figure 2.38 is particularly misleading because the images are rescaled to fit nicely on the text of this
book. Figure 2.39 clarifies the situation. It shows the two same images placed on an equally-scaled
coordinate system. It becomes clear here that an identity transform is being used to map the image
data, and that simply, we have requested to resample additional empty space around the image. The
input image is 181 × 217 pixels in size and we have requested an output of 300 × 300 pixels. In this
case, the input and output images both have spacing of 1mm × 1mm and origin of (0.0, 0.0).
Let’s now set values on the transform. Note that the supplied transform represents the mapping of
points from the output space to the input space. The following code sets up a translation.
TransformType::OutputVectorType translation;
translation[0] = -30; // X translation in millimeters
translation[1] = -50; // Y translation in millimeters
transform->Translate( translation );
The output image resulting from the translation can be seen in Figure 2.40. Again, it is better to
interpret the result in a common coordinate system as illustrated in Figure 2.41.
Probably the most important thing to keep in mind when resampling images is that the transform is
144
Chapter 2. Filtering
300 x 300 pixels
300
300
250
T={−30,−50}
250
181 x 217 pixels
200
200
Translation
Transform
(105,188)
150
150
(75,138)
100
100
50
50
0
0
0
50
100
Input Image
150
200
0
50
100
150
200
Resampled Image
250
300
Figure 2.41: ResampleImageFilter. Analysis of a translation by (−30, −50).
used to map points from the output image space into the input image space. In this case, Figure
2.41 shows that the translation is applied to every point of the output image and the resulting position
is used to read the intensity from the input image. In this way, the gray level of the point P in the
output image is taken from the point T (P) in the input image. Where T is the transformation.
In the specific case of the Figure 2.41, the value of point (105, 188) in the output image is taken
from the point (75, 138) of the input image because the transformation applied was a translation of
(−30, −50).
It is sometimes useful to intentionally set the default output value to a distinct gray value in order
to highlight the mapping of the image borders. For example, the following code sets the default
external value of 100. The result is shown in the right side of Figure 2.42.
filter->SetDefaultPixelValue( 100 );
With this change we can better appreciate the effect of the previous translation transform on the
image resampling. Figure 2.42 illustrates how the point (30, 50) of the output image gets its gray
value from the point (0, 0) of the input image.
Importance of Spacing and Origin
The source code for this section can be found in the file
ResampleImageFilter2.cxx.
2.9. Geometric Transformations
145
300 x 300 pixels
300
300
T={−30,−50}
250
250
181 x 217 pixels
200
200
Translation
Transform
150
150
100
100
50
50
(0,0)
0
0
50
100
Input Image
(30,50)
0
150
0
200
50
100
150
200
Resampled Image
250
300
Figure 2.42: ResampleImageFilter highlighting image borders with SetDefaultPixelValue().
During the computation of the resampled image all the pixels in the output region are visited. This
visit is performed using ImageIterators which walk in the integer grid-space of the image. For
each pixel, we need to convert grid position to space coordinates using the image spacing and origin.
For example, the pixel of index I = (20, 50) in an image of origin O = (19.0, 29.0) and pixel spacing
S = (1.3, 1.5) corresponds to the spatial position
P[i] = I[i] × S[i] + O[i]
(2.20)
which in this case leads to P = (20 × 1.3 + 19.0, 50 × 1.5 + 29.0) and finally P = (45.0, 104.0)
The space coordinates of P are mapped using the transform T supplied to the
itk::ResampleImageFilter in order to map the point P to the input image space point
Q = T (P).
The whole process is illustrated in Figure 2.43. In order to correctly interpret the process of the
ResampleImageFilter you should be aware of the origin and spacing settings of both the input and
output images.
In order to facilitate the interpretation of the transform we set the default pixel value to a value
distinct from the image background.
filter->SetDefaultPixelValue( 50 );
Let’s set up a uniform spacing for the output image.
146
Chapter 2. Filtering
// pixel spacing in millimeters along X & Y
const double spacing[ Dimension ] = { 1.0, 1.0 };
filter->SetOutputSpacing( spacing );
We will preserve the orientation of the input image by using the following call.
filter->SetOutputDirection( reader->GetOutput()->GetDirection() );
Additionally, we will specify a non-zero origin. Note that the values provided here will be those of
the space coordinates for the pixel of index (0, 0).
// space coordinate of origin
const double origin[ Dimension ] = { 30.0, 40.0 };
filter->SetOutputOrigin( origin );
We set the transform to identity in order to better appreciate the effect of the origin selection.
transform->SetIdentity();
filter->SetTransform( transform );
The output resulting from these filter settings is analyzed in Figure 2.43.
In the figure, the output image point with index I = (0, 0) has space coordinates P = (30, 40). The
identity transform maps this point to Q = (30, 40) in the input image space. Because the input image
in this case happens to have spacing (1.0, 1.0) and origin (0.0, 0.0), the physical point Q = (30, 40)
maps to the pixel with index I = (30, 40).
The code for a different selection of origin and image size is illustrated below. The resulting output
is presented in Figure 2.44.
size[0] = 150; // number of pixels along X
size[1] = 200; // number of pixels along Y
filter->SetSize( size );
// space coordinate of origin
const double origin[ Dimension ] = { 60.0, 30.0 };
filter->SetOutputOrigin( origin );
The output image point with index I = (0, 0) now has space coordinates P = (60, 30). The identity
transform maps this point to Q = (60, 30) in the input image space. Because the input image in this
case happens to have spacing (1.0, 1.0) and origin (0.0, 0.0), the physical point Q = (60, 30) maps
to the pixel with index I = (60, 30).
2.9. Geometric Transformations
147
Size=300x300
Spacing=(1.0,1.0)
300
300
DefaultPixelValue
250
250
Size=181x217
Spacing=(1.0,1.0)
200
200
150
150
Identity
Transform
100
100
50
50
Origin=(0,0)
Index=(0,0)
0
0
50
100
150
Input Image
Origin=(30,40)
Index=(0,0)
0
0
200
50
100
150
200
Resampled Image
250
300
Figure 2.43: ResampleImageFilter selecting the origin of the output image.
300
300
Size=150x200
Size=181x217
250
Spacing=(1.0,1.0)
250
DefaultPixelValue
Spacing=(1.0,1.0)
200
200
150
150
Identity
Transform
100
50
100
Origin=(60,30)
Index=(0,0)
50
Origin=(0,0)
Index=(0,0)
0
0
0
50
100
150
Input Image
200
0
50
100
150
Resampled Image
Figure 2.44: ResampleImageFilter origin in the output image.
200
250
300
148
Chapter 2. Filtering
Size=181x217
Spacing=(1.0,1.0)
300
300
250
Size=150x200
Spacing=(1.0,1.0)
250
200
200
I=(66,80)
Identity
Transform
150
I=(56,120)
150
DefaultPixelValue
100
100
50
Origin=(50,70)
Index=(0,0)
0
50
Origin=(60,30)
Index=(0,0)
0
0
50
100
150
200
0
Input Image
50
100
150
200
250
300
Resampled Image
Figure 2.45: Effect of selecting the origin of the input image with ResampleImageFilter.
Let’s now analyze the effect of a non-zero origin in the input image. Keeping the output image
settings of the previous example, we modify only the origin values on the file header of the input
image. The new origin assigned to the input image is O = (50, 70). An identity transform is still
used as input for the ResampleImageFilter. The result of executing the filter with these parameters
is presented in Figure 2.45.
The pixel with index I = (56, 120) on the output image has coordinates P = (116, 150) in physical
space. The identity transform maps P to the point Q = (116, 150) on the input image space. The
coordinates of Q are associated with the pixel of index I = (66, 80) on the input image.
Now consider the effect of the output spacing on the process of image resampling. In order to
simplify the analysis, let’s set the origin back to zero in both the input and output images.
// space coordinate of origin
const double origin[ Dimension ] = { 0.0, 0.0 };
filter->SetOutputOrigin( origin );
We then specify a non-unit spacing for the output image.
// pixel spacing in millimeters
const double spacing[ Dimension ] = { 2.0, 3.0 };
filter->SetOutputSpacing( spacing );
Additionally, we reduce the output image extent, since the new pixels are now covering a larger area
of 2.0mm × 3.0mm.
2.9. Geometric Transformations
149
Figure 2.46: Resampling with different spacing seen by a naive viewer (center) and a correct viewer (right),
input image (left).
size[0] = 80; // number of pixels along X
size[1] = 50; // number of pixels along Y
filter->SetSize( size );
With these new parameters the physical extent of the output image is 160 millimeters by 150 millimeters.
Before attempting to analyze the effect of the resampling image filter it is important to make sure
that the image viewer used to display the input and output images takes the spacing into account
and appropriately scales the images on the screen. Please note that images in formats like PNG are
not capable of representing origin and spacing. The toolkit assumes trivial default values for them.
Figure 2.46 (center) illustrates the effect of using a naive viewer that does not take pixel spacing into
account. A correct display is presented at the right in the same figure4.
The filter output is analyzed in a common coordinate system with the input from Figure 2.47. In this
figure, pixel I = (33, 27) of the output image is located at coordinates P = (66.0, 81.0) of the physical
space. The identity transform maps this point to Q = (66.0, 81.0) in the input image physical space.
The point Q is then associated to the pixel of index I = (66, 81) on the input image, because this
image has zero origin and unit spacing.
The input image spacing is also an important factor in the process of resampling an image. The
following example illustrates the effect of non-unit pixel spacing on the input image. An input
image similar to the those used in Figures 2.43 to 2.47 has been resampled to have pixel spacing
of 2mm × 3mm. The input image is presented in Figure 2.48 as viewed with a naive image viewer
(left) and with a correct image viewer (right).
4A
viewer is provided with ITK under the name of MetaImageViewer. This viewer takes into account pixel spacing.
150
Chapter 2. Filtering
300
300
Size=181x217
Size=80x50
Spacing=(1.0,1.0)
Spacing=(2.0,3.0)
Physical extent=(181.0,217.0)
250
Physical extent=(160.0,150.0)
250
200
200
Identity
Transform
150
150
I=(33,27)
100
P=(66.0,81.0)
100
I=(66,81)
Q=(66.0,81.0)
50
0
50
0
0
50
100
150
Input Image
200
0
50
100
150
200
250
300
Resampled Image
Figure 2.47: Effect of selecting the spacing on the output image.
Figure 2.48: Input image with 2 × 3mm spacing as seen with a naive viewer (left) and a correct viewer (right).
2.9. Geometric Transformations
151
The following code is used to transform this non-unit spacing input image into another non-unit
spacing image located at a non-zero origin. The comparison between input and output in a common
reference system is presented in figure 2.49.
Here we start by selecting the origin of the output image.
// space coordinate of origin
const double origin[ Dimension ] = { 25.0, 35.0 };
filter->SetOutputOrigin( origin );
We then select the number of pixels along each dimension.
size[0] = 40; // number of pixels along X
size[1] = 45; // number of pixels along Y
filter->SetSize( size );
Finally, we set the output pixel spacing.
const double spacing[ Dimension ] = { 4.0, 4.5 };
filter->SetOutputSpacing( spacing );
Figure 2.49 shows the analysis of the filter output under these conditions. First, notice that the origin
of the output image corresponds to the settings O = (25.0, 35.0) millimeters, spacing (4.0, 4.5)
millimeters and size (40, 45) pixels. With these parameters the pixel of index I = (10, 10) in the
output image is associated with the spatial point of coordinates P = (10 × 4.0 + 25.0, 10 × 4.5 +
35.0)) = (65.0, 80.0). This point is mapped by the transform—identity in this particular case—to
the point Q = (65.0, 80.0) in the input image space. The point Q is then associated with the pixel
of index I = ((65.0 − 0.0)/2.0 − (80.0 − 0.0)/3.0) = (32.5, 26.6). Note that the index does not
fall on a grid position. For this reason the value to be assigned to the output pixel is computed by
interpolating values on the input image around the non-integer index I = (32.5, 26.6).
Note also that the discretization of the image is more visible on the output presented on the right
side of Figure 2.49 due to the choice of a low resolution—just 40 × 45 pixels.
A Complete Example
The source code for this section can be found in the file
ResampleImageFilter3.cxx.
Previous examples have described the basic principles behind the itk::ResampleImageFilter.
Now it’s time to have some fun with it.
Figure 2.51 illustrates the general case of the resampling process. The origin and spacing of the
output image has been selected to be different from those of the input image. The circles represent
152
Chapter 2. Filtering
Size=90x72
300
Size=40x45
300
Spacing=(4.0,4.5)
Spacing=(2.0,3.0)
Physical extent=(160.0,202.5)
Physical extent=(180.0,216.0)
250
250
200
200
Identity
Transform
150
150
I=(10,10)
100
P=(65.0,80.0)
100
I=(32.5,26.6)
Q=(65.0,80.0)
50
0
50
Origin=(25.0,35.0)
0
0
50
100
150
200
Input Image
0
50
100
150
200
250
300
Resampled Image
Figure 2.49: Effect of non-unit spacing on the input and output images.
the center of pixels. They are inscribed in a rectangle representing the coverage of this pixel. The
spacing specifies the distance between pixel centers along every dimension.
The transform applied is a rotation of 30 degrees. It is important to note here that the transform
supplied to the itk::ResampleImageFilter is a clockwise rotation. This transform rotates the
coordinate system of the output image 30 degrees clockwise. When the two images are relocated in
a common coordinate system—as in Figure 2.51—the result is that the frame of the output image
appears rotated 30 degrees clockwise. If the output image is seen with its coordinate system vertically aligned—as in Figure 2.50—the image content appears rotated 30 degrees counter-clockwise.
Before continuing to read this section, you may want to meditate a bit on this fact while enjoying a
cup of (Colombian) coffee.
The following code implements the conditions illustrated in Figure 2.51 with two differences: the
output spacing is 40 times smaller and there are 40 times more pixels in both dimensions. Without
these changes, few details will be recognizable in the images. Note that the spacing and origin of
the input image should be prepared in advance by using other means since this filter cannot alter the
actual content of the input image in any way.
In order to facilitate the interpretation of the transform we set the default pixel value to value be
distinct from the image background.
filter->SetDefaultPixelValue( 100 );
The spacing is selected here to be 40 times smaller than the one illustrated in Figure 2.51.
2.9. Geometric Transformations
153
Figure 2.50: Effect of a rotation on the resampling filter. Input image at left, output image at right.
Size=7x6
Spacing=( 20.0, 30.0 )
300
Spacing=( 40.0, 30.0 )
Physical extent=( 200.0, 120.0 )
20.0
250
Size=5x4
300
Physical extent=( 140.0, 180.0 )
250
30.0
200
200
Rotation 30
Transform
150
100
40
.0
150
100
50
30
.0
50
Origin=(60.0,70.0)
0
Origin=(50.0,130.0)
0
0
50
100
150
Input Image
200
0
50
100
150
200
Resampled Image
Figure 2.51: Input and output image placed in a common reference system.
250
300
154
Chapter 2. Filtering
double spacing[ Dimension ];
spacing[0] = 40.0 / 40.0; // pixel spacing in millimeters along X
spacing[1] = 30.0 / 40.0; // pixel spacing in millimeters along Y
filter->SetOutputSpacing( spacing );
We will preserve the orientation of the input image by using the following call.
filter->SetOutputDirection( reader->GetOutput()->GetDirection() );
Let us now set up the origin of the output image. Note that the values provided here will be those of
the space coordinates for the output image pixel of index (0, 0).
double origin[ Dimension ];
origin[0] = 50.0; // X space coordinate of origin
origin[1] = 130.0; // Y space coordinate of origin
filter->SetOutputOrigin( origin );
The output image size is defined to be 40 times the one illustrated on the Figure 2.51.
InputImageType::SizeType size;
size[0] = 5 * 40; // number of pixels along X
size[1] = 4 * 40; // number of pixels along Y
filter->SetSize( size );
Rotations are performed around the origin of physical coordinates—not the image origin nor the
image center. Hence, the process of positioning the output image frame as it is shown in Figure 2.51
requires three steps. First, the image origin must be moved to the origin of the coordinate system.
This is done by applying a translation equal to the negative values of the image origin.
TransformType::OutputVectorType translation1;
translation1[0] =
-origin[0];
translation1[1] =
-origin[1];
transform->Translate( translation1 );
In a second step, a rotation of 30 degrees is performed. In the itk::AffineTransform, angles are
specified in radians. Also, a second boolean argument is used to specify if the current modification
of the transform should be pre-composed or post-composed with the current transform content. In
this case the argument is set to false to indicate that the rotation should be applied after the current
transform content.
const double degreesToRadians = std::atan(1.0) / 45.0;
transform->Rotate2D( -30.0 * degreesToRadians, false );
2.9. Geometric Transformations
155
The third and final step implies translating the image origin back to its previous location. This is be
done by applying a translation equal to the origin values.
TransformType::OutputVectorType translation2;
translation2[0] =
origin[0];
translation2[1] =
origin[1];
transform->Translate( translation2, false );
filter->SetTransform( transform );
Figure 2.50 presents the actual input and output images of this example as shown by a correct viewer
which takes spacing into account. Note the clockwise versus counter-clockwise effect discussed
previously between the representation in Figure 2.51 and Figure 2.50.
As a final exercise, let’s track the mapping of an individual pixel. Keep in mind that the transformation is initiated by walking through the pixels of the output image. This is the only way to
ensure that the image will be generated without holes or redundant values. When you think about
transformation it is always useful to analyze things from the output image towards the input image.
Let’s take the pixel with index I = (1, 2) from the output image. The physical coordinates of
this point in the output image reference system are P = (1 × 40.0 + 50.0, 2 × 30.0 + 130.0) =
(90.0, 190.0) millimeters.
This point P is now mapped through the itk::AffineTransform into the input image space.
The operation subtracts the origin, applies a 30 degrees rotation and adds the origin back. Let’s
follow those steps. Subtracting the origin from P leads to P1 = (40.0, 60.0), the rotation maps P1 to
P2 = (40.0 × cos(30.0) + 60.0 × sin(30.0), 40.0 × sin(30.0) − 60.0 × cos(30.0)) = (64.64, 31.96).
Finally this point is translated back by the amount of the image origin. This moves P2 to P3 =
(114.64, 161.96).
The point P3 is now in the coordinate system of the input image. The pixel of the input image
associated with this physical position is computed using the origin and spacing of the input image.
I = ((114.64 − 60.0)/20.0, (161 − 70.0)/30.0) which results in I = (2.7, 3.0). Note that this is a
non-grid position since the values are non-integers. This means that the gray value to be assigned to
the output image pixel I = (1, 2) must be computed by interpolation of the input image values.
In this particular code the interpolator used is simply a
itk::NearestNeighborInterpolateImageFunction which will assign the value of the closest
pixel. This ends up being the pixel of index I = (3, 3) and can be seen from Figure 2.51.
Rotating an Image
The source code for this section can be found in the file
ResampleImageFilter4.cxx.
The following example illustrates how to rotate an image around its center. In this particular case an
itk::AffineTransform is used to map the input space into the output space.
156
Chapter 2. Filtering
The header of the affine transform is included below.
#include "itkAffineTransform.h"
The transform type is instantiated using the coordinate representation type and the space dimension. Then a transform object is constructed with the New() method and passed to a
itk::SmartPointer.
using TransformType = itk::AffineTransform< double, Dimension >;
TransformType::Pointer transform = TransformType::New();
The parameters of the output image are taken from the input image.
reader->Update();
const InputImageType * inputImage = reader->GetOutput();
const InputImageType::SpacingType & spacing = inputImage->GetSpacing();
const InputImageType::PointType & origin = inputImage->GetOrigin();
InputImageType::SizeType size =
inputImage->GetLargestPossibleRegion().GetSize();
filter->SetOutputOrigin( origin );
filter->SetOutputSpacing( spacing );
filter->SetOutputDirection( inputImage->GetDirection() );
filter->SetSize( size );
Rotations are performed around the origin of physical coordinates—not the image origin nor the
image center. Hence, the process of positioning the output image frame as it is shown in Figure 2.52
requires three steps. First, the image origin must be moved to the origin of the coordinate system.
This is done by applying a translation equal to the negative values of the image origin.
TransformType::OutputVectorType translation1;
const double imageCenterX = origin[0] + spacing[0] * size[0] / 2.0;
const double imageCenterY = origin[1] + spacing[1] * size[1] / 2.0;
translation1[0] =
translation1[1] =
-imageCenterX;
-imageCenterY;
transform->Translate( translation1 );
In a second step, the rotation is specified using the method Rotate2D().
const double degreesToRadians = std::atan(1.0) / 45.0;
const double angle = angleInDegrees * degreesToRadians;
transform->Rotate2D( -angle, false );
2.9. Geometric Transformations
157
Figure 2.52: Effect of the resample filter rotating an image.
The third and final step requires translating the image origin back to its previous location. This is be
done by applying a translation equal to the origin values.
TransformType::OutputVectorType translation2;
translation2[0] =
imageCenterX;
imageCenterY;
translation2[1] =
transform->Translate( translation2, false );
filter->SetTransform( transform );
The output of the resampling filter is connected to a writer and the execution of the pipeline is
triggered by a writer update.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
158
Chapter 2. Filtering
Rotating and Scaling an Image
The source code for this section can be found in the file
ResampleImageFilter5.cxx.
This example illustrates the use of the itk::Similarity2DTransform. A similarity transform
involves rotation, translation and scaling. Since the parameterization of rotations is difficult to get
in a generic ND case, a particular implementation is available for 2D.
The header file of the transform is included below.
#include "itkSimilarity2DTransform.h"
The transform type is instantiated using the coordinate representation type as the single template
parameter.
using TransformType = itk::Similarity2DTransform< double >;
A transform object is constructed by calling New() and passing the result to a itk::SmartPointer.
TransformType::Pointer transform = TransformType::New();
The parameters of the output image are taken from the input image.
The Similarity2DTransform allows the user to select the center of rotation. This center is used for
both rotation and scaling operations.
TransformType::InputPointType rotationCenter;
rotationCenter[0] = origin[0] + spacing[0] * size[0] / 2.0;
rotationCenter[1] = origin[1] + spacing[1] * size[1] / 2.0;
transform->SetCenter( rotationCenter );
The rotation is specified with the method SetAngle().
const double degreesToRadians = std::atan(1.0) / 45.0;
const double angle = angleInDegrees * degreesToRadians;
transform->SetAngle( angle );
The scale change is defined using the method SetScale().
transform->SetScale( scale );
A translation to be applied after the rotation and scaling can be specified with the method
SetTranslation().
2.9. Geometric Transformations
159
Figure 2.53: Effect of the resample filter rotating and scaling an image.
TransformType::OutputVectorType translation;
translation[0] =
translation[1] =
13.0;
17.0;
transform->SetTranslation( translation );
filter->SetTransform( transform );
Note that the order in which rotation, scaling and translation are defined is irrelevant in this transform. This is not the case in the Affine transform which is very generic and allows different combinations for initialization. In the Similarity2DTransform class the rotation and scaling will always be
applied before the translation.
Figure 2.53 shows the effect of this rotation, translation and scaling on a slice of a brain MRI. The
scale applied for producing this figure was 1.2 and the rotation angle was 10◦ .
Resampling using a deformation field
The source code for this section can be found in the file
WarpImageFilter1.cxx.
This example illustrates how to use the WarpImageFilter and a deformation field for resampling an
image. This is typically done as the last step of a deformable registration algorithm.
160
Chapter 2. Filtering
#include "itkWarpImageFilter.h"
The deformation field is represented as an image of vector pixel types. The dimension of the vectors
is the same as the dimension of the input image. Each vector in the deformation field represents the
distance between a geometric point in the input space and a point in the output space such that:
pin = pout + distance
(2.21)
using VectorComponentType = float;
using VectorPixelType = itk::Vector< VectorComponentType, Dimension >;
using DisplacementFieldType = itk::Image< VectorPixelType, Dimension >;
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, Dimension >;
The field is read from a file, through a reader instantiated over the vector pixel types.
using FieldReaderType = itk::ImageFileReader< DisplacementFieldType >;
FieldReaderType::Pointer fieldReader = FieldReaderType::New();
fieldReader->SetFileName( argv[2] );
fieldReader->Update();
DisplacementFieldType::ConstPointer deformationField =
fieldReader->GetOutput();
The itk::WarpImageFilter is templated over the input image type, output image type and the
deformation field type.
using FilterType = itk::WarpImageFilter< ImageType,
ImageType,
DisplacementFieldType >;
FilterType::Pointer filter = FilterType::New();
Typically the mapped position does not correspond to an integer pixel position in the input image.
Interpolation via an image function is used to compute values at non-integer positions. This is done
via the SetInterpolator() method.
using InterpolatorType = itk::LinearInterpolateImageFunction<
ImageType, double >;
2.9. Geometric Transformations
161
InterpolatorType::Pointer interpolator = InterpolatorType::New();
filter->SetInterpolator( interpolator );
The output image spacing and origin may be set via SetOutputSpacing(), SetOutputOrigin(). This is
taken from the deformation field.
filter->SetOutputSpacing( deformationField->GetSpacing() );
filter->SetOutputOrigin( deformationField->GetOrigin() );
filter->SetOutputDirection( deformationField->GetDirection() );
filter->SetDisplacementField( deformationField );
Subsampling and image in the same space
The source code for this section can be found in the file
SubsampleVolume.cxx.
This example illustrates how to perform subsampling of a volume using ITK classes. In order
to avoid aliasing artifacts, the volume must be processed by a low-pass filter before resampling.
Here we use the itk::RecursiveGaussianImageFilter as a low-pass filter. The image is then
resampled by using three different factors, one per dimension of the image.
The most important headers to include here are those corresponding to the resampling image filter,
the transform, the interpolator and the smoothing filter.
#include "itkResampleImageFilter.h"
#include "itkIdentityTransform.h"
#include "itkRecursiveGaussianImageFilter.h"
We explicitly instantiate the pixel type and dimension of the input image, and the images that will
be used internally for computing the resampling.
constexpr unsigned int Dimension = 3;
using InputPixelType = unsigned char;
using InternalPixelType = float;
using OutputPixelType = unsigned char;
using InputImageType = itk::Image< InputPixelType,
Dimension >;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType,
Dimension >;
162
Chapter 2. Filtering
In this particular case we take the factors for resampling directly from the command line arguments.
const double factorX = std::stod( argv[3] );
const double factorY = std::stod( argv[4] );
const double factorZ = std::stod( argv[5] );
A casting filter is instantiated in order to convert the pixel type of the input image into the pixel type
desired for computing the resampling.
using CastFilterType = itk::CastImageFilter< InputImageType,
InternalImageType >;
CastFilterType::Pointer caster = CastFilterType::New();
caster->SetInput( inputImage );
The smoothing filter of choice is the RecursiveGaussianImageFilter. We create three of them
in order to have the freedom of performing smoothing with different sigma values along each dimension.
using GaussianFilterType = itk::RecursiveGaussianImageFilter<
InternalImageType,
InternalImageType >;
GaussianFilterType::Pointer smootherX = GaussianFilterType::New();
GaussianFilterType::Pointer smootherY = GaussianFilterType::New();
GaussianFilterType::Pointer smootherZ = GaussianFilterType::New();
The smoothing filters are connected in a cascade in the pipeline.
smootherX->SetInput( caster->GetOutput() );
smootherY->SetInput( smootherX->GetOutput() );
smootherZ->SetInput( smootherY->GetOutput() );
The sigma values to use in the smoothing filters are computed based on the pixel spacing of the input
image and the factors provided as arguments.
const InputImageType::SpacingType& inputSpacing = inputImage->GetSpacing();
const double sigmaX = inputSpacing[0] * factorX;
const double sigmaY = inputSpacing[1] * factorY;
const double sigmaZ = inputSpacing[2] * factorZ;
smootherX->SetSigma( sigmaX );
smootherY->SetSigma( sigmaY );
smootherZ->SetSigma( sigmaZ );
2.9. Geometric Transformations
163
We instruct each one of the smoothing filters to act along a particular direction of the image, and set
them to use normalization across scale space in order to account for the reduction of intensity that
accompanies the diffusion process associated with the Gaussian smoothing.
smootherX->SetDirection( 0 );
smootherY->SetDirection( 1 );
smootherZ->SetDirection( 2 );
smootherX->SetNormalizeAcrossScale( false );
smootherY->SetNormalizeAcrossScale( false );
smootherZ->SetNormalizeAcrossScale( false );
The type of the resampling filter is instantiated using the internal image type and the output image
type.
using ResampleFilterType = itk::ResampleImageFilter<
InternalImageType, OutputImageType >;
ResampleFilterType::Pointer resampler = ResampleFilterType::New();
Since the resampling is performed in the same physical extent of the input image, we select the
IdentityTransform as the one to be used by the resampling filter.
using TransformType = itk::IdentityTransform< double, Dimension >;
TransformType::Pointer transform = TransformType::New();
transform->SetIdentity();
resampler->SetTransform( transform );
The Linear interpolator is selected because it provides a good run-time performance.
For applications that require better precision you may want to replace this interpolator with the
itk::BSplineInterpolateImageFunction interpolator or with the
itk::WindowedSincInterpolateImageFunction interpolator.
using InterpolatorType = itk::LinearInterpolateImageFunction<
InternalImageType, double >;
InterpolatorType::Pointer interpolator = InterpolatorType::New();
resampler->SetInterpolator( interpolator );
The spacing to be used in the grid of the resampled image is computed using the input image spacing
and the factors provided in the command line arguments.
OutputImageType::SpacingType spacing;
spacing[0] = inputSpacing[0] * factorX;
spacing[1] = inputSpacing[1] * factorY;
164
Chapter 2. Filtering
spacing[2] = inputSpacing[2] * factorZ;
resampler->SetOutputSpacing( spacing );
The origin and direction of the input image are both preserved and passed to the output image.
resampler->SetOutputOrigin( inputImage->GetOrigin() );
resampler->SetOutputDirection( inputImage->GetDirection() );
The number of pixels to use along each direction on the grid of the resampled image is computed
using the number of pixels in the input image and the sampling factors.
InputImageType::SizeType inputSize =
inputImage->GetLargestPossibleRegion().GetSize();
using SizeValueType = InputImageType::SizeType::SizeValueType;
InputImageType::SizeType
size;
size[0] = static_cast< SizeValueType >( inputSize[0] / factorX );
size[1] = static_cast< SizeValueType >( inputSize[1] / factorY );
size[2] = static_cast< SizeValueType >( inputSize[2] / factorZ );
resampler->SetSize( size );
Finally, the input to the resampler is taken from the output of the smoothing filter.
resampler->SetInput( smootherZ->GetOutput() );
At this point we can trigger the execution of the resampling by calling the Update() method, or we
can choose to pass the output of the resampling filter to another section of pipeline, for example, an
image writer.
Resampling an Anisotropic image to make it Isotropic
The source code for this section can be found in the file
ResampleVolumesToBeIsotropic.cxx.
It is unfortunate that it is still very common to find medical image datasets that have been acquired
with large inter-slice spacings that result in voxels with anisotropic shapes. In many cases these
voxels have ratios of [1 : 5] or even [1 : 10] between the resolution in the plane (x, y) and the resolution along the z axis. These datasets are close to useless for the purpose of computer-assisted
image analysis. The abundance of datasets acquired with anisotropic voxel sizes bespeaks a dearth
of understanding of the third dimension and its importance for medical image analysis in clinical
2.9. Geometric Transformations
165
settings and radiology reading rooms. Datasets acquired with large anisotropies bring with them the
regressive message: “I do not think 3D is informative”. They stubbornly insist: “all that you need
to know, can be known by looking at individual slices, one by one”. However, the fallacy of this
statement is made evident by simply viewing the slices when reconstructed in any of the orthogonal planes. The rectangular pixel shape is ugly and distorted, and cripples any signal processing
algorithm not designed specifically for this type of image.
Image analysts have a long educational battle to fight in the radiological setting in order to bring the
message that 3D datasets acquired with anisotropies larger than [1 : 2] are simply dismissive of the
most fundamental concept of digital signal processing: The Shannon Sampling Theorem [57, 58].
Facing the inertia of many clinical imaging departments and their blithe insistence that these images
are “good enough” for image processing, some image analysts have stoically tried to deal with these
poor datasets. These image analysts usually proceed to subsample the high in-plane resolution and
to super-sample the inter-slice resolution with the purpose of faking the type of dataset that they
should have received in the first place: an isotropic dataset. This example is an illustration of how
such an operation can be performed using the filters available in the Insight Toolkit.
Note that this example is not presented here as a solution to the problem of anisotropic datasets. On
the contrary, this is simply a dangerous palliative which will only perpetuate the errant convictions
of image acquisition departments. The real solution to the problem of the anisotropic dataset is to
educate radiologists regarding the principles of image processing. If you really care about the technical decency of the medical image processing field, and you really care about providing your best
effort to the patients who will receive health care directly or indirectly affected by your processed
images, then it is your duty to reject anisotropic datasets and to patiently explain to your radiologist why anisotropic data are problematic for processing, and require crude workarounds which
handicap your ability to draw accurate conclusions from the data and preclude his or her ability to
provide quality care. Any barbarity such as a [1 : 5] anisotropy ratio should be considered as a mere
collection of slices, and not an authentic 3D dataset.
Please, before employing the techniques covered in this section, do kindly invite your fellow radiologist to see the dataset in an orthogonal slice. Magnify that image in a viewer without any linear
interpolation until you see the daunting reality of the rectangular pixels. Let her/him know how absurd it is to process digital data which have been sampled at ratios of [1 : 5] or [1 : 10]. Then, inform
them that your only option is to throw away all that high in-plane resolution and to make up data
between the slices in order to compensate for the low resolution. Only then will you be justified in
using the following code.
Let’s now move into the code. It is appropriate for you to experience guilt5 , because your use
the code below is the evidence that we have lost one more battle on the quest for real 3D dataset
processing.
This example performs subsampling on the in-plane resolution and performs super-sampling along
the inter-slices resolution. The subsampling process requires that we preprocess the data with a
5 A feeling of regret or remorse for having committed some improper act; a recognition of one’s own responsibility for
doing something wrong.
166
Chapter 2. Filtering
smoothing filter in order to avoid the occurrence of aliasing effects due to overlap of the spectrum
in the frequency domain [57, 58]. The smoothing is performed here using the RecursiveGaussian
filter, because it provides a convenient run-time performance.
The first thing that you will need to do in order to resample this ugly anisotropic dataset is to include
the header files for the itk::ResampleImageFilter, and the Gaussian smoothing filter.
#include "itkResampleImageFilter.h"
#include "itkRecursiveGaussianImageFilter.h"
The resampling filter will need a Transform in order to map point coordinates and will need an
interpolator in order to compute intensity values for the new resampled image. In this particular case
we use the itk::IdentityTransform because the image is going to be resampled by preserving
the physical extent of the sampled region. The Linear interpolator is used as a common trade-off6.
#include "itkIdentityTransform.h"
Note that, as part of the preprocessing of the image, in this example we are also rescaling the
range of intensities. This operation has already been described as Intensity Windowing. In a real
clinical application, this step requires careful consideration of the range of intensities that contain
information about the anatomical structures that are of interest for the current clinical application. It
practice you may want to remove this step of intensity rescaling.
#include "itkIntensityWindowingImageFilter.h"
We make explicit now our choices for the pixel type and dimension of the input image to be processed, as well as the pixel type that we intend to use for the internal computation during the smoothing and resampling.
constexpr unsigned int Dimension = 3;
using InputPixelType = unsigned short;
using InternalPixelType = float;
using InputImageType = itk::Image< InputPixelType,
Dimension >;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
We instantiate the smoothing filter that will be used on the preprocessing for subsampling the inplane resolution of the dataset.
6 Although arguably we should use one type of interpolator for the in-plane subsampling process and another one for
the inter-slice supersampling. But again, one should wonder why we apply any technical sophistication here, when we are
covering up for an improper acquisition of medical data, trying to make it look as if it was correctly acquired.
2.9. Geometric Transformations
167
using GaussianFilterType = itk::RecursiveGaussianImageFilter<
InternalImageType,
InternalImageType >;
We create two instances of the smoothing filter: one will smooth along the X direction while the
other will smooth along the Y direction. They are connected in a cascade in the pipeline, while
taking their input from the intensity windowing filter. Note that you may want to skip the intensity
windowing scale and simply take the input directly from the reader.
GaussianFilterType::Pointer smootherX = GaussianFilterType::New();
GaussianFilterType::Pointer smootherY = GaussianFilterType::New();
smootherX->SetInput( intensityWindowing->GetOutput() );
smootherY->SetInput( smootherX->GetOutput() );
We must now provide the settings for the resampling itself. This is done by searching for a value
of isotropic resolution that will provide a trade-off between the evil of subsampling and the evil
of supersampling. We advance here the conjecture that the geometrical mean between the in-plane
and the inter-slice resolutions should be a convenient isotropic resolution to use. This conjecture
is supported on nothing other than intuition and common sense. You can rightfully argue that this
choice deserves a more technical consideration, but then, if you are so concerned about the technical
integrity of the image sampling process, you should not be using this code, and should discuss these
issues with the radiologist who acquired this ugly anisotropic dataset.
We take the image from the input and then request its array of pixel spacing values.
InputImageType::ConstPointer inputImage = reader->GetOutput();
const InputImageType::SpacingType& inputSpacing = inputImage->GetSpacing();
and apply our ad-hoc conjecture that the correct anisotropic resolution to use is the geometrical mean
of the in-plane and inter-slice resolutions. Then set this spacing as the Sigma value to be used for
the Gaussian smoothing at the preprocessing stage.
const double isoSpacing = std::sqrt( inputSpacing[2] * inputSpacing[0] );
smootherX->SetSigma( isoSpacing );
smootherY->SetSigma( isoSpacing );
We instruct the smoothing filters to act along the X and Y direction respectively.
smootherX->SetDirection( 0 );
smootherY->SetDirection( 1 );
168
Chapter 2. Filtering
Now that we have taken care of the smoothing in-plane, we proceed to instantiate the resampling
filter that will reconstruct an isotropic image. We start by declaring the pixel type to be used as the
output of this filter, then instantiate the image type and the type for the resampling filter. Finally we
construct an instantiation of the filter.
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType,
Dimension >;
using ResampleFilterType = itk::ResampleImageFilter<
InternalImageType, OutputImageType >;
ResampleFilterType::Pointer resampler = ResampleFilterType::New();
The resampling filter requires that we provide a Transform, which in this particular case can simply
be an identity transform.
using TransformType = itk::IdentityTransform< double, Dimension >;
TransformType::Pointer transform = TransformType::New();
transform->SetIdentity();
resampler->SetTransform( transform );
The filter also requires an interpolator to be passed to it. In this case we chose to use a linear
interpolator.
using InterpolatorType = itk::LinearInterpolateImageFunction<
InternalImageType, double >;
InterpolatorType::Pointer interpolator = InterpolatorType::New();
resampler->SetInterpolator( interpolator );
The pixel spacing of the resampled dataset is loaded in a SpacingType and passed to the resampling
filter.
OutputImageType::SpacingType spacing;
spacing[0] = isoSpacing;
spacing[1] = isoSpacing;
spacing[2] = isoSpacing;
resampler->SetOutputSpacing( spacing );
The origin and orientation of the output image is maintained, since we decided to resample the image
in the same physical extent of the input anisotropic image.
2.9. Geometric Transformations
169
resampler->SetOutputOrigin( inputImage->GetOrigin() );
resampler->SetOutputDirection( inputImage->GetDirection() );
The number of pixels to use along each dimension in the grid of the resampled image is computed
using the ratio between the pixel spacings of the input image and those of the output image. Note that
the computation of the number of pixels along the Z direction is slightly different with the purpose
of making sure that we don’t attempt to compute pixels that are outside of the original anisotropic
dataset.
InputImageType::SizeType inputSize =
inputImage->GetLargestPossibleRegion().GetSize();
using SizeValueType = InputImageType::SizeType::SizeValueType;
const double dx = inputSize[0] * inputSpacing[0] / isoSpacing;
const double dy = inputSize[1] * inputSpacing[1] / isoSpacing;
const double dz = (inputSize[2] - 1 ) * inputSpacing[2] / isoSpacing;
Finally the values are stored in a SizeType and passed to the resampling filter. Note that this
process requires a casting since the computations are performed in double, while the elements of
the SizeType are integers.
InputImageType::SizeType
size;
size[0] = static_cast( dx );
size[1] = static_cast( dy );
size[2] = static_cast( dz );
resampler->SetSize( size );
Our last action is to take the input for the resampling image filter from the output of the cascade of
smoothing filters, and then to trigger the execution of the pipeline by invoking the Update() method
on the resampling filter.
resampler->SetInput( smootherY->GetOutput() );
resampler->Update();
At this point we should take a moment in silence to reflect on the circumstances that have led us to
accept this cover-up for the improper acquisition of medical data.
170
Chapter 2. Filtering
2.10 Frequency Domain
2.10.1 Computing a Fast Fourier Transform (FFT)
The source code for this section can be found in the file
FFTImageFilter.cxx.
In this section we assume that you are familiar with Spectral Analysis, in particular with the concepts
of the Fourier Transform and the numerical implementation of the Fast Fourier transform. If you are
not familiar with these concepts you may want to consult first any of the many available introductory
books to spectral analysis [8, 9].
This example illustrates how to use the Fast Fourier Transform filter (FFT) for processing an
image in the spectral domain. Given that FFT computation can be CPU intensive, there are
multiple hardware specific implementations of FFT. It is convenient in many cases to delegate the actual computation of the transform to local available libraries. Particular examples
of those libraries are fftw7 and the VXL implementation of FFT. For this reason ITK provides a base abstract class that factorizes the interface to multiple specific implementations of
FFT. This base class is the itk::ForwardFFTImageFilter, and two of its derived classes are
itk::VnlForwardFFTImageFilter and itk::FFTWRealToComplexConjugateImageFilter.
A typical application that uses FFT will need to include the following header files.
#include
#include
#include
#include
"itkImage.h"
"itkVnlForwardFFTImageFilter.h"
"itkComplexToRealImageFilter.h"
"itkComplexToImaginaryImageFilter.h"
The first decision to make is related to the pixel type and dimension of the images on which we want
to compute the Fourier transform.
using PixelType = float;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
We use the same image type in order to instantiate the FFT filter, in this case the
itk::VnlForwardFFTImageFilter. Once the filter type is instantiated, we can use it for creating one object by invoking the New() method and assigning the result to a SmartPointer.
using FFTFilterType = itk::VnlForwardFFTImageFilter< ImageType >;
FFTFilterType::Pointer fftFilter = FFTFilterType::New();
7 http://www.fftw.org
2.10. Frequency Domain
171
The input to this filter can be taken from a reader, for example.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
fftFilter->SetInput( reader->GetOutput() );
The execution of the filter can be triggered by invoking the Update() method. Since this invocation
can eventually throw an exception, the call must be placed inside a try/catch block.
try
{
fftFilter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Error: " << std::endl;
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
In general the output of the FFT filter will be a complex image. We can proceed to save this image
in a file for further analysis. This can be done by simply instantiating an itk::ImageFileWriter
using the trait of the output image from the FFT filter. We construct one instance of the writer and
pass the output of the FFT filter as the input of the writer.
using ComplexImageType = FFTFilterType::OutputImageType;
using ComplexWriterType = itk::ImageFileWriter< ComplexImageType >;
ComplexWriterType::Pointer complexWriter = ComplexWriterType::New();
complexWriter->SetFileName( argv[4] );
complexWriter->SetInput( fftFilter->GetOutput() );
Finally we invoke the Update() method placed inside a try/catch block.
try
{
complexWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Error: " << std::endl;
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
172
Chapter 2. Filtering
In addition to saving the complex image into a file, we could also extract its real and imaginary
parts for further analysis. This can be done with the itk::ComplexToRealImageFilter and the
itk::ComplexToImaginaryImageFilter.
We instantiate first the ImageFilter that will help us to extract the real part from the complex image.
The ComplexToRealImageFilter takes as its first template parameter the type of the complex
image and as its second template parameter it takes the type of the output image pixel. We create
one instance of this filter and connect as its input the output of the FFT filter.
using RealFilterType = itk::ComplexToRealImageFilter<
ComplexImageType, ImageType >;
RealFilterType::Pointer realFilter = RealFilterType::New();
realFilter->SetInput( fftFilter->GetOutput() );
Since the range of intensities in the Fourier domain can be quite concentrated, it is convenient to rescale the image in order to visualize it. For this purpose we instantiate a
itk::RescaleIntensityImageFilter that will rescale the intensities of the real image into a
range suitable for writing in a file. We also set the minimum and maximum values of the output to
the range of the pixel type used for writing.
using RescaleFilterType = itk::RescaleIntensityImageFilter<
ImageType,
WriteImageType >;
RescaleFilterType::Pointer intensityRescaler = RescaleFilterType::New();
intensityRescaler->SetInput( realFilter->GetOutput() );
intensityRescaler->SetOutputMinimum( 0 );
intensityRescaler->SetOutputMaximum( 255 );
We can now instantiate the ImageFilter that will help us to extract the imaginary part from the
complex image. The filter that we use here is the itk::ComplexToImaginaryImageFilter. It
takes as first template parameter the type of the complex image and as second template parameter it
takes the type of the output image pixel. An instance of the filter is created, and its input is connected
to the output of the FFT filter.
using ComplexImageType = FFTFilterType::OutputImageType;
using ImaginaryFilterType = itk::ComplexToImaginaryImageFilter<
ComplexImageType, ImageType >;
ImaginaryFilterType::Pointer imaginaryFilter = ImaginaryFilterType::New();
imaginaryFilter->SetInput( fftFilter->GetOutput() );
2.10. Frequency Domain
173
The Imaginary image can then be rescaled and saved into a file, just as we did with the Real part.
For the sake of illustrating the use of a itk::ImageFileReader on Complex images, here we
instantiate a reader that will load the Complex image that we just saved. Note that nothing special
is required in this case. The instantiation is done just the same as for any other type of image, which
once again illustrates the power of Generic Programming.
using ComplexReaderType = itk::ImageFileReader< ComplexImageType >;
ComplexReaderType::Pointer complexReader = ComplexReaderType::New();
complexReader->SetFileName( argv[4] );
complexReader->Update();
2.10.2 Filtering on the Frequency Domain
The source code for this section can be found in the file
FFTImageFilterFourierDomainFiltering.cxx.
One of the most common image processing operations performed in the Fourier Domain is the
masking of the spectrum in order to eliminate a range of spatial frequencies from the input image.
This operation is typically performed by taking the input image, computing its Fourier transform
using a FFT filter, masking the resulting image in the Fourier domain with a mask, and finally
taking the result of the masking and computing its inverse Fourier transform.
This typical process is illustrated in the example below.
We start by including the headers of the FFT filters and the Mask image filter. Note that we use two
different types of FFT filters here. The first one expects as input an image of real pixel type (real
in the sense of complex numbers) and produces as output a complex image. The second FFT filter
expects as in put a complex image and produces a real image as output.
#include "itkVnlForwardFFTImageFilter.h"
#include "itkVnlInverseFFTImageFilter.h"
#include "itkMaskImageFilter.h"
The first decision to make is related to the pixel type and dimension of the images on which we want
to compute the Fourier transform.
using InputPixelType = float;
constexpr unsigned int Dimension = 2;
using InputImageType = itk::Image< InputPixelType, Dimension >;
Then we select the pixel type to use for the mask image and instantiate the image type of the mask.
174
Chapter 2. Filtering
using MaskPixelType = unsigned char;
using MaskImageType = itk::Image< MaskPixelType, Dimension >;
Both the input image and the mask image can be read from files or could be obtained as the output
of a preprocessing pipeline. We omit here the details of reading the image since the process is quite
standard.
Now the itk::VnlForwardFFTImageFilter can be instantiated. Like most ITK filters, the FFT
filter is instantiated using the full image type. By not setting the output image type, we decide to use
the default one provided by the filter. Using this type we construct one instance of the filter.
using FFTFilterType = itk::VnlForwardFFTImageFilter< InputImageType >;
FFTFilterType::Pointer fftFilter = FFTFilterType::New();
fftFilter->SetInput( inputReader->GetOutput() );
Since our purpose is to perform filtering in the frequency domain by altering the weights of the
image spectrum, we need a filter that will mask the Fourier transform of the input image with a
binary image. Note that the type of the spectral image is taken here from the traits of the FFT filter.
using SpectralImageType = FFTFilterType::OutputImageType;
using MaskFilterType = itk::MaskImageFilter< SpectralImageType,
MaskImageType, SpectralImageType >;
MaskFilterType::Pointer maskFilter = MaskFilterType::New();
We connect the inputs to the mask filter by taking the outputs from the first FFT filter and from the
reader of the Mask image.
maskFilter->SetInput1( fftFilter->GetOutput() );
maskFilter->SetInput2( maskReader->GetOutput() );
For the purpose of verifying the aspect of the spectrum after being filtered with the mask, we can
write out the output of the Mask filter to a file.
using SpectralWriterType = itk::ImageFileWriter< SpectralImageType >;
SpectralWriterType::Pointer spectralWriter = SpectralWriterType::New();
spectralWriter->SetFileName("filteredSpectrum.mhd");
spectralWriter->SetInput( maskFilter->GetOutput() );
spectralWriter->Update();
The output of the mask filter will contain the filtered spectrum of the input image. We must then
2.11. Extracting Surfaces
175
apply an inverse Fourier transform on it in order to obtain the filtered version of the input image.
For that purpose we create another instance of the FFT filter.
using IFFTFilterType = itk::VnlInverseFFTImageFilter;
IFFTFilterType::Pointer fftInverseFilter = IFFTFilterType::New();
fftInverseFilter->SetInput( maskFilter->GetOutput() );
The execution of the pipeline can be triggered by invoking the Update() method in this last filter.
Since this invocation can eventually throw an exception, the call must be placed inside a try/catch
block.
try
{
fftInverseFilter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Error: " << std::endl;
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
The result of the filtering can now be saved into an image file, or be passed to a subsequent processing pipeline. Here we simply write it out to an image file.
using WriterType = itk::ImageFileWriter< InputImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[3] );
writer->SetInput( fftInverseFilter->GetOutput() );
Note that this example is just a minimal illustration of the multiple types of processing that are
possible in the Fourier domain.
2.11 Extracting Surfaces
2.11.1 Surface extraction
The source code for this section can be found in the file
SurfaceExtraction.cxx.
Surface extraction has attracted continuous interest since the early days of image analysis, especially
in the context of medical applications. Although it is commonly associated with image segmentation, surface extraction is not in itself a segmentation technique, instead it is a transformation that
176
Chapter 2. Filtering
changes the way a segmentation is represented. In its most common form, isosurface extraction is
the equivalent of image thresholding followed by surface extraction.
Probably the most widely known method of surface extraction is the Marching Cubes algorithm [36].
Although it has been followed by a number of variants [54], Marching Cubes has become an icon in
medical image processing. The following example illustrates how to perform surface extraction in
ITK using an algorithm similar to Marching Cubes 8 .
The representation of unstructured data in ITK is done with the itk::Mesh. This class enables us
to represent N-Dimensional grids of varied topology. It is natural for the filter that extracts surfaces
from an image to produce a mesh as its output.
We initiate our example by including the header files of the surface extraction filter, the image and
the mesh.
#include "itkBinaryMask3DMeshSource.h"
#include "itkImage.h"
We define then the pixel type and dimension of the image from which we are going to extract the
surface.
constexpr unsigned int Dimension = 3;
using PixelType = unsigned char;
using ImageType = itk::Image< PixelType, Dimension >;
With the same image type we instantiate the type of an ImageFileReader and construct one with the
purpose of reading in the input image.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
The type of the itk::Mesh is instantiated by specifying the type to be associated with the pixel
value of the Mesh nodes. This particular pixel type happens to be irrelevant for the purpose of
extracting the surface.
using MeshType = itk::Mesh;
Having declared the Image and Mesh types we can now instantiate the surface extraction filter, and
construct one by invoking its New() method.
8 Note
that the Marching Cubes algorithm is covered by a patent that expired on June 5th 2005.
2.11. Extracting Surfaces
177
using MeshSourceType = itk::BinaryMask3DMeshSource< ImageType, MeshType >;
MeshSourceType::Pointer meshSource = MeshSourceType::New();
In this example, the pixel value associated with the object to be extracted is read from the command
line arguments and it is passed to the filter by using the SetObjectValue() method. Note that this
is different from the traditional isovalue used in the Marching Cubes algorithm. In the case of the
BinaryMask3DMeshSource filter, the object values define the membership of pixels to the object
from which the surface will be extracted. In other words, the surface will be surrounding all pixels
with value equal to the ObjectValue parameter.
const auto objectValue = static_cast( std::stod( argv[2] ) );
meshSource->SetObjectValue( objectValue );
The input to the surface extraction filter is taken from the output of the image reader.
meshSource->SetInput( reader->GetOutput() );
Finally we trigger the execution of the pipeline by invoking the Update() method. Given that the
pipeline may throw an exception this call must be place inside a try/catch block.
try
{
meshSource->Update();
}
catch( itk::ExceptionObject & exp )
{
std::cerr << "Exception thrown during Update() " << std::endl;
std::cerr << exp << std::endl;
return EXIT_FAILURE;
}
We print out the number of nodes and cells in order to inspect the output mesh.
std::cout << "Nodes = " << meshSource->GetNumberOfNodes() << std::endl;
std::cout << "Cells = " << meshSource->GetNumberOfCells() << std::endl;
This resulting Mesh could be used as input for a deformable model segmentation algorithm, or it
could be converted to a format suitable for visualization in an interactive application.
CHAPTER
THREE
REGISTRATION
This chapter introduces ITK’s capabilities for performing image registration.
T
q
p
Image registration is the process of
determining the spatial transform that
maps points from one image to homologous points on a object in the second
image. This concept is schematically
represented in Figure 3.1. In ITK, regis- Figure 3.1: Image registration is the task of finding a spatration is performed within a framework tial transform mapping one image into another.
of pluggable components that can easily
be interchanged. This flexibility means that a combinatorial variety of registration methods can be
created, allowing users to pick and choose the right tools for their specific application.
3.1 Registration Framework
Let’s begin with a simplified typical registration framework where its components and their interconnections are shown in Figure 3.2. The basic input data to the registration process are two images:
one is defined as the fixed image f (X) and the other as the moving image m(X), where X represents
a position in N-dimensional space. Registration is treated as an optimization problem with the goal
of finding the spatial mapping that will bring the moving image into alignment with the fixed image.
The transform component T (X) represents the spatial mapping of points from the fixed image space
to points in the moving image space. The interpolator is used to evaluate moving image intensities
at non-grid positions. The metric component S( f , m ◦ T ) provides a measure of how well the fixed
image is matched by the transformed moving image. This measure forms a quantitative criterion to
be optimized by the optimizer over the search space defined by the parameters of the transform.
ITKv4 registration framework provides more flexibility to the above traditional registration concept.
In this new framework, the registration computations can happen on a physical grid completely
different than the fixed image domain having different sampling density. This “sampling domain” is
180
Chapter 3. Registration
pixels
Fixed Image
fitness value
Metric
pixels
Optimizer
Interpolator
Transform
parameters
Moving Image
pixels
points
Transform
Figure 3.2: The basic components of a typical registration framework are two input images, a transform, a
metric, an interpolator and an optimizer.
Figure 3.3: The basic components of the ITKv4 registration framework.
considered as a new component in the registration framework known as virtual image that can be
an arbitrary set of physical points, not necessarily a uniform grid of points.
Various ITKv4 registration components are illustrated in Figure 3.3. Boxes with dashed borders
show data objects, while those with solid borders show process objects.
The matching Metric class is a key component that controls most parts of the registration process
since it handles fixed, moving and virtual images as well as fixed and moving transforms and interpolators.
Fixed and moving transforms and interpolators are used by the metric to evaluate the intensity values
of the fixed and moving images at each physical point of the virtual space. Those intensity values
are then used by the metric cost function to evaluate the fitness value and derivatives, which are
passed to the optimizer that asks the moving transform to update its parameters based on the outputs
of the cost function. Since the moving transform is shared between metric and optimizer, the above
3.2. ”Hello World” Registration
181
process will be repeated till the convergence criteria are met.
Later in section 3.3 you will get a better understanding of the behind-the-scenes processes of ITKv4
registration framework. First, we begin with some simple registration examples.
3.2 ”Hello World” Registration
The source code for this section can be found in the file
ImageRegistration1.cxx.
This example illustrates the use of the image registration framework in Insight. It should be read as
a “Hello World” for ITK registration. Instead of means to an end, this example should be read as a
basic introduction to the elements typically involved when solving a problem of image registration.
A registration method requires the following set of components: two input images, a transform, a
metric and an optimizer. Some of these components are parameterized by the image type for which
the registration is intended. The following header files provide declarations of common types used
for these components.
#include
#include
#include
#include
"itkImageRegistrationMethodv4.h"
"itkTranslationTransform.h"
"itkMeanSquaresImageToImageMetricv4.h"
"itkRegularStepGradientDescentOptimizerv4.h"
The type of each registration component should be instantiated first. We start by selecting the image
dimension and the types to be used for representing image pixels.
constexpr unsigned int Dimension = 2;
using PixelType = float;
The types of the input images are instantiated by the following lines.
using FixedImageType = itk::Image< PixelType, Dimension >;
using MovingImageType = itk::Image< PixelType, Dimension >;
The transform that will map the fixed image space into the moving image space is defined below.
using TransformType = itk::TranslationTransform< double, Dimension >;
An optimizer is required to explore the parameter space of the transform in search of optimal values
of the metric.
182
Chapter 3. Registration
using OptimizerType = itk::RegularStepGradientDescentOptimizerv4;
The metric will compare how well the two images match each other. Metric types are usually
templated over the image types as seen in the following type declaration.
using MetricType = itk::MeanSquaresImageToImageMetricv4<
FixedImageType,
MovingImageType >;
The registration method type is instantiated using the types of the fixed and moving images as well
as the output transform type. This class is responsible for interconnecting all the components that
we have described so far.
using RegistrationType = itk::ImageRegistrationMethodv4<
FixedImageType,
MovingImageType,
TransformType
>;
Each one of the registration components is created using its New() method and is assigned to its
respective itk::SmartPointer.
MetricType::Pointer
OptimizerType::Pointer
RegistrationType::Pointer
metric
= MetricType::New();
optimizer
= OptimizerType::New();
registration = RegistrationType::New();
Each component is now connected to the instance of the registration method.
registration->SetMetric(
registration->SetOptimizer(
metric
optimizer
);
);
In this example the transform object does not need to be created and passed to the registration
method like above since the registration filter will instantiate an internal transform object using the
transform type that is passed to it as a template parameter.
Metric needs an interpolator to evaluate the intensities of the fixed and moving images at non-grid
positions. The types of fixed and moving interpolators are declared here.
using FixedLinearInterpolatorType = itk::LinearInterpolateImageFunction<
FixedImageType,
double >;
using MovingLinearInterpolatorType = itk::LinearInterpolateImageFunction<
MovingImageType,
double >;
3.2. ”Hello World” Registration
183
Then, fixed and moving interpolators are created and passed to the metric. Since linear interpolators
are used as default, we could skip the following step in this example.
FixedLinearInterpolatorType::Pointer fixedInterpolator =
FixedLinearInterpolatorType::New();
MovingLinearInterpolatorType::Pointer movingInterpolator =
MovingLinearInterpolatorType::New();
metric->SetFixedInterpolator( fixedInterpolator );
metric->SetMovingInterpolator( movingInterpolator );
In this example, the fixed and moving images are read from files.
This requires the
itk::ImageRegistrationMethodv4 to acquire its inputs from the output of the readers.
registration->SetFixedImage(
registration->SetMovingImage(
fixedImageReader->GetOutput()
movingImageReader->GetOutput()
);
);
Now the registration process should be initialized. ITKv4 registration framework provides initial
transforms for both fixed and moving images. These transforms can be used to setup an initial
known correction of the misalignment between the virtual domain and fixed/moving image spaces.
In this particular case, a translation transform is being used for initialization of the moving image
space. The array of parameters for the initial moving transform is simply composed of the translation
values along each dimension. Setting the values of the parameters to zero initializes the transform to
an Identity transform. Note that the array constructor requires the number of elements to be passed
as an argument.
TransformType::Pointer movingInitialTransform = TransformType::New();
TransformType::ParametersType initialParameters(
movingInitialTransform->GetNumberOfParameters() );
initialParameters[0] = 0.0; // Initial offset in mm along X
initialParameters[1] = 0.0; // Initial offset in mm along Y
movingInitialTransform->SetParameters( initialParameters );
registration->SetMovingInitialTransform( movingInitialTransform );
In the registration filter this moving initial transform will be added to a composite transform that
already includes an instantiation of the output optimizable transform; then, the resultant composite
transform will be used by the optimizer to evaluate the metric values at each iteration.
Despite this, the fixed initial transform does not contribute to the optimization process. It is only
used to access the fixed image from the virtual image space where the metric evaluation happens.
Virtual images are a new concept added to the ITKv4 registration framework, which potentially lets
us to do the registration process in a physical domain totally different from the fixed and moving
image domains. In fact, the region over which metric evaluation is performed is called virtual image
184
Chapter 3. Registration
domain. This domain defines the resolution at which the evaluation is performed, as well as the
physical coordinate system.
The virtual reference domain is taken from the “virtual image” buffered region, and the input images
should be accessed from this reference space using the fixed and moving initial transforms.
The legacy intuitive registration framework can be considered as a special case where the virtual
domain is the same as the fixed image domain. As this case practically happens in most of the real
life applications, the virtual image is set to be the same as the fixed image by default. However,
the user can define the virtual domain differently than the fixed image domain by calling either
SetVirtualDomain or SetVirtualDomainFromImage.
In this example, like the most examples of this chapter, the virtual image is considered the same as
the fixed image. Since the registration process happens in the fixed image physical domain, the fixed
initial transform maintains its default value of identity and does not need to be set.
However, a “Hello World!” example should show all the basics, so all the registration components
are explicity set here.
In the next section of this chapter, you will get a better understanding from behind the scenes of the
registration process when the initial fixed transform is not identity.
TransformType::Pointer identityTransform = TransformType::New();
identityTransform->SetIdentity();
registration->SetFixedInitialTransform( identityTransform );
Note that the above process shows only one way of initializing the registration configuration. Another option is to initialize the output optimizable transform directly. In this approach, a transform object is created, initialized, and then passed to the registration method via
SetInitialTransform(). This approach is shown in section 3.6.1.
At this point the registration method is ready for execution. The optimizer is the component that
drives the execution of the registration. However, the ImageRegistrationMethodv4 class orchestrates
the ensemble to make sure that everything is in place before control is passed to the optimizer.
It is usually desirable to fine tune the parameters of the optimizer. Each optimizer has particular
parameters that must be interpreted in the context of the optimization strategy it implements. The
optimizer used in this example is a variant of gradient descent that attempts to prevent it from taking
steps that are too large. At each iteration, this optimizer will take a step along the direction of
the itk::ImageToImageMetricv4 derivative. Each time the direction of the derivative abruptly
changes, the optimizer assumes that a local extrema has been passed and reacts by reducing the step
length by a relaxation factor. The reducing factor should have a value between 0 and 1. This factor
is set to 0.5 by default, and it can be changed to a different value via SetRelaxationFactor().
Also, the default value for the initial step length is 1, and this value can be changed manually with
the method SetLearningRate().
In addition to manual settings, the initial step size can also be estimated automatically, either at
3.2. ”Hello World” Registration
185
each iteration or only at the first iteration, by assigning a ScalesEstimator (as will be seen in later
examples).
After several reductions of the step length, the optimizer may be moving in a very restricted area of
the transform parameter space. By the method SetMinimumStepLength(), the user can define how
small the step length should be to consider convergence to have been reached. This is equivalent to
defining the precision with which the final transform should be known. User can also set some other
stop criteria manually like maximum number of iterations.
In
other
gradient
descent-based
optimizers
of
the
ITKv4
framework,
such
as
itk::GradientDescentLineSearchOptimizerv4
and
itk::ConjugateGradientLineSearchOptimizerv4, the convergence criteria are set via
SetMinimumConvergenceValue() which is computed based on the results of the last few iterations. The number of iterations involved in computations are defined by the convergence window
size via SetConvergenceWindowSize() which is shown in later examples of this chapter.
Also note that unlike the previous versions, ITKv4 optimizers do not have a “maximize/minimize”
option to modify the effect of the metric derivatives. Each assigned metric is assumed to return a
parameter derivative result that ”improves” the optimization.
optimizer->SetLearningRate( 4 );
optimizer->SetMinimumStepLength( 0.001 );
optimizer->SetRelaxationFactor( 0.5 );
In case the optimizer never succeeds reaching the desired precision tolerance, it is prudent to establish a limit on the number of iterations to be performed. This maximum number is defined with the
method SetNumberOfIterations().
optimizer->SetNumberOfIterations( 200 );
ITKv4 facilitates a multi-level registration framework whereby each stage is different in the resolution of its virtual space and the smoothness of the fixed and moving images. These criteria need to
be defined before registration starts. Otherwise, the default values will be used. In this example, we
run a simple registration in one level with no space shrinking or smoothing on the input data.
constexpr unsigned int numberOfLevels = 1;
RegistrationType::ShrinkFactorsArrayType shrinkFactorsPerLevel;
shrinkFactorsPerLevel.SetSize( 1 );
shrinkFactorsPerLevel[0] = 1;
RegistrationType::SmoothingSigmasArrayType smoothingSigmasPerLevel;
smoothingSigmasPerLevel.SetSize( 1 );
smoothingSigmasPerLevel[0] = 0;
registration->SetNumberOfLevels ( numberOfLevels );
registration->SetSmoothingSigmasPerLevel( smoothingSigmasPerLevel );
registration->SetShrinkFactorsPerLevel( shrinkFactorsPerLevel );
186
Chapter 3. Registration
The registration process is triggered by an invocation of the Update() method. If something goes
wrong during the initialization or execution of the registration an exception will be thrown. We
should therefore place the Update() method inside a try/catch block as illustrated in the following
lines.
try
{
registration->Update();
std::cout << "Optimizer stop condition: "
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
In a real life application, you may attempt to recover from the error by taking more effective actions
in the catch block. Here we are simply printing out a message and then terminating the execution of
the program.
The result of the registration process is obtained using the GetTransform() method that returns a
constant pointer to the output transform.
TransformType::ConstPointer transform = registration->GetTransform();
In the case of the itk::TranslationTransform, there is a straightforward interpretation of the
parameters. Each element of the array corresponds to a translation along one spatial dimension.
TransformType::ParametersType finalParameters = transform->GetParameters();
const double TranslationAlongX = finalParameters[0];
const double TranslationAlongY = finalParameters[1];
The optimizer can be queried for the actual number of iterations performed to reach convergence.
The GetCurrentIteration() method returns this value. A large number of iterations may be an
indication that the learning rate has been set too small, which is undesirable since it results in long
computational times.
const unsigned int numberOfIterations = optimizer->GetCurrentIteration();
The value of the image metric corresponding to the last set of parameters can be obtained with the
GetValue() method of the optimizer.
3.2. ”Hello World” Registration
187
Figure 3.4: Fixed and Moving image provided as input to the registration method.
const double bestValue = optimizer->GetValue();
Let’s execute this example over two of the images provided in Examples/Data:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceShifted13x17y.png
The second image is the result of intentionally translating the first image by (13, 17) millimeters.
Both images have unit-spacing and are shown in Figure 3.4. The registration takes 20 iterations and
the resulting transform parameters are:
Translation X = 13.0012
Translation Y = 16.9999
As expected, these values match quite well the misalignment that we intentionally introduced in the
moving image.
It is common, as the last step of a registration task, to use the resulting transform to map the moving
image into the fixed image space.
Before the mapping process, notice that we have not used the direct initialization of the output
transform in this example, so the parameters of the moving initial transform are not reflected in the
188
Chapter 3. Registration
output parameters of the registration filter. Hence, a composite transform is needed to concatenate
both initial and output transforms together.
using CompositeTransformType = itk::CompositeTransform<
double,
Dimension >;
CompositeTransformType::Pointer outputCompositeTransform =
CompositeTransformType::New();
outputCompositeTransform->AddTransform( movingInitialTransform );
outputCompositeTransform->AddTransform(
registration->GetModifiableTransform() );
Now the mapping process is easily done with the itk::ResampleImageFilter. Please refer to
Section 2.9.4 for details on the use of this filter. First, a ResampleImageFilter type is instantiated
using the image types. It is convenient to use the fixed image type as the output type since it is likely
that the transformed moving image will be compared with the fixed image.
using ResampleFilterType = itk::ResampleImageFilter<
MovingImageType,
FixedImageType >;
A resampling filter is created and the moving image is connected as its input.
ResampleFilterType::Pointer resampler = ResampleFilterType::New();
resampler->SetInput( movingImageReader->GetOutput() );
The created output composite transform is also passed as input to the resampling filter.
resampler->SetTransform( outputCompositeTransform );
As described in Section 2.9.4, the ResampleImageFilter requires additional parameters to be specified, in particular, the spacing, origin and size of the output image. The default pixel value is also
set to a distinct gray level in order to highlight the regions that are mapped outside of the moving
image.
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
resampler->SetSize( fixedImage->GetLargestPossibleRegion().GetSize() );
resampler->SetOutputOrigin( fixedImage->GetOrigin() );
resampler->SetOutputSpacing( fixedImage->GetSpacing() );
resampler->SetOutputDirection( fixedImage->GetDirection() );
resampler->SetDefaultPixelValue( 100 );
The output of the filter is passed to a writer that will store the image in a file. An
itk::CastImageFilter is used to convert the pixel type of the resampled image to the final type
used by the writer. The cast and writer filters are instantiated below.
3.2. ”Hello World” Registration
189
Figure 3.5: Mapped moving image and its difference with the fixed image before and after registration
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
using CastFilterType = itk::CastImageFilter<
FixedImageType,
OutputImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The filters are created by invoking their New() method.
WriterType::Pointer
writer =
CastFilterType::Pointer caster =
WriterType::New();
CastFilterType::New();
The filters are connected together and the Update() method of the writer is invoked in order to
trigger the execution of the pipeline.
caster->SetInput( resampler->GetOutput() );
writer->SetInput( caster->GetOutput()
);
writer->Update();
The fixed image and the transformed moving image can easily be compared using the
itk::SubtractImageFilter. This pixel-wise filter computes the difference between homologous
pixels of its two input images.
using DifferenceFilterType = itk::SubtractImageFilter<
FixedImageType,
FixedImageType,
190
Chapter 3. Registration
Registration Method
Reader
Fixed Image
Resample
Filter
Metric
Optimizer
Parameters
Interpolator
Reader
Moving Image
Transform
Subtract
Filter
Writer
Subtract
Filter
Writer
Transform
Resample
Filter
Figure 3.6: Pipeline structure of the registration example.
FixedImageType >;
DifferenceFilterType::Pointer difference = DifferenceFilterType::New();
difference->SetInput1( fixedImageReader->GetOutput() );
difference->SetInput2( resampler->GetOutput() );
Note that the use of subtraction as a method for comparing the images is appropriate here because
we chose to represent the images using a pixel type float. A different filter would have been used
if the pixel type of the images were any of the unsigned integer types.
Since the differences between the two images may correspond to very low values of intensity, we
rescale those intensities with a itk::RescaleIntensityImageFilter in order to make them more
visible. This rescaling will also make it possible to visualize the negative values even if we save the
difference image in a file format that only supports unsigned pixel values1 . We also reduce the
DefaultPixelValue to “1” in order to prevent that value from absorbing the dynamic range of the
differences between the two images.
using RescalerType = itk::RescaleIntensityImageFilter<
FixedImageType,
OutputImageType >;
RescalerType::Pointer intensityRescaler = RescalerType::New();
intensityRescaler->SetInput( difference->GetOutput() );
intensityRescaler->SetOutputMinimum( 0 );
intensityRescaler->SetOutputMaximum( 255 );
resampler->SetDefaultPixelValue( 1 );
Its output can be passed to another writer.
1 This
is the case of PNG, BMP, JPEG and TIFF among other common file formats.
3.2. ”Hello World” Registration
191
4500
4000
20
3500
18
16
3000
Mean Squares Metric
Y Translations (mm)
14
12
10
2500
2000
8
1500
6
4
1000
2
2
4
6
8
10
12
14
X Translations (mm)
500
0
0
5
10
15
20
Iteration No.
Figure 3.7: The sequence of translations and metric values at each iteration of the optimizer.
WriterType::Pointer writer2 = WriterType::New();
writer2->SetInput( intensityRescaler->GetOutput() );
For the purpose of comparison, the difference between the fixed image and the moving image before
registration can also be computed by simply setting the transform to an identity transform. Note
that the resampling is still necessary because the moving image does not necessarily have the same
spacing, origin and number of pixels as the fixed image. Therefore a pixel-by-pixel operation cannot
in general be performed. The resampling process with an identity transform will ensure that we have
a representation of the moving image in the grid of the fixed image.
resampler->SetTransform( identityTransform );
The complete pipeline structure of the current example is presented in Figure 3.6. The components
of the registration method are depicted as well. Figure 3.5 (left) shows the result of resampling the
moving image in order to map it onto the fixed image space. The top and right borders of the image
appear in the gray level selected with the SetDefaultPixelValue() in the ResampleImageFilter.
The center image shows the difference between the fixed image and the original moving image
(i.e. the difference before the registration is performed). The right image shows the difference
between the fixed image and the transformed moving image (i.e. after the registration has been
performed). Both difference images have been rescaled in intensity in order to highlight those pixels
where differences exist. Note that the final registration is still off by a fraction of a pixel, which
causes bands around edges of anatomical structures to appear in the difference image. A perfect
registration would have produced a null difference image.
It is always useful to keep in mind that registration is essentially an optimization problem. Figure
192
Chapter 3. Registration
3.7 helps to reinforce this notion by showing the trace of translations and values of the image metric
at each iteration of the optimizer. It can be seen from the top figure that the step length is reduced
progressively as the optimizer gets closer to the metric extrema. The bottom plot clearly shows how
the metric value decreases as the optimization advances. The log plot helps to highlight the normal
oscillations of the optimizer around the extrema value.
In this section, we used a very simple example to introduce the basic components of a registration process in ITKv4. However, studying this example alone is not enough to start using the
itk::ImageRegistrationMethodv4. In order to choose the best registration practice for a specific application, knowledge of other registration method instantiations and their capabilities are
required. For example, direct initialization of the output optimizable transform is shown in section 3.6.1. This method can simplify the registration process in many cases. Also, multi-resolution
and multistage registration approaches are illustrated in sections 3.7 and 3.8. These examples illustrate the flexibility in the usage of ITKv4 registration method framework that can help to provide
faster and more reliable registration processes.
3.3 Features of the Registration Framework
This section presents internals of the registration process in ITKv4. Understanding what actually
happens is necessary to have a correct interpretation of the results of a registration filter. It also
helps to understand the most common difficulties that users encounter when they start using the
ITKv4 registration framework:
• Registration is done in physical coordinates
• The direction of the transform maps from the space of the virtual image to that of the moving
image
These two topics tend to create confusion because they are implemented in different ways in other
systems, and community members tend to have different expectations regarding how registration
should work in ITKv4. The situation is further complicated by the way most people describe image
operations, as if they were manually performed on a continuous picture on a piece of paper.
These concepts are discussed in this section through a general example shown in Figure 3.8.
Recall that ITKv4 does the registration in “physical” space where fixed, moving and virtual images
are placed. Also, note that the term of virtual image is deceptive here since it does not refer to
any actual image. In fact, the virtual image defines the origin, direction and the spacing of a space
lattice that holds the output resampled image of the registration process. The virtual pixel lattice is
illustrated in green at the top left side of Figure 3.8.
As shown in this figure, generally there are two transforms involved in the registration process even
though only one of them is being optimized. Tvm maps points from physical virtual space onto the
physical space of the moving image, and in the same way Tv f finds homologous points between
3.3. Features of the Registration Framework
193
Figure 3.8: Different coordinate systems involved in the image registration process. Note that the transform
being optimized is the one mapping from the physical space of the virtual image into the physical space of the
moving image.
194
Chapter 3. Registration
physical virtual space and the physical space of the fixed image. Note that only Tvm is optimized
during the registration process. Tv f cannot be optimized. The fixed transform usually is an identity
transform since the virtual image lattice is commonly defined as the fixed image lattice.
When the registration starts, the algorithm goes through each grid point of the virtual lattice in a
raster sweep. At each point the fixed and moving transforms find coordinates of the homologous
points in the fixed and moving image physical spaces, and interpolators are used to find the pixel
intensities if mapped points are in non-grid positions. These intensity values are passed to a cost
function to find the current metric value.
Note the direction of the mapping transforms here. For example, if you consider the Tvm transform,
confusion often occurs since the transform shifts a virtual lattice point on the positive X direction.
The visual effect of this mapping, once the moving image is resampled, is equivalent to manually
shifting the moving image along the negative X direction. In the same way, when the Tvm transform
applies a clock-wise rotation to the virtual space points, the visual effect of this mapping, once the
moving image has been resampled, is equivalent to manually rotating the moving image counterclock-wise. The same relationships also occur with the Tv f transform between the virtual space and
the fixed image space.
This mapping direction is chosen because the moving image is resampled on the grid of the virtual
image. In the resampling process, an algorithm iterates through every pixel of the output image and
computes the intensity assigned to this pixel by mapping to its location in the moving image.
Instead, if we were to use the transform mapping coordinates from the moving image physical space
into the virtual image physical space, then the resampling process would not guarantee that every
pixel in the grid of the virtual image would receive one and only one value. In other words, the
resampling would result in an image with holes and redundant or overlapping pixel values.
As seen in the previous examples, and as corroborated in the remaining examples in this chapter,
the transform computed by the registration framework can be used directly in the resampling filter
in order to map the moving image onto the discrete grid of the virtual image.
There are exceptional cases in which the transform desired is actually the inverse transform of the
one computed by the ITK registration framework. Only those cases may require invoking the
GetInverse() method that most transforms offer. Before attempting this, read the examples on
resampling illustrated in section 2.9 in order to familiarize yourself with the correct interpretation of
the transforms.
Now we come back to the situation illustrated in Figure 3.8. This figure shows the flexibility of
the ITKv4 registration framework. We can register two images with different scales, sizes and
resolutions. Also, we can create the output warped image with any desired size and resolution.
Nevertheless, note that the spatial transform computed during the registration process does not need
to be concerned about a different number of pixels and different pixel sizes between fixed, moving
and output images because the conversion from index space to the physical space implicitly takes
care of the required scaling factor between the involved images.
One important consequence of this fact is that having the correct image origin, image pixel size, and
3.4. Monitoring Registration
195
image direction is fundamental for the success of the registration process in ITK, since we need this
information to compute the exact location of each pixel lattice in the physical space; we must make
sure that the correct values for the origin, spacing, and direction of all fixed, moving and virtual
images are provided.
In this example, the spatial transform computed will physically map the brain from the moving
image onto the virtual space and minimize its difference with the resampled brain from the fixed
image into the virtual space. Fortunately in practice there is no need to resample the fixed image
since the virtual image physical domain is often assumed to be the same as physical domain of the
fixed image.
3.4 Monitoring Registration
The source code for this section can be found in the file
ImageRegistration3.cxx.
Given the numerous parameters involved in tuning a registration method for a particular application, it is not uncommon for a registration process to run for several minutes and still produce a
useless result. To avoid this situation it is quite helpful to track the evolution of the registration as
it progresses. The following section illustrates the mechanisms provided in ITK for monitoring the
activity of the ImageRegistrationMethodv4 class.
Insight implements the Observer/Command design pattern [20]. The classes involved in this implementation are the itk::Object, itk::Command and itk::EventObject classes. The Object is
the base class of most ITK objects. This class maintains a linked list of pointers to event observers.
The role of observers is played by the Command class. Observers register themselves with an Object, declaring that they are interested in receiving notification when a particular event happens. A
set of events is represented by the hierarchy of the Event class. Typical events are Start, End,
Progress and Iteration.
Registration is controlled by an itk::Optimizer, which generally executes an iterative process.
Most Optimizer classes invoke an itk::IterationEvent at the end of each iteration. When an
event is invoked by an object, this object goes through its list of registered observers (Commands)
and checks whether any one of them has expressed interest in the current event type. Whenever such
an observer is found, its corresponding Execute() method is invoked. In this context, Execute()
methods should be considered callbacks. As such, some of the common sense rules of callbacks
should be respected. For example, Execute() methods should not perform heavy computational
tasks. They are expected to execute rapidly, for example, printing out a message or updating a value
in a GUI.
The following code illustrates a simple way of creating a Observer/Command to monitor a registration process. This new class derives from the Command class and provides a specific implementation
of the Execute() method. First, the header file of the Command class must be included.
196
Chapter 3. Registration
#include "itkCommand.h"
Our custom command class is called CommandIterationUpdate. It derives from the Command
class and declares for convenience the types Self and Superclass. This facilitates the use of
standard macros later in the class implementation.
class CommandIterationUpdate : public itk::Command
{
public:
using Self = CommandIterationUpdate;
using Superclass = itk::Command;
The following type alias declares the type of the SmartPointer capable of holding a reference to this
object.
using Pointer = itk::SmartPointer;
The itkNewMacro takes care of defining all the necessary code for the New() method. Those
with curious minds are invited to see the details of the macro in the file itkMacro.h in the
Insight/Code/Common directory.
itkNewMacro( Self );
In order to ensure that the New() method is used to instantiate the class (and not the C++ new
operator), the constructor is declared protected.
protected:
CommandIterationUpdate() = default;
Since this Command object will be observing the optimizer, the following type alias are useful
for converting pointers when the Execute() method is invoked. Note the use of const on the
declaration of OptimizerPointer. This is relevant since, in this case, the observer is not intending
to modify the optimizer in any way. A const interface ensures that all operations invoked on the
optimizer are read-only.
using OptimizerType = itk::RegularStepGradientDescentOptimizerv4;
using OptimizerPointer = const OptimizerType *;
ITK enforces const-correctness. There is hence a distinction between the Execute() method that
can be invoked from a const object and the one that can be invoked from a non-const object. In
this particular example the non-const version simply invoke the const version. In a more elaborate
situation the implementation of both Execute() methods could be quite different. For example,
3.4. Monitoring Registration
197
you could imagine a non-const interaction in which the observer decides to stop the optimizer
in response to a divergent behavior. A similar case could happen when a user is controlling the
registration process from a GUI.
void Execute(itk::Object *caller,
const itk::EventObject & event) override
{
Execute( (const itk::Object *)caller, event);
}
Finally we get to the heart of the observer, the Execute() method. Two arguments are passed to this
method. The first argument is the pointer to the object that invoked the event. The second argument
is the event that was invoked.
void Execute(const itk::Object * object,
const itk::EventObject & event) override
{
Note that the first argument is a pointer to an Object even though the actual object invoking the event
is probably a subclass of Object. In our case we know that the actual object is an optimizer. Thus
we can perform a dynamic cast to the real type of the object.
auto optimizer = static_cast< OptimizerPointer >( object );
The next step is to verify that the event invoked is actually the one in which we are interested. This is
checked using the RTTI2 support. The CheckEvent() method allows us to compare the actual type
of two events. In this case we compare the type of the received event with an IterationEvent. The
comparison will return true if event is of type IterationEvent or derives from IterationEvent.
If we find that the event is not of the expected type then the Execute() method of this command
observer should return without any further action.
if( ! itk::IterationEvent().CheckEvent( &event ) )
{
return;
}
If the event matches the type we are looking for, we are ready to query data from the optimizer.
Here, for example, we get the current number of iterations, the current value of the cost function and
the current position on the parameter space. All of these values are printed to the standard output.
You could imagine more elaborate actions like updating a GUI or refreshing a visualization pipeline.
2 RTTI
stands for: Run-Time Type Information
198
Chapter 3. Registration
Figure 3.9: Interaction between the Command/Observer and the Registration Method.
std::cout << optimizer->GetCurrentIteration() << " = ";
std::cout << optimizer->GetValue() << " : ";
std::cout << optimizer->GetCurrentPosition() << std::endl;
This concludes our implementation of a minimal Command class capable of observing our registration method. We can now move on to configuring the registration process.
Once all the registration components are in place we can create one instance of our observer. This
is done with the standard New() method and assigned to a SmartPointer.
CommandIterationUpdate::Pointer observer = CommandIterationUpdate::New();
The newly created command is registered as observer on the optimizer, using the AddObserver()
method. Note that the event type is provided as the first argument to this method. In order for the
RTTI mechanism to work correctly, a newly created event of the desired type must be passed as
the first argument. The second argument is simply the smart pointer to the observer. Figure 3.9
illustrates the interaction between the Command/Observer class and the registration method.
optimizer->AddObserver( itk::IterationEvent(), observer );
At this point, we are ready to execute the registration. The typical call to Update() will do it. Note
again the use of the try/catch block around the Update() method in case an exception is thrown.
try
{
registration->Update();
std::cout << "Optimizer stop condition: "
3.4. Monitoring Registration
199
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
The registration process is applied to the following images in Examples/Data:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceShifted13x17y.png
It produces the following output.
0 = 4499.45 : [2.9286959512455857, 2.7244705953923805]
1 = 3860.84 : [6.135143776902402, 5.115849348610004]
2 = 3508.02 : [8.822660051952475, 8.078492808653918]
3 = 3117.31 : [10.968558473732326, 11.454158663474674]
4 = 2125.43 : [13.105290365964755, 14.835634202454191]
5 = 911.308 : [12.75173580401588, 18.819978461140323]
6 = 741.417 : [13.139053510563274, 16.857840597942413]
7 = 16.8918 : [12.356787624301035, 17.480785285045815]
8 = 233.714 : [12.79212443526829, 17.234854683011704]
9 = 39.8027 : [13.167510875734614, 16.904574468172815]
10 = 16.5731 : [12.938831371165355, 17.005597654570586]
11 = 1.68763 : [13.063495692092735, 16.996443033457986]
12 = 1.79437 : [13.001061362657559, 16.999307384689935]
13 = 0.000762481 : [12.945418587211314, 17.0277701944711]
14 = 1.74802 : [12.974454390534774, 17.01621663980765]
15 = 0.430253 : [13.002439510423766, 17.002309966416835]
16 = 0.00531816 : [12.989877586882951, 16.99301810428082]
17 = 0.0721346 : [12.996759235073881, 16.996716492365685]
18 = 0.00996773 : [13.00288423694971, 17.00156618393022]
19 = 0.00516378 : [12.99928608126834, 17.000045636412015]
20 = 0.000228075 : [13.00123653240422, 16.999943471681494]
You can verify from the code in the Execute() method that the first column is the iteration number, the second column is the metric value and the third and fourth columns are the parameters of
the transform, which is a 2D translation transform in this case. By tracking these values as the
registration progresses, you will be able to determine whether the optimizer is advancing in the
200
Chapter 3. Registration
right direction and whether the step-length is reasonable or not. That will allow you to interrupt
the registration process and fine-tune parameters without having to wait until the optimizer stops by
itself.
3.5 Multi-Modality Registration
Some of the most challenging cases of image registration arise when images of different modalities
are involved. In such cases, metrics based on direct comparison of gray levels are not applicable.
It has been extensively shown that metrics based on the evaluation of mutual information are well
suited for overcoming the difficulties of multi-modality registration.
The concept of Mutual Information is derived from Information Theory and its application to image
registration has been proposed in different forms by different groups [12, 37, 63]; a more detailed
review can be found in [23, 46]. The Insight Toolkit currently provides two different implementations of Mutual Information metrics (see section 3.11 for details). The following example illustrates
the practical use of one of these metrics.
3.5.1 Mattes Mutual Information
The source code for this section can be found in the file
ImageRegistration4.cxx.
In this example, we will solve a simple multi-modality problem using an implementation of mutual
information. This implementation was published by Mattes et. al [40].
First, we include the header files of the components used in this example.
#include
#include
#include
#include
"itkImageRegistrationMethodv4.h"
"itkTranslationTransform.h"
"itkMattesMutualInformationImageToImageMetricv4.h"
"itkRegularStepGradientDescentOptimizerv4.h"
In this example the image types and all registration components, except the metric, are declared as
in Section 3.2. The Mattes mutual information metric type is instantiated using the image types.
using MetricType = itk::MattesMutualInformationImageToImageMetricv4<
FixedImageType,
MovingImageType >;
The metric is created using the New() method and then connected to the registration object.
3.5. Multi-Modality Registration
201
MetricType::Pointer metric = MetricType::New();
registration->SetMetric( metric );
The metric requires the user to specify the number of bins used to compute the entropy. In a typical
application, 50 histogram bins are sufficient. Note however, that the number of bins may have
dramatic effects on the optimizer’s behavior.
unsigned int numberOfBins = 24;
metric->SetNumberOfHistogramBins( numberOfBins );
To calculate the image gradients, an image gradient calculator based on ImageFunction is
used instead of image gradient filters. Image gradient methods are defined in the superclass
ImageToImageMetricv4.
metric->SetUseMovingImageGradientFilter( false );
metric->SetUseFixedImageGradientFilter( false );
Notice that in the ITKv4 registration framework, optimizers always try to minimize the cost function,
and the metrics always return a parameter and derivative result that improves the optimization, so
this metric computes the negative mutual information. The optimization parameters are tuned for
this example, so they are not exactly the same as the parameters used in Section 3.2.
optimizer->SetLearningRate( 8.00 );
optimizer->SetMinimumStepLength( 0.001 );
optimizer->SetNumberOfIterations( 200 );
optimizer->ReturnBestParametersAndValueOn();
Note that large values of the learning rate will make the optimizer unstable. Small values, on the
other hand, may result in the optimizer needing too many iterations in order to walk to the extrema of
the cost function. The easy way of fine tuning this parameter is to start with small values, probably
in the range of {1.0, 5.0}. Once the other registration parameters have been tuned for producing
convergence, you may want to revisit the learning rate and start increasing its value until you observe
that the optimization becomes unstable. The ideal value for this parameter is the one that results in
a minimum number of iterations while still keeping a stable path on the parametric space of the
optimization. Keep in mind that this parameter is a multiplicative factor applied on the gradient of
the metric. Therefore, its effect on the optimizer step length is proportional to the metric values
themselves. Metrics with large values will require you to use smaller values for the learning rate in
order to maintain a similar optimizer behavior.
Whenever the regular step gradient descent optimizer encounters change in the direction of movement in the parametric space, it reduces the size of the step length. The rate at which the step length
202
Chapter 3. Registration
is reduced is controlled by a relaxation factor. The default value of the factor is 0.5. This value,
however may prove to be inadequate for noisy metrics since they tend to induce erratic movements
on the optimizers and therefore result in many directional changes. In those conditions, the optimizer will rapidly shrink the step length while it is still too far from the location of the extrema in
the cost function. In this example we set the relaxation factor to a number higher than the default in
order to prevent the premature shrinkage of the step length.
optimizer->SetRelaxationFactor( 0.8 );
Instead of using the whole virtual domain (usually fixed image domain) for the registration, we can use a spatial sampled point set by supplying an arbitrary point list over which
to evaluate the metric. The point list is expected to be in the fixed image domain, and
the points are transformed into the virtual domain internally as needed. The user can define the point set via SetFixedSampledPointSet(), and the point set is used by calling
SetUsedFixedSampledPointSet().
Also, instead of dealing with the metric directly, the user may define the sampling percentage and
sampling strategy for the registration framework at each level. In this case, the registration filter
manages the sampling operation over the fixed image space based on the input strategy (REGULAR,
RANDOM) and passes the sampled point set to the metric internally.
RegistrationType::MetricSamplingStrategyType samplingStrategy =
RegistrationType::RANDOM;
The number of spatial samples to be used depends on the content of the image. If the images are
smooth and do not contain many details, the number of spatial samples can usually be as low as
1% of the total number of pixels in the fixed image. On the other hand, if the images are detailed,
it may be necessary to use a much higher proportion, such as 20% to 50%. Increasing the number
of samples improves the smoothness of the metric, and therefore helps when this metric is used in
conjunction with optimizers that rely of the continuity of the metric values. The trade-off, of course,
is that a larger number of samples results in longer computation times per every evaluation of the
metric.
One mechanism for bringing the metric to its limit is to disable the sampling and use all the pixels
present in the FixedImageRegion. This can be done with the SetUseSampledPointSet( false )
method. You may want to try this option only while you are fine tuning all other parameters of your
registration. We don’t use this method in this current example though.
It has been demonstrated empirically that the number of samples is not a critical parameter for the
registration process. When you start fine tuning your own registration process, you should start
using high values of number of samples, for example in the range of 20% to 50% of the number of
pixels in the fixed image. Once you have succeeded to register your images you can then reduce the
number of samples progressively until you find a good compromise on the time it takes to compute
one evaluation of the metric. Note that it is not useful to have very fast evaluations of the metric
if the noise in their values results in more iterations being required by the optimizer to converge.
3.5. Multi-Modality Registration
203
You must then study the behavior of the metric values as the iterations progress, just as illustrated in
section 3.4.
double samplingPercentage = 0.20;
In ITKv4, a single virtual domain or spatial sample point set is used for the all iterations of the
registration process. The use of a single sample set results in a smooth cost function that can improve
the functionality of the optimizer.
The spatial point set is pseudo randomly generated. For reproducible results an integer seed should
set.
registration->SetMetricSamplingStrategy( samplingStrategy );
registration->SetMetricSamplingPercentage( samplingPercentage );
registration->MetricSamplingReinitializeSeed( 121213 );
Let’s execute this example over two of the images provided in Examples/Data:
• BrainT1SliceBorder20.png
• BrainProtonDensitySliceShifted13x17y.png
The second image is the result of intentionally translating the image BrainProtonDensitySliceBorder20.png by (13, 17) millimeters. Both images have unit-spacing and are shown in Figure
3.10. The registration process converges after 46 iterations and produces the following results:
Translation X = 13.0204
Translation Y = 17.0006
These values are a very close match to the true misalignment introduced in the moving image.
The result of resampling the moving image is presented on the left of Figure 3.11. The center and
right parts of the figure present a checkerboard composite of the fixed and moving images before
and after registration respectively.
Figure 3.12 (upper-left) shows the sequence of translations followed by the optimizer as it searched
the parameter space. The upper-right figure presents a closer look at the convergence basin for the
last iterations of the optimizer. The bottom of the same figure shows the sequence of metric values
computed as the optimizer searched the parameter space.
You must note however that there are a number of non-trivial issues involved in the fine tuning of
parameters for the optimization. For example, the number of bins used in the estimation of Mutual
Information has a dramatic effect on the performance of the optimizer. In order to illustrate this
effect, the same example has been executed using a range of different values for the number of
204
Chapter 3. Registration
Figure 3.10: A T1 MRI (fixed image) and a proton density MRI (moving image) are provided as input to the
registration method.
Figure 3.11: The mapped moving image (left) and the composition of fixed and moving images before (center)
and after (right) registration with Mattes mutual information.
3.5. Multi-Modality Registration
205
18
17.4
16
Y Translations (mm)
17.2
12
10
17
16.8
8
6
16.6
4
6
7
8
9
10
11
12
13
X Translations (mm)
14
15
16
12.6
17
12.8
13
X Translations (mm)
13.2
13.4
-0.3
-0.4
-0.5
Mutual Information Mattes
Y Translations (mm)
14
-0.6
-0.7
-0.8
-0.9
-1
0
5
10
15
20
25
30
35
40
45
50
Iteration No.
Figure 3.12: Sequence of translations and metric values at each iteration of the optimizer.
206
Chapter 3. Registration
20
18
16
Y Translations (mm)
14
12
10
8
6
4
2
0
-2
0
2
4
6
8
10
12
14
16
X Translations (mm)
Figure 3.13: Sensitivity of the optimization path to the number of Bins used for estimating the value of Mutual
Information with Mattes et al. approach.
bins, from 10 to 30. If you repeat this experiment, you will notice that depending on the number
of bins used, the optimizer’s path may get trapped early on in local minima. Figure 3.13 shows the
multiple paths that the optimizer took in the parametric space of the transform as a result of different
selections on the number of bins used by the Mattes Mutual Information metric. Note that many of
the paths die in local minima instead of reaching the extrema value on the upper right corner.
Effects such as the one illustrated here highlight how useless is to compare different algorithms based
on a non-exhaustive search of their parameter setting. It is quite difficult to be able to claim that a
particular selection of parameters represent the best combination for running a particular algorithm.
Therefore, when comparing the performance of two or more different algorithms, we are faced with
the challenge of proving that none of the algorithms involved in the comparison are being run with
a sub-optimal set of parameters.
The plots in Figures 3.12 and 3.13 were generated using Gnuplot3. The scripts used for this purpose
3
http://www.gnuplot.info/
3.6. Center Initialization
207
are available in the ITKSoftwareGuide Git repository under the directory
ITKSoftwareGuide/SoftwareGuide/Art.
Data for the plots were taken directly from the output that the Command/Observer in this example
prints out to the console. The output was processed with the UNIX editor sed4 in order to remove
commas and brackets that were confusing for Gnuplot’s parser. Both the shell script for running sed
and for running Gnuplot are available in the directory indicated above. You may find useful to run
them in order to verify the results presented here, and to eventually modify them for profiling your
own registrations.
Open Science is not just an abstract concept. Open Science is something to be practiced every day
with the simple gesture of sharing information with your peers, and by providing all the tools that
they need for replicating the results that you are reporting. In Open Science, the only bad results are
those that can not be replicated5 . Science is dead when people blindly trust authorities 6 instead of
verifying their statements by performing their own experiments [47, 48].
3.6
Center Initialization
The ITK image coordinate origin is typically located in one of the image corners (see the Defining Origin and Spacing section of Book 1 for details). This results in counter-intuitive transform
behavior when rotations and scaling are involved. Users tend to assume that rotations and scaling
are performed around a fixed point at the center of the image. In order to compensate for this difference in expected interpretation, the concept of center of transform has been introduced into the
toolkit. This parameter is generally a fixed parameter that is not optimized during registration, so
initialization is crucial to get efficient and accurate results. The following sections describe the main
characteristics and effects of initializing the center of a transform.
3.6.1 Rigid Registration in 2D
The source code for this section can be found in the file
ImageRegistration5.cxx.
This example illustrates the use of the itk::Euler2DTransform for performing rigid registration
in 2D. The example code is for the most part identical to that presented in Section 3.2. The main
difference is the use of the Euler2DTransform here instead of the itk::TranslationTransform.
In addition to the headers included in previous examples, the following header must also be included.
4 http://www.gnu.org/software/sed/sed.html
5 http://science.creativecommons.org/
6 For
example: Reviewers of Scientific Journals.
208
Chapter 3. Registration
#include "itkEuler2DTransform.h"
The transform type is instantiated using the code below. The only template parameter for this class
is the representation type of the space coordinates.
using TransformType = itk::Euler2DTransform< double >;
In the Hello World! example, we used Fixed/Moving initial transforms to initialize the registration
configuration. That approach was good to get an intuition of the registration method, specifically
when we aim to run a multistage registration process, from which the output of each stage can be
used to initialize the next registration stage.
To get a better underestanding of the registration process in such situations, consider an example of 3
stages registration process that is started using an initial moving transform (Γmi ). Multiple stages are
handled by linking multiple instantiations of the itk::ImageRegistrationMethodv4 class. Inside
the registration filter of the first stage, the initial moving transform is added to an internal composite
transform along with an updatable identity transform (Γu ). Although the whole composite transform
is used for metric evaluation, only the Γu is set to be updated by the optimizer at each iteration. The
Γu will be considered as the output transform of the current stage when the optimization process is
converged. This implies that the output of this stage does not include the initialization parameters,
so we need to concatenate the output and the initialization transform into a composite transform to
be considered as the final transform of the first registration stage.
T1 (x) = Γmi (Γstage1 (x))
Consider that, as explained in section 3.3, the above transform is a mapping from the vitual domain
(i.e. fixed image space, when no fixed initial transform) to the moving image space.
Then, the result transform of the first stage will be used as the initial moving transform for the second
stage of the registration process, and this approach goes on until the last stage of the registration
process.
At the end of the registration process, the Γmi and the outputs of each stage can be concatenated into
a final composite transform that is considered to be the final output of the whole registration process.
Im′ (x) = Im (Γmi (Γstage1 (Γstage2 (Γstage3 (x)))))
The above approach is especially useful if individual stages are characterized by different types of
transforms, e.g. when we run a rigid registration process that is proceeded by an affine registration
which is completed by a BSpline registration at the end.
In addition to the above method, there is also a direct initialization method in which the initial
transform will be optimized directly. In this way the initial transform will be modified during the
registration process, so it can be used as the final transform when the registration process is completed. This direct approach is conceptually close to what was happening in ITKv3 registration.
Using this method is very simple and efficient when we have only one level of registration, which is
3.6. Center Initialization
209
the case in this example. Also, a good application of this initialization method in a multi-stage scenario is when two consequent stages have the same transform types, or at least the initial parameters
can easily be inferred from the result of the previous stage, such as when a translation transform is
followed by a rigid transform.
The direct initialization approach is shown by the current example in which we try to initialize the
parameters of the optimizable transform (Γu ) directly.
For this purpose, first, the initial transform object is constructed below. This transform will be
initialized, and its initial parameters will be used when the registration process starts.
TransformType::Pointer initialTransform = TransformType::New();
In this example, the input images are taken from readers. The code below updates the readers in
order to ensure that the image parameters (size, origin and spacing) are valid when used to initialize
the transform. We intend to use the center of the fixed image as the rotation center and then use the
vector between the fixed image center and the moving image center as the initial translation to be
applied after the rotation.
fixedImageReader->Update();
movingImageReader->Update();
The center of rotation is computed using the origin, size and spacing of the fixed image.
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
const
const
const
const
SpacingType
OriginType
RegionType
SizeType
fixedSpacing
fixedOrigin
fixedRegion
fixedSize
=
=
=
=
fixedImage->GetSpacing();
fixedImage->GetOrigin();
fixedImage->GetLargestPossibleRegion();
fixedRegion.GetSize();
TransformType::InputPointType centerFixed;
centerFixed[0] = fixedOrigin[0] + fixedSpacing[0] * fixedSize[0] / 2.0;
centerFixed[1] = fixedOrigin[1] + fixedSpacing[1] * fixedSize[1] / 2.0;
The center of the moving image is computed in a similar way.
MovingImageType::Pointer movingImage = movingImageReader->GetOutput();
const
const
const
const
SpacingType
OriginType
RegionType
SizeType
movingSpacing
movingOrigin
movingRegion
movingSize
=
=
=
=
movingImage->GetSpacing();
movingImage->GetOrigin();
movingImage->GetLargestPossibleRegion();
movingRegion.GetSize();
TransformType::InputPointType centerMoving;
210
Chapter 3. Registration
centerMoving[0] = movingOrigin[0] + movingSpacing[0] * movingSize[0] / 2.0;
centerMoving[1] = movingOrigin[1] + movingSpacing[1] * movingSize[1] / 2.0;
Then, we initialize the transform by passing the center of the fixed image as the rotation center
with the SetCenter() method. Also, the translation is set as the vector relating the center of
the moving image to the center of the fixed image. This last vector is passed with the method
SetTranslation().
initialTransform->SetCenter( centerFixed );
initialTransform->SetTranslation( centerMoving - centerFixed );
Let’s finally initialize the rotation with a zero angle.
initialTransform->SetAngle( 0.0 );
Now the current parameters of the initial transform will be set to a registration method, so they can
be assigned to the Γu directly. Note that you should not confuse the following function with the
SetMoving(Fixed)InitialTransform() methods that were used in Hello World! example.
registration->SetInitialTransform( initialTransform );
Keep in mind that the scale of units in rotation and translation is quite different. For example, here
we know that the first element of the parameters array corresponds to the angle that is measured in
radians, while the other parameters correspond to the translations that are measured in millimeters,
so a naive application of gradient descent optimizer will not produce a smooth change of parameters,
because a similar change of δ to each parameter will produce a different magnitude of impact on
the transform. As the result, we need “parameter scales” to customize the learning rate for each
parameter. We can take advantage of the scaling functionality provided by the optimizers.
In this example we use small factors in the scales associated with translations. However, for the
transforms with larger parameters sets, it is not intuitive for a user to set the scales. Fortunately, a
framework for automated estimation of parameter scales is provided by ITKv4 that will be discussed
later in the example of section 3.8.
using OptimizerScalesType = OptimizerType::ScalesType;
OptimizerScalesType optimizerScales(
initialTransform->GetNumberOfParameters() );
const double translationScale = 1.0 / 1000.0;
optimizerScales[0] = 1.0;
optimizerScales[1] = translationScale;
optimizerScales[2] = translationScale;
optimizer->SetScales( optimizerScales );
3.6. Center Initialization
211
Next we set the normal parameters of the optimization method. In this case we are using an
itk::RegularStepGradientDescentOptimizerv4. Below, we define the optimization parameters like the relaxation factor, learning rate (initial step length), minimal step length and number of
iterations. These last two act as stopping criteria for the optimization.
double initialStepLength = 0.1;
optimizer->SetRelaxationFactor( 0.6 );
optimizer->SetLearningRate( initialStepLength );
optimizer->SetMinimumStepLength( 0.001 );
optimizer->SetNumberOfIterations( 200 );
Let’s execute this example over two of the images provided in Examples/Data:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceRotated10.png
The second image is the result of intentionally rotating the first image by 10 degrees around the
geometrical center of the image. Both images have unit-spacing and are shown in Figure 3.14. The
registration takes 17 iterations and produces the results:
[0.177612, 0.00681015, 0.00396471]
These results are interpreted as
• Angle = 0.177612 radians
• Translation = (0.00681015, 0.00396471) millimeters
As expected, these values match the misalignment intentionally introduced into the moving image
quite well, since 10 degrees is about 0.174532 radians.
Figure 3.15 shows from left to right the resampled moving image after registration, the difference
between the fixed and moving images before registration, and the difference between the fixed and
resampled moving image after registration. It can be seen from the last difference image that the
rotational component has been solved but that a small centering misalignment persists.
Figure 3.16 shows plots of the main output parameters produced from the registration process. This
includes the metric values at every iteration, the angle values at every iteration, and the translation
components of the transform as the registration progresses.
Let’s now consider the case in which rotations and translations are present in the initial registration,
as in the following pair of images:
212
Chapter 3. Registration
Figure 3.14: Fixed and moving images are provided as input to the registration method using the CenteredRigid2D transform.
Figure 3.15: Resampled moving image (left). Differences between the fixed and moving images, before (center) and after (right) registration using the Euler2D transform.
3.6. Center Initialization
213
2500
0.18
0.16
0.02
2000
0.14
0.01
0
-0.01
Y Translations (mm)
1500
Rotation Angle (radians)
Square Differences Metric
0.12
0.1
1000
-0.02
-0.03
-0.04
0.08
-0.05
-0.06
0.06
500
-0.07
-0.05
0
0.05
0.1
0.15
0.2
X Translations (mm)
0.04
0
0.02
0
2
4
6
8
10
12
14
16
Iteration No.
18
0
2
4
6
8
10
12
14
16
18
Iteration No.
Figure 3.16: Metric values, rotation angle and translations during registration with the Euler2D transform.
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceR10X13Y17.png
The second image is the result of intentionally rotating the first image by 10 degrees and then translating it 13mm in X and 17mm in Y . Both images have unit-spacing and are shown in Figure 3.17.
In order to accelerate convergence it is convenient to use a larger step length as shown here.
optimizer->SetMaximumStepLength( 1.3 );
The registration now takes 37 iterations and produces the following results:
[0.174582, 13.0002, 16.0007]
These parameters are interpreted as
• Angle = 0.174582 radians
• Translation = (13.0002, 16.0007) millimeters
These values approximately match the initial misalignment intentionally introduced into the moving
image, since 10 degrees is about 0.174532 radians. The horizontal translation is well resolved while
the vertical translation ends up being off by about one millimeter.
Figure 3.18 shows the output of the registration. The rightmost image of this figure shows the
difference between the fixed image and the resampled moving image after registration.
Figure 3.19 shows plots of the main output registration parameters when the rotation and translations
are combined. These results include the metric values at every iteration, the angle values at every
iteration, and the translation components of the registration as the registration converges. It can be
seen from the smoothness of these plots that a larger step length could have been supported easily
by the optimizer. You may want to modify this value in order to get a better idea of how to tune the
parameters.
214
Chapter 3. Registration
Figure 3.17: Fixed and moving images provided as input to the registration method using the CenteredRigid2D
transform.
Figure 3.18: Resampled moving image (left). Differences between the fixed and moving images, before (center) and after (right) registration with the CenteredRigid2D transform.
3.6. Center Initialization
215
4500
0.5
0.45
4000
18
0.4
3500
16
0.35
14
12
2500
2000
Y Translations (mm)
0.3
Rotation Angle (radians)
Square Differences Metric
3000
0.25
0.2
10
8
6
1500
4
0.15
2
1000
0.1
0
0
2
4
6
8
10
12
14
X Translations (mm)
500
0.05
0
0
0
5
10
15
20
25
30
35
0
5
Iteration No.
10
15
20
25
30
35
Iteration No.
Figure 3.19: Metric values, rotation angle and translations during the registration using the Euler2D transform
on an image with rotation and translation mis-registration.
3.6.2 Initializing with Image Moments
The source code for this section can be found in the file
ImageRegistration6.cxx.
This example illustrates the use of the itk::Euler2DTransform for performing registration. The
example code is for the most part identical to the one presented in Section 3.6.1. Even though
this current example is done in 2D, the class itk::CenteredTransformInitializer is quite
generic and could be used in other dimensions. The objective of the initializer class is to simplify
the computation of the center of rotation and the translation required to initialize certain transforms
such as the Euler2DTransform. The initializer accepts two images and a transform as inputs. The
images are considered to be the fixed and moving images of the registration problem, while the
transform is the one used to register the images.
The CenteredTransformInitializer supports two modes of operation. In the first mode, the centers
of the images are computed as space coordinates using the image origin, size and spacing. The
center of the fixed image is assigned as the rotational center of the transform while the vector going
from the fixed image center to the moving image center is passed as the initial translation of the
transform. In the second mode, the image centers are not computed geometrically but by using
the moments of the intensity gray levels. The center of mass of each image is computed using the
helper class itk::ImageMomentsCalculator. The center of mass of the fixed image is passed as
the rotational center of the transform while the vector going from the fixed image center of mass to
the moving image center of mass is passed as the initial translation of the transform. This second
mode of operation is quite convenient when the anatomical structures of interest are not centered
in the image. In such cases the alignment of the centers of mass provides a better rough initial
registration than the simple use of the geometrical centers. The validity of the initial registration
should be questioned when the two images are acquired in different imaging modalities. In those
cases, the center of mass of intensities in one modality does not necessarily match the center of mass
of intensities in the other imaging modality.
216
Chapter 3. Registration
The following are the most relevant headers in this example.
#include "itkEuler2DTransform.h"
#include "itkCenteredTransformInitializer.h"
The transform type is instantiated using the code below. The only template parameter of this class
is the representation type of the space coordinates.
using TransformType = itk::Euler2DTransform< double >;
Like the previous section, a direct initialization method is used here. The transform object is constructed below. This transform will be initialized, and its initial parameters will be considered as the
parameters to be used when the registration process begins.
TransformType::Pointer transform = TransformType::New();
The input images are taken from readers. It is not necessary to explicitly call Update() on the
readers since the CenteredTransformInitializer class will do it as part of its initialization. The following code instantiates the initializer. This class is templated over the fixed and moving images
type as well as the transform type. An initializer is then constructed by calling the New() method
and assigning the result to a itk::SmartPointer.
using TransformInitializerType = itk::CenteredTransformInitializer<
TransformType,
FixedImageType,
MovingImageType >;
TransformInitializerType::Pointer initializer =
TransformInitializerType::New();
The initializer is now connected to the transform and to the fixed and moving images.
initializer->SetTransform( transform );
initializer->SetFixedImage( fixedImageReader->GetOutput() );
initializer->SetMovingImage( movingImageReader->GetOutput() );
The use of the geometrical centers is selected by calling GeometryOn() while the use of center of
mass is selected by calling MomentsOn(). Below we select the center of mass mode.
initializer->MomentsOn();
Finally, the computation of the center and translation is triggered by the InitializeTransform()
method. The resulting values will be passed directly to the transform.
3.6. Center Initialization
217
initializer->InitializeTransform();
The remaining parameters of the transform are initialized as before.
transform->SetAngle( 0.0 );
Now the initialized transform object will be set to the registration method, and the starting point of
the registration is defined by its initial parameters.
If the InPlaceOn() method is called, this initialized transform will be the output transform object
or “grafted” to the output. Otherwise, this “InitialTransform” will be deep-copied or “cloned” to the
output.
registration->SetInitialTransform( transform );
registration->InPlaceOn();
Since the registration filter has InPlace set, the transform object is grafted to the output and is
updated by the registration method.
Let’s execute this example over some of the images provided in Examples/Data, for example:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceR10X13Y17.png
The second image is the result of intentionally rotating the first image by 10 degrees around the
geometric center and shifting it 13mm in X and 17mm in Y . Both images have unit-spacing and are
shown in Figure 3.14. The registration takes 21 iterations and produces:
[ 0.174527, 12.4528, 16.0766]
These parameters are interpreted as
• Angle = 0.174527 radians
• Translation = (12.4528, 16.0766) millimeters
Note that the reported translation is not the translation of (13, 17) that might be expected. The
reason is that we used the center of mass (111.204, 131.591) for the fixed center, while the input
was rotated about the geometric center (110.5, 128.5). It is more illustrative in this case to take a
look at the actual rotation matrix and offset resulting from the five parameters.
218
Chapter 3. Registration
TransformType::MatrixType matrix = transform->GetMatrix();
TransformType::OffsetType offset = transform->GetOffset();
std::cout << "Matrix = " << std::endl << matrix << std::endl;
std::cout << "Offset = " << std::endl << offset << std::endl;
Which produces the following output.
Matrix =
0.984809 -0.173642
0.173642 0.984809
Offset =
[36.9919, -1.23402]
This output illustrates how counter-intuitive the mix of center of rotation and translations can be.
Figure 3.20 will clarify this situation. The figure shows the original image on the left. A rotation of
10◦ around the center of the image is shown in the middle. The same rotation performed around the
origin of coordinates is shown on the right. It can be seen here that changing the center of rotation
introduces additional translations.
Let’s analyze what happens to the center of the image that we just registered. Under the point of view
of rotating 10◦ around the center and then applying a translation of (13mm, 17mm). The image has
a size of (221 × 257) pixels and unit spacing. Hence its center has coordinates (110.5, 128.5). Since
the rotation is done around this point, the center behaves as the fixed point of the transformation and
remains unchanged. Then with the (13mm, 17mm) translation it is mapped to (123.5, 145.5) which
becomes its final position.
The matrix and offset that we obtained at the end of the registration indicate that this should be
equivalent to a rotation of 10◦ around the origin, followed by a translation of (36.99, −1.23). Let’s
compute this in detail. First the rotation of the image center by 10◦ around the origin will move
the point (110.5, 128.5) to (86.51, 145.74). Now, applying a translation of (36.99, −1.23) maps this
point to (123.50, 144.50), which is very close to the result of our previous computation.
It is unlikely that we could have chosen these translations as the initial guess, since we tend to think
about images in a coordinate system whose origin is in the center of the image.
This underscores the importance of using good initialization for the center for a transform fixed
parameter. By using either the center of geometry or center of mass for initialization the rotation
and translation parameters may have a more intuitive interpretation than if only the optimization
parameters of translation and rotation are initialized.
Figure 3.22 shows the output of the registration. The image on the right of this figure shows the
differences between the fixed image and the resampled moving image after registration.
Figure 3.23 plots the output parameters of the registration process. It includes the metric values at
every iteration, the angle values at every iteration, and the values of the translation components as
3.6. Center Initialization
219
60
60
60
50
50
50
40
40
40
10
30
30
30
20
20
20
10
10
10
0
0
0
0
10 20
30
40 50
Original Image
0
10
20
30 40
50
Rotation around image center
10
0
10 20
30
40 50
Rotation around origin
Figure 3.20: Effect of changing the center of rotation.
Figure 3.21: Fixed and moving images provided as input to the registration method using CenteredTransformInitializer.
220
Chapter 3. Registration
Figure 3.22: Resampled moving image (left). Differences between fixed and moving images, before registration
(center) and after registration (right) with the CenteredTransformInitializer.
1600
0.18
0.16
1400
16.1
0.14
1200
16.05
0.12
800
Y Translations (mm)
16
Rotation Angle (radians)
Square Differences Metric
1000
0.1
0.08
15.95
15.9
600
0.06
15.85
400
0.04
15.8
12.3
200
12.35
12.4
12.45
12.5 12.55 12.6 12.65
X Translations (mm)
12.7
12.75
12.8
12.85
0.02
0
0
0
5
10
15
Iteration No.
20
25
0
5
10
15
20
25
Iteration No.
Figure 3.23: Plots of the Metric, rotation angle, center of rotation and translations during the registration using
CenteredTransformInitializer.
3.6. Center Initialization
221
the registration progresses. Note that this is the complementary translation as used in the transform,
not the actual total translation that is used in the transform offset. We could modify the observer to
print the total offset instead of printing the array of parameters. Let’s call that an exercise for the
reader!
3.6.3 Similarity Transform in 2D
The source code for this section can be found in the file
ImageRegistration7.cxx.
This example illustrates the use of the itk::Simularity2DTransform class for performing registration in 2D. The example code is for the most part identical to the code presented in Section 3.6.2. The main difference is the use of itk::Simularity2DTransform here rather than the
itk::Euler2DTransform class.
A similarity transform can be seen as a composition of rotations, translations and uniform (isotropic)
scaling. It preserves angles and maps lines into lines. This transform is implemented in the toolkit
as deriving from a rigid 2D transform and with a scale parameter added.
When using this transform, attention should be paid to the fact that scaling and translations are not
independent. In the same way that rotations can locally be seen as translations, scaling also results
in local displacements. Scaling is performed in general with respect to the origin of coordinates.
However, we already saw how ambiguous that could be in the case of rotations. For this reason,
this transform also allows users to setup a specific center. This center is used both for rotation and
scaling.
In addition to the headers included in previous examples, here the following header must be included.
#include "itkSimilarity2DTransform.h"
The Transform class is instantiated using the code below. The only template parameter of this class
is the representation type of the space coordinates.
using TransformType = itk::Similarity2DTransform< double >;
As before, the transform object is constructed and initialized before it is passed to the registration
filter.
TransformType::Pointer transform = TransformType::New();
In this example, we again use the helper class itk::CenteredTransformInitializer to compute
a reasonable value for the initial center of rotation and scaling along with an initial translation.
222
Chapter 3. Registration
using TransformInitializerType = itk::CenteredTransformInitializer<
TransformType,
FixedImageType,
MovingImageType >;
TransformInitializerType::Pointer initializer
= TransformInitializerType::New();
initializer->SetTransform( transform );
initializer->SetFixedImage( fixedImageReader->GetOutput() );
initializer->SetMovingImage( movingImageReader->GetOutput() );
initializer->MomentsOn();
initializer->InitializeTransform();
The remaining parameters of the transform are initialized below.
transform->SetScale( initialScale );
transform->SetAngle( initialAngle );
Now the initialized transform object will be set to the registration method, and its initial parameters
are used to initialize the registration process.
Also, by calling the InPlaceOn() method, this initialized transform will be the output transform
object or “grafted” to the output of the registration process.
registration->SetInitialTransform( transform );
registration->InPlaceOn();
Keeping in mind that the scale of units in scaling, rotation and translation are quite different, we take
advantage of the scaling functionality provided by the optimizers. We know that the first element of
the parameters array corresponds to the scale factor, the second corresponds to the angle, third and
fourth are the remaining translation. We use henceforth small factors in the scales associated with
translations.
using OptimizerScalesType = OptimizerType::ScalesType;
OptimizerScalesType optimizerScales( transform->GetNumberOfParameters() );
const double translationScale = 1.0 / 100.0;
optimizerScales[0]
optimizerScales[1]
optimizerScales[2]
optimizerScales[3]
= 10.0;
= 1.0;
= translationScale;
= translationScale;
optimizer->SetScales( optimizerScales );
3.6. Center Initialization
223
We also set the ordinary parameters of the optimization method. In this case we are using a
itk::RegularStepGradientDescentOptimizerv4. Below we define the optimization parameters, i.e. initial learning rate (step length), minimal step length and number of iterations. The last
two act as stopping criteria for the optimization.
optimizer->SetLearningRate( steplength );
optimizer->SetMinimumStepLength( 0.0001 );
optimizer->SetNumberOfIterations( 500 );
Let’s execute this example over some of the images provided in Examples/Data, for example:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceR10X13Y17S12.png
The second image is the result of intentionally rotating the first image by 10 degrees, scaling by
1/1.2 and then translating by (−13, −17). Both images have unit-spacing and are shown in Figure
3.24. The registration takes 53 iterations and produces:
[0.833237, -0.174511, -12.8065, -12.7244 ]
That are interpreted as
• Scale factor = 0.833237
• Angle = −0.174511 radians
• Translation = (−12.8065, −12.7244) millimeters
These values approximate the misalignment intentionally introduced into the moving image. Since
10 degrees is about 0.174532 radians.
Figure 3.25 shows the output of the registration. The right image shows the squared magnitude of
pixel differences between the fixed image and the resampled moving image.
Figure 3.26 shows the plots of the main output parameters of the registration process. The metric
values at every iteration are shown on the left. The rotation angle and scale factor values are shown
in the two center plots while the translation components of the registration are presented in the plot
on the right.
3.6.4 Rigid Transform in 3D
The source code for this section can be found in the file
ImageRegistration8.cxx.
224
Chapter 3. Registration
Figure 3.24: Fixed and Moving image provided as input to the registration method using the Similarity2D
transform.
Figure 3.25: Resampled moving image (left). Differences between fixed and moving images, before (center)
and after (right) registration with the Similarity2D transform.
3.6. Center Initialization
225
16000
-0.1
1.2
-0.15
14000
1.1
-12.6
-0.2
12000
-12.8
1
-0.25
-13
Y Translations (mm)
8000
0.9
-0.3
Scale Factor
Rotation Angle (radians)
Square Differences Metric
10000
-0.35
0.8
6000
-13.2
-13.4
-13.6
-0.4
0.7
-13.8
4000
-0.45
-14
-13.1
2000
-13
-12.9
-12.8
-12.7
-12.6
-12.5
-12.4
-12.3
-12.2
-12.1
X Translations (mm)
0.6
-0.5
0
-0.55
0
10
20
30
40
Iteration No.
50
60
0.5
0
10
20
30
40
50
60
0
10
20
Iteration No.
30
40
50
60
Iteration No.
Figure 3.26: Plots of the Metric, rotation angle, scale factor, and translations during the registration using
Similarity2D transform.
This example illustrates the use of the itk::VersorRigid3DTransform class for performing registration of two 3D images. The class itk::CenteredTransformInitializer is used to initialize
the center and translation of the transform. The case of rigid registration of 3D images is probably
one of the most common uses of image registration.
The following are the most relevant headers of this example.
#include "itkVersorRigid3DTransform.h"
#include "itkCenteredTransformInitializer.h"
The parameter space of the VersorRigid3DTransform is not a vector space, because addition is
not a closed operation in the space of versor components. Hence, we need to use Versor composition operation to update the first three components of the parameter array (rotation parameters), and
Vector addition for updating the last three components of the parameters array (translation parameters) [24, 27].
In the previous version of ITK, a special optimizer, itk::VersorRigid3DTransformOptimizer
was needed for registration to deal with versor computations. Fortunately in ITKv4, the
itk::RegularStepGradientDescentOptimizerv4 can be used for both vector and versor transform optimizations because, in the new registration framework, the task of updating parameters
is delegated to the moving transform itself. The UpdateTransformParameters method is implemented in the itk::Transform class as a virtual function, and all the derived transform classes
can have their own implementations of this function. Due to this fact, the updating function is reimplemented for versor transforms so it can handle versor composition of the rotation parameters.
#include "itkRegularStepGradientDescentOptimizerv4.h"
The Transform class is instantiated using the code below. The only template parameter to this class
is the representation type of the space coordinates.
226
Chapter 3. Registration
using TransformType = itk::VersorRigid3DTransform< double >;
The initial transform object is constructed below. This transform will be initialized, and its initial
parameters will be used when the registration process starts.
TransformType::Pointer initialTransform = TransformType::New();
The input images are taken from readers. It is not necessary here to explicitly call Update() on the
readers since the itk::CenteredTransformInitializer will do it as part of its computations.
The following code instantiates the type of the initializer. This class is templated over the fixed and
moving image types as well as the transform type. An initializer is then constructed by calling the
New() method and assigning the result to a smart pointer.
using TransformInitializerType = itk::CenteredTransformInitializer<
TransformType,
FixedImageType,
MovingImageType >;
TransformInitializerType::Pointer initializer =
TransformInitializerType::New();
The initializer is now connected to the transform and to the fixed and moving images.
initializer->SetTransform( initialTransform );
initializer->SetFixedImage( fixedImageReader->GetOutput() );
initializer->SetMovingImage( movingImageReader->GetOutput() );
The use of the geometrical centers is selected by calling GeometryOn() while the use of center of
mass is selected by calling MomentsOn(). Below we select the center of mass mode.
initializer->MomentsOn();
Finally, the computation of the center and translation is triggered by the InitializeTransform()
method. The resulting values will be passed directly to the transform.
initializer->InitializeTransform();
The rotation part of the transform is initialized using a itk::Versor which is simply a unit quaternion. The VersorType can be obtained from the transform traits. The versor itself defines the type
of the vector used to indicate the rotation axis. This trait can be extracted as VectorType. The
following lines create a versor object and initialize its parameters by passing a rotation axis and an
angle.
3.6. Center Initialization
227
using VersorType = TransformType::VersorType;
using VectorType = VersorType::VectorType;
VersorType
rotation;
VectorType
axis;
axis[0] = 0.0;
axis[1] = 0.0;
axis[2] = 1.0;
constexpr double angle = 0;
rotation.Set( axis, angle );
initialTransform->SetRotation( rotation );
Now the current initialized transform will be set to the registration method, so its initial parameters
can be used to initialize the registration process.
registration->SetInitialTransform( initialTransform );
Let’s execute this example over some of the images available in the following website
http://public.kitware.com/pub/itk/Data/BrainWeb.
Note that the images in this website are compressed in .tgz files. You should download these files
and decompress them in your local system. After decompressing and extracting the files you could
take a pair of volumes, for example the pair:
• brainweb1e1a10f20.mha
• brainweb1e1a10f20Rot10Tx15.mha
The second image is the result of intentionally rotating the first image by 10 degrees around the
origin and shifting it 15mm in X.
Also, instead of doing the above steps manually, you can turn on the following flag in your build
environment:
ITK USE BRAINWEB DATA
Then, the above data will be loaded to your local ITK build directory.
The registration takes 21 iterations and produces:
[7.2295e-05, -7.20626e-05, -0.0872168, 2.64765, -17.4626, -0.00147153]
That are interpreted as
• Versor = (7.2295e − 05, −7.20626e − 05, −0.0872168)
• Translation = (2.64765, −17.4626, −0.00147153) millimeters
228
Chapter 3. Registration
This Versor is equivalent to a rotation of 9.98 degrees around the Z axis.
Note that the reported translation is not the translation of (15.0, 0.0, 0.0) that we may be naively expecting. The reason is that the VersorRigid3DTransform is applying the rotation around the center found by the CenteredTransformInitializer and then adding the translation vector shown
above.
It is more illustrative in this case to take a look at the actual rotation matrix and offset resulting from
the 6 parameters.
TransformType::MatrixType matrix = finalTransform->GetMatrix();
TransformType::OffsetType offset = finalTransform->GetOffset();
std::cout << "Matrix = " << std::endl << matrix << std::endl;
std::cout << "Offset = " << std::endl << offset << std::endl;
The output of this print statements is
Matrix =
0.984786 0.173769 -0.000156187
-0.173769 0.984786 -0.000131469
0.000130965 0.000156609 1
Offset =
[-15, 0.0189186, -0.0305439]
From the rotation matrix it is possible to deduce that the rotation is happening in the X,Y plane and
that the angle is on the order of arcsin (0.173769) which is very close to 10 degrees, as we expected.
Figure 3.28 shows the output of the registration. The center image in this figure shows the differences
between the fixed image and the resampled moving image before the registration. The image on the
right side presents the difference between the fixed image and the resampled moving image after
the registration has been performed. Note that these images are individual slices extracted from the
actual volumes. For details, look at the source code of this example, where the ExtractImageFilter
is used to extract a slice from the the center of each one of the volumes. One of the main purposes
of this example is to illustrate that the toolkit can perform registration on images of any dimension.
The only limitations are, as usual, the amount of memory available for the images and the amount
of computation time that it will take to complete the optimization process.
Figure 3.29 shows the plots of the main output parameters of the registration process. The Z component of the versor is plotted as an indication of how the rotation progresses. The X,Y translation
components of the registration are plotted at every iteration too.
Shell and Gnuplot scripts for generating the diagrams in Figure 3.29 are available in the
ITKSoftwareGuide Git repository under the directory
ITKSoftwareGuide/SoftwareGuide/Art.
3.6. Center Initialization
229
Figure 3.27: Fixed and moving image provided as input to the registration method using CenteredTransformInitializer.
Figure 3.28: Resampled moving image (left). Differences between fixed and moving images, before (center)
and after (right) registration with the CenteredTransformInitializer.
230
Chapter 3. Registration
2500
-0.01
-15.8
-16
-0.02
2000
-16.2
1500
1000
Y Translations (mm)
Z Versor Component
Square Differences Metric
-0.03
-0.04
-0.05
-0.06
-16.4
-16.6
-16.8
-17
-0.07
-17.2
500
-0.08
0
-17.4
-0.09
0
5
10
15
Iteration No.
20
-17.6
0
5
10
15
Iteration No.
20
2.3 2.35 2.4 2.45 2.5 2.55 2.6 2.65 2.7
X Translations (mm)
Figure 3.29: Plots of the metric, rotation angle, center of rotation and translations during the registration using
CenteredTransformInitializer.
You are strongly encouraged to run the example code, since only in this way can you gain first-hand
experience with the behavior of the registration process. Once again, this is a simple reflection of
the philosophy that we put forward in this book:
If you can not replicate it, then it does not exist!
We have seen enough published papers with pretty pictures, presenting results that in practice are
impossible to replicate. That is vanity, not science.
3.6.5 Centered Initialized Affine Transform
The source code for this section can be found in the file
ImageRegistration9.cxx.
This example illustrates the use of the itk::AffineTransform for performing registration in 2D.
The example code is, for the most part, identical to that in 3.6.2. The main difference is the use of the
AffineTransform here instead of the itk::Euler2DTransform. We will focus on the most relevant
changes in the current code and skip the basic elements already explained in previous examples.
Let’s start by including the header file of the AffineTransform.
#include "itkAffineTransform.h"
We then define the types of the images to be registered.
constexpr unsigned int Dimension = 2;
using PixelType = float;
using FixedImageType = itk::Image< PixelType, Dimension >;
using MovingImageType = itk::Image< PixelType, Dimension >;
3.6. Center Initialization
231
The transform type is instantiated using the code below. The template parameters of this class are
the representation type of the space coordinates and the space dimension.
using TransformType = itk::AffineTransform< double, Dimension >;
The transform object is constructed below and is initialized before the registration process starts.
TransformType::Pointer transform = TransformType::New();
In this example, we again use the itk::CenteredTransformInitializer helper class in order to
compute reasonable values for the initial center of rotation and the translations. The initializer is set
to use the center of mass of each image as the initial correspondence correction.
using TransformInitializerType = itk::CenteredTransformInitializer<
TransformType,
FixedImageType,
MovingImageType >;
TransformInitializerType::Pointer initializer
= TransformInitializerType::New();
initializer->SetTransform( transform );
initializer->SetFixedImage( fixedImageReader->GetOutput() );
initializer->SetMovingImage( movingImageReader->GetOutput() );
initializer->MomentsOn();
initializer->InitializeTransform();
Now we pass the transform object to the registration filter, and it will be grafted to the output transform of the registration filter by updating its parameters during the the registration process.
registration->SetInitialTransform( transform );
registration->InPlaceOn();
Keeping in mind that the scale of units in scaling, rotation and translation are quite different, we
take advantage of the scaling functionality provided by the optimizers. We know that the first N × N
elements of the parameters array correspond to the rotation matrix factor, and the last N are the
components of the translation to be applied after multiplication with the matrix is performed.
using OptimizerScalesType = OptimizerType::ScalesType;
OptimizerScalesType optimizerScales( transform->GetNumberOfParameters() );
optimizerScales[0]
optimizerScales[1]
optimizerScales[2]
optimizerScales[3]
optimizerScales[4]
optimizerScales[5]
=
=
=
=
=
=
1.0;
1.0;
1.0;
1.0;
translationScale;
translationScale;
optimizer->SetScales( optimizerScales );
232
Chapter 3. Registration
We also set the usual parameters of the optimization method. In this case we are using an
itk::RegularStepGradientDescentOptimizerv4 as before. Below, we define the optimization
parameters like learning rate (initial step length), minimum step length and number of iterations.
These last two act as stopping criteria for the optimization.
optimizer->SetLearningRate( steplength );
optimizer->SetMinimumStepLength( 0.0001 );
optimizer->SetNumberOfIterations( maxNumberOfIterations );
Finally we trigger the execution of the registration method by calling the Update() method. The
call is placed in a try/catch block in the case any exceptions are thrown.
try
{
registration->Update();
std::cout << "Optimizer stop condition: "
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
Once the optimization converges, we recover the parameters from the registration method. We can
also recover the final value of the metric with the GetValue() method and the final number of
iterations with the GetCurrentIteration() method.
const TransformType::ParametersType finalParameters =
registration->GetOutput()->Get()->GetParameters();
const
const
const
const
double
double
double
double
finalRotationCenterX
finalRotationCenterY
finalTranslationX
finalTranslationY
=
=
=
=
transform->GetCenter()[0];
transform->GetCenter()[1];
finalParameters[4];
finalParameters[5];
const unsigned int numberOfIterations = optimizer->GetCurrentIteration();
const double bestValue = optimizer->GetValue();
Let’s execute this example over two of the images provided in Examples/Data:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceR10X13Y17.png
3.6. Center Initialization
233
Figure 3.30: Fixed and moving images provided as input to the registration method using the AffineTransform.
The second image is the result of intentionally rotating the first image by 10 degrees and then translating by (−13, −17). Both images have unit-spacing and are shown in Figure 3.30. We execute
the code using the following parameters: step length=1.0, translation scale= 0.0001 and maximum
number of iterations = 300. With these images and parameters the registration takes 92 iterations
and produces
90
44.0851 [0.9849, -0.1729, 0.1725, 0.9848, 12.4541, 16.0759] AffineAngle: 9.9494
These results are interpreted as
• Iterations = 92
• Final Metric = 44.0386
• Center = (111.204, 131.591) millimeters
• Translation = (12.4542, 16.076) millimeters
• Affine scales = (1.00014, .999732)
The second component of the matrix values is usually associated with sin θ. We obtain the rotation through SVD of the affine matrix. The value is 9.9494 degrees, which is approximately the
intentional misalignment of 10.0 degrees.
234
Chapter 3. Registration
Figure 3.31: The resampled moving image (left), and the difference between the fixed and moving images
before (center) and after (right) registration with the AffineTransform transform.
5000
10
4500
9
8
16.4
3500
16.2
3000
2500
2000
Y Translations (mm)
7
Angle (degrees)
Square Differences Metric
4000
6
5
4
15.6
15.4
15.2
1500
15
3
1000
16
15.8
12
2
500
0
12.5
13 13.5 14 14.5
X Translations (mm)
15
1
0 10 20 30 40 50 60 70 80 90
Iteration No.
0
10 20 30 40 50 60 70 80 90
Iteration No.
Figure 3.32: Metric values, rotation angle and translations during the registration using the AffineTransform
transform.
3.7. Multi-Resolution Registration
235
Figure 3.31 shows the output of the registration. The right most image of this figure shows the
squared magnitude difference between the fixed image and the resampled moving image.
Figure 3.32 shows the plots of the main output parameters of the registration process. The metric
values at every iteration are shown on the left plot. The angle values are shown on the middle plot,
while the translation components of the registration are presented on the right plot. Note that the
final total offset of the transform is to be computed as a combination of the shift due to rotation plus
the explicit translation set on the transform.
3.7 Multi-Resolution Registration
Performing image registration using a multi-resolution approach is widely used to improve speed,
accuracy and robustness. The basic idea is that registration is first performed at a coarse scale where
the images have fewer pixels. The spatial mapping determined at the coarse level is then used
to initialize registration at the next finer scale. This process is repeated until it reaches the finest
possible scale. This coarse-to-fine strategy greatly improves the registration success rate and also
increases robustness by eliminating local optima at coarser scales. Robustness can be improved even
more by smoothing the images at coarse scales.
In all previous examples we ran the registration process at a single resolution. However, the ITKv4
registration framework is structured to provide a multi-resolution registration method. For this purpose we only need to define the number of levels as well as the resolution and smoothness of the
input images at each level. The registration filter smoothes and subsamples the images according to
user-defined ShrinkFactor and SmoothingSigma vectors.
We now present the multi-resolution capabilities of the framework by way of an example.
3.7.1 Fundamentals
In ITK, the itk::MultiResolutionPyramidImageFilter can be used to create a sequence of
reduced resolution images from the input image. The down-sampling is performed according to a
user defined multi-resolution schedule. The schedule is specified as an itk::Array2D of integers,
containing shrink factors for each multi-resolution level (rows) for each dimension (columns). For
example,
8 4 4
4 4 2
is a schedule for a three dimensional image for two multi-resolution levels. In the first (coarsest)
level, the image is reduced by a factor of 8 in the column dimension, factor of 4 in the row dimension
and a factor of 4 in the slice dimension. In the second level, the image reduced by a factor of 4 in
the column dimension, 4 in the row dimension and 2 in the slice dimension.
236
Chapter 3. Registration
The method SetNumberOfLevels() is used to set the number of resolution levels in the pyramid.
This method will allocate memory for the schedule and generate a default table with the starting
(coarsest) shrink factors for all dimensions set to 2( M − 1), where M is the number of levels. All
factors are halved for all subsequent levels. For example, if we set the number of levels to 4, the
default schedule is then:
8
4
2
1
8
4
2
1
8
4
2
1
The user can get a copy of the schedule using method GetSchedule(), make modifications, and
reset it using method SetSchedule(). Alternatively, a user can create a default table by specifying
the starting (coarsest) shrink factors using the method SetStartingShrinkFactors(). The factors
for the subsequent levels are generated by halving the factor or setting it to one, depending on which
is larger. For example, for a 4 level pyramid and starting factors of 8, 8 and 4, the generated schedule
would be:
8
4
2
1
8
4
2
1
4
2
1
1
When this filter is triggered by Update(), M outputs are produced where the m-th output corresponds to the m-th level of the pyramid. To generate these images, Gaussian smoothing is first performed using a itk::DiscreteGaussianImageFilter with the variance set to (s/2)2 , where s is
the shrink factor. The smoothed images are then sub-sampled using a itk::ShrinkImageFilter.
3.7.2 Fundamentals
The source code for this section can be found in the file
MultiResImageRegistration1.cxx.
This example illustrates the use of the itk::ImageRegistrationMethodv4 to solve a simple
multi-modality registration problem by a multi-resolution approach. Since ITKv4 registration
method is designed based on a multi-resolution structure, a separate set of classes are no longer
required to run the registration process of this example.
This a great advantage over the previous versions of ITK, as in ITKv3 we had to use a different filter (
itk::MultiResolutionImageRegistrationMethod) to run a multi-resolution process. Also, we
had to use image pyramids filters ( itk::MultiResolutionPyramidImageFilter) for creating the
sequence of downsampled images. Hence, you can see how ITKv4 framework is more user-friendly
in more complex situations.
3.7. Multi-Resolution Registration
237
Registration Level 0
Transform
Registration Level 1
Transform
Registration Level 2
Transform
Registration Level 3
Transform
Registration Level 4
Moving Image
Pyramid
Fixed Image
Pyramid
Transform
Figure 3.33: Conceptual representation of the multi-resolution registration process.
To begin the example, we include the headers of the registration components we will use.
#include
#include
#include
#include
"itkImageRegistrationMethodv4.h"
"itkTranslationTransform.h"
"itkMattesMutualInformationImageToImageMetricv4.h"
"itkRegularStepGradientDescentOptimizerv4.h"
The ImageRegistrationMethodv4 solves a registration problem in a coarse-to-fine manner as illustrated in Figure 3.33. The registration is first performed at the coarsest level using the images at the
first level of the fixed and moving image pyramids. The transform parameters determined by the
registration are then used to initialize the registration at the next finer level using images from the
second level of the pyramids. This process is repeated as we work up to the finest level of image
resolution.
In a typical registration scenario, a user will tweak component settings or even swap out components
between multi-resolution levels. For example, when optimizing at a coarse resolution, it may be
possible to take more aggressive step sizes and have a more relaxed convergence criterion.
Tweaking the components between resolution levels can be done using ITK’s implementation of
the Command/Observer design pattern. Before beginning registration at each resolution level,
where ImageRegistrationMethodv4 invokes a MultiResolutionIterationEvent(). The registration components can be changed by implementing a itk::Command which responds to the event.
A brief description of the interaction between events and commands was previously presented in
Section 3.4.
We will illustrate this mechanism by changing the parameters of the optimizer between each resolution level by way of a simple interface command. First, we include the header file of the Command
238
Chapter 3. Registration
class.
#include "itkCommand.h"
Our new interface command class is called RegistrationInterfaceCommand. It derives from
Command and is templated over the multi-resolution registration type.
template
class RegistrationInterfaceCommand : public itk::Command
{
We then define Self, Superclass, Pointer, New() and a constructor in a similar fashion to the
CommandIterationUpdate class in Section 3.4.
public:
using Self = RegistrationInterfaceCommand;
using Superclass = itk::Command;
using Pointer = itk::SmartPointer;
itkNewMacro( Self );
protected:
RegistrationInterfaceCommand() = default;
For convenience, we declare types useful for converting pointers in the Execute() method.
public:
using
using
using
using
RegistrationType = TRegistration;
RegistrationPointer = RegistrationType *;
OptimizerType = itk::RegularStepGradientDescentOptimizerv4;
OptimizerPointer = OptimizerType *;
Two arguments are passed to the Execute() method: the first is the pointer to the object which
invoked the event and the second is the event that was invoked.
void Execute( itk::Object * object,
const itk::EventObject & event) override
{
First
we
verify
that
the
event
invoked
is
of
the
right
itk::MultiResolutionIterationEvent(). If not, we return without any further action.
if( !(itk::MultiResolutionIterationEvent().CheckEvent( &event ) ) )
{
return;
}
type,
3.7. Multi-Resolution Registration
239
We then convert the input object pointer to a RegistrationPointer. Note that no error checking is done
here to verify the dynamic cast was successful since we know the actual object is a registration
method. Then we ask for the optimizer object from the registration method.
auto registration = static_cast( object );
auto optimizer = static_cast< OptimizerPointer >(
registration->GetModifiableOptimizer() );
If this is the first resolution level we set the learning rate (representing the first step size) and the
minimum step length (representing the convergence criterion) to large values. At each subsequent
resolution level, we will reduce the minimum step length by a factor of 5 in order to allow the
optimizer to focus on progressively smaller regions. The learning rate is set up to the current step
length. In this way, when the optimizer is reinitialized at the beginning of the registration process
for the next level, the step length will simply start with the last value used for the previous level.
This will guarantee the continuity of the path taken by the optimizer through the parameter space.
if ( registration->GetCurrentLevel() == 0 )
{
optimizer->SetLearningRate( 16.00 );
optimizer->SetMinimumStepLength( 2.5 );
}
else
{
optimizer->SetLearningRate( optimizer->GetCurrentStepLength() );
optimizer->SetMinimumStepLength(
optimizer->GetMinimumStepLength() * 0.2 );
}
Another version of the Execute() method accepting a const input object is also required since this
method is defined as pure virtual in the base class. This version simply returns without taking any
action.
void Execute(const itk::Object * , const itk::EventObject & ) override
{
return;
}
};
The fixed and moving image types are defined as in previous examples. The downsampled images
for different resolution levels are created internally by the registration method based on the values
provided for ShrinkFactor and SmoothingSigma vectors.
The types for the registration components are then derived using the fixed and moving image type,
as in previous examples.
To set the optimizer parameters, note that LearningRate and MinimumStepLength are set in the
obsever at the begining of each resolution level. The other optimizer parameters are set as follows.
240
Chapter 3. Registration
optimizer->SetNumberOfIterations( 200 );
optimizer->SetRelaxationFactor( 0.5 );
We set the number of multi-resolution levels to three and set the corresponding shrink factor and
smoothing sigma values for each resolution level. Using smoothing in the subsampled images in
low-resolution levels can avoid large fluctuations in the metric function, which prevents the optimizer from becoming trapped in local minima. In this simple example we have no smoothing, and
we have used small shrinkings for the first two resolution levels.
constexpr unsigned int numberOfLevels = 3;
RegistrationType::ShrinkFactorsArrayType shrinkFactorsPerLevel;
shrinkFactorsPerLevel.SetSize( 3 );
shrinkFactorsPerLevel[0] = 3;
shrinkFactorsPerLevel[1] = 2;
shrinkFactorsPerLevel[2] = 1;
RegistrationType::SmoothingSigmasArrayType smoothingSigmasPerLevel;
smoothingSigmasPerLevel.SetSize( 3 );
smoothingSigmasPerLevel[0] = 0;
smoothingSigmasPerLevel[1] = 0;
smoothingSigmasPerLevel[2] = 0;
registration->SetNumberOfLevels ( numberOfLevels );
registration->SetShrinkFactorsPerLevel( shrinkFactorsPerLevel );
registration->SetSmoothingSigmasPerLevel( smoothingSigmasPerLevel );
Once all the registration components are in place we can create an instance of our interface command
and connect it to the registration object using the AddObserver() method.
using CommandType = RegistrationInterfaceCommand;
CommandType::Pointer command = CommandType::New();
registration->AddObserver( itk::MultiResolutionIterationEvent(), command );
Then we trigger the registration process by calling Update().
Let’s execute this example using the following images
• BrainT1SliceBorder20.png
• BrainProtonDensitySliceShifted13x17y.png
The output produced by the execution of the method is
0
-0.316956
[11.4200, 11.2063]
3.7. Multi-Resolution Registration
241
Figure 3.34: Mapped moving image (left) and composition of fixed and moving images before (center) and
after (right) registration.
1
2
3
4
-0.562048
-0.407696
-0.5702
-0.803252
[18.2938,
[11.3643,
[13.7244,
[11.1634,
25.6545]
21.6569]
18.4274]
15.3547]
0
1
2
3
4
5
-0.697586
-0.901984
-0.827423
-0.92754
-0.902671
-0.941212
[12.8778,
[13.1794,
[13.0545,
[12.8528,
[12.9426,
[13.1402,
16.3846]
18.3617]
17.3695]
16.3901]
16.8819]
17.3413]
0
1
2
-0.922239
-0.930203
-0.930959
[13.0364, 17.1138]
[12.9463, 16.8806]
[13.0191, 16.9822]
Result =
Translation X
Translation Y
Iterations
Metric value
=
=
=
=
13.0192
16.9823
4
-0.929237
These values are a close match to the true misalignment of (13, 17) introduced in the moving image.
The result of resampling the moving image is presented in the left image of Figure 3.34. The center
and right images of the figure depict a checkerboard composite of the fixed and moving images
242
Chapter 3. Registration
-0.3
-0.4
26
-0.5
22
20
-0.6
Metric
Y Translations (mm)
24
18
-0.7
16
14
-0.8
12
10
11
12
13 14 15 16 17
X Translations (mm)
18
19
-0.9
-1
0
2
4
6
8
10
Iteration No.
12
14
Figure 3.35: Sequence of translations and metric values at each iteration of the optimizer.
before and after registration.
Figure 3.35 (left) shows the sequence of translations followed by the optimizer as it searched the
parameter space. The right side of the same figure shows the sequence of metric values computed
as the optimizer searched the parameter space. From the trace, we can see that with the more
aggressive optimization parameters we get quite close to the optimal value within 5 iterations with
the remaining iterations just doing fine adjustments. It is interesting to compare these results with
those of the single resolution example in Section 3.5.1, where 46 iterations were required as more
conservative optimization parameters had to be used.
3.8 Multi-Stage Registration
In section 3.7 you noticed how to tweak component settings between multi-resolution levels and saw
how it can benefit the registration process. That is, the matching metric gets close to the optimal
value before final parameter adjustments in full resolution. This approach saves large amounts of
time in most practical cases, since fewer iterations are required at the full resolution level. This is
helpful in cases like a deformable registration process on a large dataset, e.g. a high-resolution 3D
image.
Another possible scheme is to apply a simple rigid transform for the initial coarse registration, then
upgrade to an affine transform at the finer level. Finally, proceed to a deformable transform at the
last level when we are close enough to the optimal value.
Fortunately, itk::ImageRegistrationMethodv4 allows for multistage registration whereby each
stage is characterized by possibly different transforms and different image metrics. As in the above
3.8. Multi-Stage Registration
243
situation, you may want to perform a linear registration followed by a deformable registration with
both stages performed across multiple resolutions.
Multiple stages are handled by linking multiple instantiations of this class. An optional composite
transform can be used as a container to concatenate the output transforms of multiple stages.
We now present the multistage capabilities of the framework by way of an example.
3.8.1 Fundamentals
The source code for this section can be found in the file
MultiStageImageRegistration1.cxx.
This example illustrates the use of more complex components of the registration framework. In
particular, it introduces a multistage, multi-resolution approach to run a multi-modal registration
process using two linear itk::TranslationTransform and itk::AffineTransform. Also, it
shows the use of Scale Estimators for fine-tuning the scale parameters of the optimizer when an
Affine transform is used. The itk::RegistrationParameterScalesFromPhysicalShift filter
is used for automatic estimation of the parameters scales.
To begin the example, we include the headers of the registration components we will use.
#include "itkImageRegistrationMethodv4.h"
#include "itkMattesMutualInformationImageToImageMetricv4.h"
#include "itkRegularStepGradientDescentOptimizerv4.h"
#include "itkConjugateGradientLineSearchOptimizerv4.h"
#include "itkTranslationTransform.h"
#include "itkAffineTransform.h"
#include "itkCompositeTransform.h"
In a multistage scenario, each stage needs an individual instantiation of the
itk::ImageRegistrationMethodv4, so each stage can possibly have a different transform,
a different optimizer, and a different image metric and can be performed in multiple levels. The
configuration of the registration method at each stage closely follows the procedure in the previous
section.
In early stages we can use simpler transforms and more aggressive optimization parameters to take
big steps toward the optimal value. Then, at the final stage we can have a more complex transform
to do fine adjustments of the final parameters.
A possible scheme is to use a simple translation transform for initial coarse registration levels and
upgrade to an affine transform at the finer level. Since we have two different types of transforms, we
can use a multistage registration approach as shown in the current example.
First we need to configure the registration components of the initial stage. The instantiation of the
244
Chapter 3. Registration
transform type requires only the dimension of the space and the type used for representing space
coordinates.
using TTransformType = itk::TranslationTransform< double, Dimension >;
The types of other registration components are defined here.
itk::RegularStepGradientDescentOptimizerv4 is used as the optimizer of the first stage.
Also, we use itk::MattesMutualInformationImageToImageMetricv4 as the metric since it
is fitted for a multi-modal registration.
using TOptimizerType = itk::RegularStepGradientDescentOptimizerv4< double >;
using MetricType = itk::MattesMutualInformationImageToImageMetricv4<
FixedImageType,
MovingImageType >;
using TRegistrationType = itk::ImageRegistrationMethodv4<
FixedImageType,
MovingImageType,
TTransformType >;
Then, all the components are instantiated using their New() method and connected to the registration
object as in previous examples.
The output transform of the registration process will be constructed internally in the registration filter
since the related TransformType is already passed to the registration method as a template parameter.
However, we should provide an initial moving transform for the registration method if needed.
TTransformType::Pointer
movingInitTx
= TTransformType::New();
After setting the initial parameters, the initial transform can be passed to the registration filter by
SetMovingInitialTransform() method.
transRegistration->SetMovingInitialTransform( movingInitTx );
We can use a itk::CompositeTransform to stack all the output transforms resulted from multiple
stages. This composite transform should also hold the moving initial transform (if it exists) because
as explained in section 3.6.1, the output of each registration stage does not include the input initial
transform to that stage.
using CompositeTransformType = itk::CompositeTransform< double,
Dimension >;
CompositeTransformType::Pointer compositeTransform =
CompositeTransformType::New();
compositeTransform->AddTransform( movingInitTx );
3.8. Multi-Stage Registration
245
In the case of this simple example, the first stage is run only in one level of registration at a coarse
resolution.
constexpr unsigned int numberOfLevels1 = 1;
TRegistrationType::ShrinkFactorsArrayType shrinkFactorsPerLevel1;
shrinkFactorsPerLevel1.SetSize( numberOfLevels1 );
shrinkFactorsPerLevel1[0] = 3;
TRegistrationType::SmoothingSigmasArrayType smoothingSigmasPerLevel1;
smoothingSigmasPerLevel1.SetSize( numberOfLevels1 );
smoothingSigmasPerLevel1[0] = 2;
transRegistration->SetNumberOfLevels ( numberOfLevels1 );
transRegistration->SetShrinkFactorsPerLevel( shrinkFactorsPerLevel1 );
transRegistration->SetSmoothingSigmasPerLevel( smoothingSigmasPerLevel1 );
Also, for this initial stage we can use a more agressive parameter set for the optimizer by taking a
big step size and relaxing stop criteria.
transOptimizer->SetLearningRate( 16 );
transOptimizer->SetMinimumStepLength( 1.5 );
Once all the registration components are in place, we trigger the registration process by calling
Update() and add the result output transform to the final composite transform, so this composite
transform can be used to initialize the next registration stage.
try
{
transRegistration->Update();
std::cout << "Optimizer stop condition: "
<< transRegistration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
compositeTransform->AddTransform(
transRegistration->GetModifiableTransform() );
Now we can upgrade to an Affine transform as the second stage of registration process. The AffineTransform is a linear transformation that maps lines into lines. It can be used to represent translations, rotations, anisotropic scaling, shearing or any combination of them. Details about the affine
transform can be seen in Section 3.9.16. The instantiation of the transform type requires only the
dimension of the space and the type used for representing space coordinates.
246
Chapter 3. Registration
using ATransformType = itk::AffineTransform< double, Dimension >;
We also use a different optimizer in configuration of the second stage while the metric is kept the
same as before.
using AOptimizerType =
itk::ConjugateGradientLineSearchOptimizerv4Template;
using ARegistrationType = itk::ImageRegistrationMethodv4<
FixedImageType,
MovingImageType,
ATransformType >;
Again all the components are instantiated using their New() method and connected to the registration
object like in previous stages.
The current stage can be initialized using the initial transform of the registration and the result
transform of the previous stage, so that both are concatenated into the composite transform.
affineRegistration->SetMovingInitialTransform( compositeTransform );
In Section 3.6.2 we showed the importance of center of rotation in the registration process. In Affine
transforms, the center of rotation is defined by the fixed parameters set, which are set by default to
[0, 0]. However, consider a situation where the origin of the virtual space, in which the registration
is run, is far away from the zero origin. In such cases, leaving the center of rotation as the default
value can make the optimization process unstable. Therefore, we are always interested to set the
center of rotation to the center of virtual space which is usually the fixed image space.
Note that either center of gravity or geometrical center can be used as the center of rotation. In this
example center of rotation is set to the geometrical center of the fixed image. We could also use
itk::ImageMomentsCalculator filter to compute the center of mass.
Based on the above discussion, the user must set the fixed parameters of the registration transform
outside of the registraton method, so first we instantiate an object of the output transform type.
ATransformType::Pointer affineTx = ATransformType::New();
Then, we compute the physical center of the fixed image and set that as the center of the output
Affine transform.
using
using
using
using
SpacingType = FixedImageType::SpacingType;
OriginType = FixedImageType::PointType;
RegionType = FixedImageType::RegionType;
SizeType = FixedImageType::SizeType;
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
3.8. Multi-Stage Registration
const
const
const
const
SpacingType
OriginType
RegionType
SizeType
fixedSpacing
fixedOrigin
fixedRegion
fixedSize
247
=
=
=
=
fixedImage->GetSpacing();
fixedImage->GetOrigin();
fixedImage->GetLargestPossibleRegion();
fixedRegion.GetSize();
ATransformType::InputPointType centerFixed;
centerFixed[0] =
fixedOrigin[0] + fixedSpacing[0] * fixedSize[0] / 2.0;
centerFixed[1] =
fixedOrigin[1] + fixedSpacing[1] * fixedSize[1] / 2.0;
const unsigned int numberOfFixedParameters =
affineTx->GetFixedParameters().Size();
ATransformType::ParametersType fixedParameters( numberOfFixedParameters );
for (unsigned int i = 0; i < numberOfFixedParameters; ++i)
{
fixedParameters[i] = centerFixed[i];
}
affineTx->SetFixedParameters( fixedParameters );
Then, the initialized output transform should be connected to the registration object by using
SetInitialTransform() method.
It
is
important
to
distinguish
between
the
SetInitialTransform()
and
SetMovingInitialTransform() that was used to initialize the registration stage based on
the results of the previous stages. You can assume that the first one is used for direct manipulation
of the optimizable transform in current registration process.
affineRegistration->SetInitialTransform( affineTx
);
The set of optimizable parameters in the Affine transform have different dynamic ranges. Typically the parameters associated with the matrix have values around [−1 : 1], although they are not
restricted to this interval. Parameters associated with translations, on the other hand, tend to have
much higher values, typically on the order of 10.0 to 100.0. This difference in dynamic range negatively affects the performance of gradient descent optimizers. ITK provides some mechanisms
to compensate for such differences in values among the parameters when they are passed to the
optimizer.
The first mechanism consists of providing an array of scale factors to the optimizer. These factors
re-normalize the gradient components before they are used to compute the step of the optimizer
at the current iteration. These scales are estimated by the user intuitively as shown in previous
examples of this chapter. In our particular case, a common choice for the scale parameters is to set
all those associated with the matrix coefficients to 1.0, that is, the first N × N factors. Then, we set
the remaining scale factors to a small value.
Here the affine transform is represented by the matrix M and the vector T. The transformation of a
248
Chapter 3. Registration
point P into P′ is expressed as
P′ x
P′ y
=
M11
M21
M12
M22
Px
Tx
·
+
Py
Ty
(3.1)
Based on the above discussion, we need much smaller scales for translation parameters of vector
T (Tx , Ty ) compared to the parameters of matrix M (M11 , M12 , M21 , M22 ). However, it is not easy
to have an intuitive estimation of all parameter scales when we have to deal with a large parameter
space.
Fortunately,
ITKv4 provides a framework for automated parameter scaling.
itk::RegistrationParameterScalesEstimator vastly reduces the difficulty of tuning parameters for different transform/metric combinations. Parameter scales are estimated by analyzing
the result of a small parameter update on the change in the magnitude of physical space deformation
induced by the transformation.
The impact from a unit change of a parameter may be defined in multiple ways, such
as the maximum shift of voxels in index or physical space, or the average norm of
transform Jacobian. Filters itk::RegistrationParameterScalesFromPhysicalShift and
itk::RegistrationParameterScalesFromIndexShift use the first definition to estimate the
scales, while the itk::RegistrationParameterScalesFromJacobian filter estimates scales
based on the later definition. In all methods, the goal is to rescale the transform parameters such that
a unit change of each scaled parameter will have the same impact on deformation.
In this example the first filter is chosen to estimate the parameter scales. The scales estimator will
then be passed to optimizer.
using ScalesEstimatorType =
itk::RegistrationParameterScalesFromPhysicalShift;
ScalesEstimatorType::Pointer scalesEstimator =
ScalesEstimatorType::New();
scalesEstimator->SetMetric( affineMetric );
scalesEstimator->SetTransformForward( true );
affineOptimizer->SetScalesEstimator( scalesEstimator );
The step length has to be proportional to the expected values of the parameters in the search space.
Since the expected values of the matrix coefficients are around 1.0, the initial step of the optimization
should be a small number compared to 1.0. As a guideline, it is useful to think of the matrix
coefficients as combinations of cos(θ) and sin(θ). This leads to use values close to the expected
rotation measured in radians. For example, a rotation of 1.0 degree is about 0.017 radians.
However, we need not worry about the above considerations. Thanks to the ScalesEstimator, the
initial step size can also be estimated automatically, either at each iteration or only at the first iteration. In this example we choose to estimate learning rate once at the begining of the registration
process.
3.8. Multi-Stage Registration
249
affineOptimizer->SetDoEstimateLearningRateOnce( true );
affineOptimizer->SetDoEstimateLearningRateAtEachIteration( false );
At the second stage, we run two levels of registration, where the second level is run in full resolution
in which we do the final adjustments of the output parameters.
constexpr unsigned int numberOfLevels2 = 2;
ARegistrationType::ShrinkFactorsArrayType shrinkFactorsPerLevel2;
shrinkFactorsPerLevel2.SetSize( numberOfLevels2 );
shrinkFactorsPerLevel2[0] = 2;
shrinkFactorsPerLevel2[1] = 1;
ARegistrationType::SmoothingSigmasArrayType smoothingSigmasPerLevel2;
smoothingSigmasPerLevel2.SetSize( numberOfLevels2 );
smoothingSigmasPerLevel2[0] = 1;
smoothingSigmasPerLevel2[1] = 0;
affineRegistration->SetNumberOfLevels ( numberOfLevels2 );
affineRegistration->SetShrinkFactorsPerLevel( shrinkFactorsPerLevel2 );
affineRegistration->SetSmoothingSigmasPerLevel( smoothingSigmasPerLevel2 );
Finally we trigger the registration process by calling Update() and add the output transform of
the last stage to the composite transform. This composite transform will be considered as the final
transform of this multistage registration process and will be used by the resampler to resample the
moving image in to the virtual domain space (fixed image space if there is no fixed initial transform).
try
{
affineRegistration->Update();
std::cout << "Optimizer stop condition: "
<< affineRegistration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
compositeTransform->AddTransform(
affineRegistration->GetModifiableTransform() );
Let’s execute this example using the following multi-modality images:
• BrainT1SliceBorder20.png
• BrainProtonDensitySliceR10X13Y17.png
250
Chapter 3. Registration
The second image is the result of intentionally rotating the first image by 10 degrees and then translating by (−13, −17). Both images have unit-spacing and are shown in Figure 3.36.
The registration converges after 5 iterations in the translation stage. Also, in the second stage, the
registration converges after 46 iterations in the first level, and 6 iterations in the second level. The
final results when printed as an array of parameters are:
Initial parameters of the registration process:
[3, 5]
Translation parameters after first registration stage:
[9.0346, 10.8303]
Affine parameters after second registration stage:
[0.9864, -0.1733, 0.1738, 0.9863, 0.9693, 0.1482]
As it can be seen, the translation parameters after the first stage compensate most of the offset between the fixed and moving images. When the images are close to each other, the affine registration
is run for the rotation and the final match. By reordering the Affine array of parameters as coefficients of matrix M and vector T they can now be seen as
M=
0.9864 −0.1733
0.1738 0.9863
and T =
0.9693
0.1482
(3.2)
In this form, it is easier to interpret the effect of the transform. The matrix M is responsible for
scaling, rotation and shearing while T is responsible for translations.
The second component of the matrix values is usually associated with sin θ. We obtain the rotation
through SVD of the affine matrix. The value is 9.975 degrees, which is approximately the intentional
misalignment of 10.0 degrees.
Also, let’s compute the total translation values resulting from initial transform, translation transform,
and the Affine transform together.
In X direction:
3 + 9.0346 + 0.9693 = 13.0036
(3.3)
5 + 10.8303 + 0.1482 = 15.9785
(3.4)
In Y direction:
It can be seen that the translation values closely match the true misalignment introduced in the
moving image.
It is important to note that once the images are registered at a sub-pixel level, any further improvement of the registration relies heavily on the quality of the interpolator. It may then be reasonable to
use a coarse and fast interpolator in the lower resolution levels and switch to a high-quality but slow
3.8. Multi-Stage Registration
251
Figure 3.36: Fixed and moving images provided as input to the registration method using the AffineTransform.
interpolator in the final resolution level. However, in this example we used a linear interpolator for
all stages and different registration levels since it is so fast.
The result of resampling the moving image is presented in the left image of Figure 3.37. The center
and right images of the figure depict a checkerboard composite of the fixed and moving images
before and after registration.
3.8.2 Cascaded Multistage Registration
The source code for this section can be found in the file
MultiStageImageRegistration2.cxx.
This examples shows how different stages can be cascaded together directly in a multistage registration process. The example code is, for the most part, identical to the previous multistage example.
The main difference is that no initial transform is used, and the output of the first stage is directly
linked to the second stage, and the whole registration process is triggered only once by calling
Update() after the last stage stage.
We will focus on the most relevent changes in current code and skip all the similar parts already
explained in the previous example.
Let’s start by defining different types of the first stage.
252
Chapter 3. Registration
Figure 3.37: Mapped moving image (left) and composition of fixed and moving images before (center) and
after (right) registration.
using TTransformType = itk::TranslationTransform< double, Dimension >;
using TOptimizerType = itk::RegularStepGradientDescentOptimizerv4;
using MetricType = itk::MattesMutualInformationImageToImageMetricv4<
FixedImageType,
MovingImageType >;
using TRegistrationType = itk::ImageRegistrationMethodv4<
FixedImageType,
MovingImageType >;
Type definitions are the same as previous example with an important subtle change: the transform
type is not passed to the registration method as a template parameter anymore. In this case, the
registration filter will consider the transform base class itk::Transform as the type of its output
transform.
Instead of passing the transform type, we create an explicit instantiation of the transform object outside of the registration filter, and connect that to the registration object using the
SetInitialTransform() method. Also, by calling InPlaceOn() method, this transform object
will be the output transform of the registration filter or will be grafted to the output.
TTransformType::Pointer translationTx = TTransformType::New();
transRegistration->SetInitialTransform( translationTx );
transRegistration->InPlaceOn();
Also, there is no initial transform defined for this example.
As in the previous example, the first stage is run using only one level of registration at a coarse
resolution level. However, notice that we do not need to update the translation registration filter at
this step since the output of this stage will be directly connected to the initial input of the next stage.
3.8. Multi-Stage Registration
253
Due to ITK’s pipeline structure, when we call the Update() at the last stage, the first stage will be
updated as well.
Now we upgrade to an Affine transform as the second stage of registration process, and as before,
we initially define and instantiate different components of the current registration stage. We have
used a new optimizer but the same metric in new configurations.
using ATransformType =
itk::AffineTransform< double, Dimension >;
using AOptimizerType =
itk::ConjugateGradientLineSearchOptimizerv4Template;
using ARegistrationType = itk::ImageRegistrationMethodv4<
FixedImageType,
MovingImageType >;
Again notice that TransformType is not passed to the type definition of the registration filter. It is
important because when the registration filter considers transform base class itk::Transform as
the type of its output transform, it prevents the type mismatch when the two stages are cascaded to
each other.
Then, all components are instantiated using their New() method and connected to the registration
object among the transform type. Despite the previous example, here we use the fixed image’s center
of mass to initialize the fixed parameters of the Affine transform. itk::ImageMomentsCalculator
filter is used for this purpose.
using FixedImageCalculatorType = itk::ImageMomentsCalculator;
FixedImageCalculatorType::Pointer fixedCalculator =
FixedImageCalculatorType::New();
fixedCalculator->SetImage( fixedImage );
fixedCalculator->Compute();
FixedImageCalculatorType::VectorType fixedCenter =
fixedCalculator->GetCenterOfGravity();
Then, we initialize the fixed parameters (center of rotation) in the Affine transform and connect that
to the registration object.
ATransformType::Pointer
affineTx
= ATransformType::New();
const unsigned int numberOfFixedParameters =
affineTx->GetFixedParameters().Size();
ATransformType::ParametersType fixedParameters( numberOfFixedParameters );
for (unsigned int i = 0; i < numberOfFixedParameters; ++i)
{
fixedParameters[i] = fixedCenter[i];
}
affineTx->SetFixedParameters( fixedParameters );
254
affineRegistration->SetInitialTransform( affineTx
affineRegistration->InPlaceOn();
Chapter 3. Registration
);
Now, the output of the first stage is wrapped through a itk::DataObjectDecorator
and is passed to the input of the second stage as the moving initial transform via
SetMovingInitialTransformInput() method. Note that this API has an “Input” word attached
to the name of another initialization method SetMovingInitialTransform() that already has been
used in previous example. This extension means that the following API expects a data object decorator type.
affineRegistration->SetMovingInitialTransformInput(
transRegistration->GetTransformOutput() );
Second stage runs two levels of registration, where the second level is run in full resolution.
Once all the registration components are in place, finally we trigger the whole registration process,
including two cascaded registration stages, by calling Update() on the registration filter of the last
stage, which causes both stages be updated.
try
{
affineRegistration->Update();
std::cout << "Optimizer stop condition: "
<< affineRegistration->
GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & err )
{
std::cout << "ExceptionObject caught !" << std::endl;
std::cout << err << std::endl;
return EXIT_FAILURE;
}
Finally, a composite transform is used to concatenate the results of all stages together, which will
be considered as the final output of this multistage process and will be passed to the resampler to
resample the moving image into the virtual domain space (fixed image space if there is no fixed
initial transform).
using CompositeTransformType = itk::CompositeTransform< double,
Dimension >;
CompositeTransformType::Pointer compositeTransform =
CompositeTransformType::New();
compositeTransform->AddTransform( translationTx );
compositeTransform->AddTransform( affineTx );
Let’s execute this example using the same multi-modality images as before. The registration con-
3.8. Multi-Stage Registration
255
Figure 3.38: Mapped moving image (left) and composition of fixed and moving images before (center) and
after (right) registration.
verges after 6 iterations in the first stage, also in 45 and 11 iterations corresponding to the first level
and second level of the Affine stage. The final results when printed as an array of parameters are:
Translation parameters after first registration stage:
[11.600, 15.1814]
Affine parameters after second registration stage:
[0.9860, -0.1742, 0.1751, 0.9862, 0.9219, 0.8023]
Let’s reorder the Affine array of parameters again as coefficients of matrix M and vector T. They
can now be seen as
M=
0.9860 −0.1742
0.1751 0.9862
and T =
0.9219
0.8023
(3.5)
10.02 degrees is the rotation value computed from the affine matrix parameters, which approximately equals the intentional misalignment.
Also for the total translation value resulted from both transforms, we have:
In X direction:
11.6004 + 0.9219 = 12.5223
(3.6)
15.1814 + 0.8023 = 15.9837
(3.7)
In Y direction:
These results closely match the true misalignment introduced in the moving image.
256
Chapter 3. Registration
The result of resampling the moving image is presented in the left image of Figure 3.38. The center
and right images of the figure depict a checkerboard composite of the fixed and moving images
before and after registration.
With the completion of these examples, we will now review the main features of the components
forming the registration framework.
3.9. Transforms
257
Point
Vector
Covariant
Vectors
Figure 3.39: Geometric representation objects in ITK.
3.9 Transforms
In the Insight Toolkit, itk::Transform objects encapsulate the mapping of points and vectors
from an input space to an output space. If a transform is invertible, back transform methods are
also provided. Currently, ITK provides a variety of transforms from simple translation, rotation and
scaling to general affine and kernel transforms. Note that, while in this section we discuss transforms
in the context of registration, transforms are general and can be used for other applications. Some
of the most commonly used transforms will be discussed in detail later. Let’s begin by introducing
the objects used in ITK for representing basic spatial concepts.
3.9.1 Geometrical Representation
ITK implements a consistent geometric representation of space. The characteristics of classes involved in this representation are summarized in Table 3.1. In this regard, ITK takes full advantage
of the capabilities of Object Oriented programming and resists the temptation of using simple arrays
of float or double in order to represent geometrical objects. The use of basic arrays would have
blurred the important distinction between the different geometrical concepts and would have allowed
for the innumerable conceptual and programming errors that result from using a vector where a point
is needed or vice versa.
Additional uses of the itk::Point, itk::Vector and itk::CovariantVector classes have
been discussed in the Data Representation chaper of Book 1. Each one of these classes behaves
differently under spatial transformations. It is therefore quite important to keep their distinction
clear. Figure 3.39 illustrates the differences between these concepts.
Transform classes provide different methods for mapping each one of the basic spacerepresentation objects. Points, vectors and covariant vectors are transformed using the methods
TransformPoint(), TransformVector() and TransformCovariantVector() respectively.
One of the classes that deserves further comments is the itk::Vector. This ITK class tends to
be misinterpreted as a container of elements instead of a geometrical object. This is a common
misconception originating from the colloquial use by computer scientists and software engineers of
258
Chapter 3. Registration
Class
itk::Point
itk::Vector
itk::CovariantVector
Geometrical concept
Position in space. In N-dimensional space it is represented
by an array of N numbers associated with space coordinates.
Relative position between two points. In N-dimensional
space it is represented by an array of N numbers, each one
associated with the distance along a coordinate axis. Vectors do not have a position in space. A vector is defined as
the subtraction of two points.
Orthogonal direction to a (N − 1)-dimensional manifold in
space. For example, in 3D it corresponds to the vector orthogonal to a surface. This is the appropriate class for representing gradients of functions. Covariant vectors do not
have a position in space. Covariant vector should not be
added to Points, nor to Vectors.
Table 3.1: Summary of objects representing geometrical concepts in ITK.
the term “Vector”. The actual word “Vector” is relatively young. It was coined by William Hamilton
in his book “Elements of Quaternions” published in 1886 (post-mortem)[24]. In the same text
Hamilton coined the terms: “Scalar”, “Versor” and “Tensor”. Although the modern term of “Tensor”
is used in Calculus in a different sense of what Hamilton defined in his book at the time [17].
A “Vector” is, by definition, a mathematical object that embodies the concept of “direction in space”.
Strictly speaking, a Vector describes the relationship between two Points in space, and captures both
their relative distance and orientation.
Computer scientists and software engineers misused the term vector in order to represent the concept
of an “Indexed Set” [5]. Mechanical Engineers and Civil Engineers, who deal with the real world
of physical objects will not commit this mistake and will keep the word “Vector” attached to a
geometrical concept. Biologists, on the other hand, will associate “Vector” to a “vehicle” that allows
them to direct something in a particular direction, for example, a virus that allows them to insert
pieces of code into a DNA strand [35].
Textbooks in programming do not help to clarify those concepts and loosely use the term “Vector”
for the purpose of representing an “enumerated set of common elements”. STL follows this trend and
continues using the word “Vector” in this manner [5, 1]. Linear algebra separates the “Vector” from
its notion of geometric reality and makes it an abstract set of numbers with arithmetic operations
associated.
For those of you who are looking for the “Vector” in the Software Engineering sense, please look
at the itk::Array and itk::FixedArray classes that actually provide such functionalities. Additionally, the itk::VectorContainer and itk::MapContainer classes may be of interest too.
These container classes are intended for algorithms which require insertion and deletion of elements,
and those which may have large numbers of elements.
3.9. Transforms
259
The Insight Toolkit deals with real objects that inhabit the physical space. This is particularly true
in the context of the image registration framework. We chose to give the appropriate name to the
mathematical objects that describe geometrical relationships in N-Dimensional space. It is for this
reason that we explicitly make clear the distinction between Point, Vector and CovariantVector,
despite the fact that most people would be happy with a simple use of double[3] for the three
concepts and then will proceed to perform all sort of conceptually flawed operations such as
• Adding two Points
• Dividing a Point by a Scalar
• Adding a Covariant Vector to a Point
• Adding a Covariant Vector to a Vector
In order to enforce the correct use of the geometrical concepts in ITK we organized these classes
in a hierarchy that supports reuse of code and compartmentalizes the behavior of the individual
classes. The use of the itk::FixedArray as the base class of the itk::Point, the itk::Vector
and the itk::CovariantVector was a design decision based on the decision to use the correct
nomenclature.
An itk::FixedArray is an enumerated collection with a fixed number of elements. You can
instantiate a fixed array of letters, or a fixed array of images, or a fixed array of transforms, or a
fixed array of geometrical shapes. Therefore, the FixedArray only implements the functionality that
is necessary to access those enumerated elements. No assumptions can be made at this point on
any other operations required by the elements of the FixedArray, except that it will have a default
constructor.
The itk::Point is a type that represents the spatial coordinates of a spatial location. Based on
geometrical concepts we defined the valid operations of the Point class. In particular we made
sure that no operator+() was defined between Points, and that no operator*( scalar ) nor
operator/( scalar ) were defined for Points.
In other words, you can perform ITK operations such as:
• Vector = Point - Point
• Point += Vector
• Point -= Vector
• Point = BarycentricCombination( Point, Point )
and you cannot (because you should not) perform operations such as
• Point = Point * Scalar
260
Chapter 3. Registration
• Point = Point + Point
• Point = Point / Scalar
The itk::Vector is, by Hamilton’s definition, the subtraction between two points. Therefore a
Vector must satisfy the following basic operations:
• Vector = Point - Point
• Point = Point + Vector
• Point = Point - Vector
• Vector = Vector + Vector
• Vector = Vector - Vector
An itk::Vector object is intended to be instantiated over elements that support mathematical
operation such as addition, subtraction and multiplication by scalars.
3.9.2 Transform General Properties
Each transform class typically has several methods for setting its parameters. For example,
itk::Euler2DTransform provides methods for specifying the offset, angle, and the entire rotation matrix. However, for use in the registration framework, the parameters are represented by a
flat Array of doubles to facilitate communication with generic optimizers. In the case of the Euler2DTransform, the transform is also defined by three doubles: the first representing the angle,
and the last two the offset. The flat array of parameters is defined using SetParameters(). A
description of the parameters and their ordering is documented in the sections that follow.
In the context of registration, the transform parameters define the search space for optimizers. That
is, the goal of the optimization is to find the set of parameters defining a transform that results in
the best possible value of an image metric. The more parameters a transform has, the longer its
computational time will be when used in a registration method since the dimension of the search
space will be equal to the number of transform parameters.
Another requirement that the registration framework imposes on the transform classes is the computation of their Jacobians. In general, metrics require the knowledge of the Jacobian in order to
compute Metric derivatives. The Jacobian is a matrix whose elements are the partial derivatives of
3.9. Transforms
261
Behavior
Maps every point to
itself, every vector to
itself and every covariant vector to itself.
Number of
Parameters
0
Parameter
Ordering
NA
Restrictions
Only defined when the input and output space has the
same number of dimensions.
Table 3.2: Characteristics of the identity transform.
the output point with respect to the array of parameters that defines the transform:7
J=
∂x1
∂p1
∂x2
∂p1
∂x1
∂p2
∂x2
∂p2
∂xn
∂p1
∂xn
∂p2
..
.
..
.
···
···
..
.
···
∂x1
∂pm
∂x2
∂pm
..
.
∂xn
∂pm
(3.8)
where {pi } are the transform parameters and {xi } are the coordinates of the output point. Within
this framework, the Jacobian is represented by an itk::Array2D of doubles and is obtained from
the transform by method GetJacobian(). The Jacobian can be interpreted as a matrix that indicates
for a point in the input space how much its mapping on the output space will change as a response
to a small variation in one of the transform parameters. Note that the values of the Jacobian matrix
depend on the point in the input space. So actually the Jacobian can be noted as J(X), where
X = {xi }. The use of transform Jacobians enables the efficient computation of metric derivatives.
When Jacobians are not available, metrics derivatives have to be computed using finite differences
at a price of 2M evaluations of the metric value, where M is the number of transform parameters.
The following sections describe the main characteristics of the transform classes available in ITK.
3.9.3 Identity Transform
The identity transform itk::IdentityTransform is mainly used for debugging purposes. It is
provided to methods that require a transform and in cases where we want to have the certainty that
the transform will have no effect whatsoever in the outcome of the process. The main characteristics
of the identity transform are summarized in Table 3.2
262
Chapter 3. Registration
Behavior
Represents a simple
translation of points
in the input space and
has no effect on vectors or covariant vectors.
Number of
Parameters
Same as the
input space
dimension.
Parameter
Ordering
The i-th parameter represents the
translation in the
i-th dimension.
Restrictions
Only defined when the input
and output space have the
same number of dimensions.
Table 3.3: Characteristics of the TranslationTransform class.
3.9.4 Translation Transform
The itk::TranslationTransform is probably the simplest yet one of the most useful transformations. It maps all Points by adding a Vector to them. Vector and covariant vectors remain unchanged
under this transformation since they are not associated with a particular position in space. Translation is the best transform to use when starting a registration method. Before attempting to solve
for rotations or scaling it is important to overlap the anatomical objects in both images as much as
possible. This is done by resolving the translational misalignment between the images. Translations
also have the advantage of being fast to compute and having parameters that are easy to interpret.
The main characteristics of the translation transform are presented in Table 3.3.
3.9.5 Scale Transform
The itk::ScaleTransform represents a simple scaling of the vector space. Different scaling
factors can be applied along each dimension. Points are transformed by multiplying each one of their
coordinates by the corresponding scale factor for the dimension. Vectors are transformed in the same
way as points. Covariant vectors, on the other hand, are transformed differently since anisotropic
scaling does not preserve angles. Covariant vectors are transformed by dividing their components
by the scale factor of the corresponding dimension. In this way, if a covariant vector was orthogonal
to a vector, this orthogonality will be preserved after the transformation. The following equations
summarize the effect of the transform on the basic geometric objects.
Point
P′
Vector
V′
CovariantVector C′
=
=
=
T (P) : P′ i
T (V) : V′ i
T (C) : C′ i
=
=
=
Pi · Si
Vi · Si
Ci /Si
(3.9)
where Pi , Vi and Ci are the point, vector and covariant vector i-th components while Si is the scaling
7 Note that the term Jacobian is also commonly used for the matrix representing the derivatives of output point coordinates
with respect to input point coordinates. Sometimes the term is loosely used to refer to the determinant of such a matrix. [17]
3.9. Transforms
263
Behavior
Points are transformed by multiplying each one of
their coordinates by
the
corresponding
scale factor for the
dimension. Vectors
are transformed as
points.
Covariant
vectors are transformed by dividing
their components by
the scale factor in
the
corresponding
dimension.
Number of
Parameters
Same as the
input space
dimension.
Parameter
Ordering
The i-th parameter represents the
scaling in the i-th
dimension.
Restrictions
Only defined when the input
and output space have the
same number of dimensions.
Table 3.4: Characteristics of the ScaleTransform class.
factor along dimension i − th. The following equation illustrates the effect of the scaling transform
on a 3D point.
x′
S1
y′ = 0
z′
0
0
S2
0
0
x
0 · y
S3
z
(3.10)
Scaling appears to be a simple transformation but there are actually a number of issues to keep
in mind when using different scale factors along every dimension. There are subtle effects—for
example, when computing image derivatives. Since derivatives are represented by covariant vectors,
their values are not intuitively modified by scaling transforms.
One of the difficulties with managing scaling transforms in a registration process is that typical optimizers manage the parameter space as a vector space where addition is the basic operation. Scaling
is better treated in the frame of a logarithmic space where additions result in regular multiplicative
increments of the scale. Gradient descent optimizers have trouble updating step length, since the
effect of an additive increment on a scale factor diminishes as the factor grows. In other words, a
scale factor variation of (1.0 + ε) is quite different from a scale variation of (5.0 + ε).
Registrations involving scale transforms require careful monitoring of the optimizer parameters in
order to keep it progressing at a stable pace. Note that some of the transforms discussed in following
sections, for example, the AffineTransform, have hidden scaling parameters and are therefore subject
to the same vulnerabilities of the ScaleTransform.
264
Chapter 3. Registration
Behavior
Points are transformed by multiplying each one of
their coordinates by
the
corresponding
scale factor for the
dimension. Vectors
are transformed as
points.
Covariant
vectors are transformed by dividing
their components by
the scale factor in
the
corresponding
dimension.
Number of
Parameters
Same as the
input space
dimension.
Parameter
Ordering
The i-th parameter represents the
scaling in the i-th
dimension.
Restrictions
Only defined when the input and output space have
the same number of dimensions. The difference between this transform and
the ScaleTransform is that
here the scaling factors are
passed as logarithms, in this
way their behavior is closer
to the one of a Vector space.
Table 3.5: Characteristics of the ScaleLogarithmicTransform class.
In cases involving misalignments with simultaneous translation, rotation and scaling components it
may be desirable to solve for these components independently. The main characteristics of the scale
transform are presented in Table 3.4.
3.9.6 Scale Logarithmic Transform
The itk::ScaleLogarithmicTransform is a simple variation of the itk::ScaleTransform. It
is intended to improve the behavior of the scaling parameters when they are modified by optimizers. The difference between this transform and the ScaleTransform is that the parameter factors
are passed here as logarithms. In this way, multiplicative variations in the scale become additive
variations in the logarithm of the scaling factors.
3.9.7 Euler2DTransform
itk::Euler2DTransform implements a rigid transformation in 2D. It is composed of a plane
rotation and a two-dimensional translation. The rotation is applied first, followed by the translation.
The following equation illustrates the effect of this transform on a 2D point,
x′
y′
=
cos θ − sin θ
sin θ cos θ
x
Tx
·
+
y
Ty
(3.11)
3.9. Transforms
265
Behavior
Represents a 2D rotation and a 2D translation.
Note that
the translation component has no effect
on the transformation
of vectors and covariant vectors.
Number of
Parameters
3
Parameter
Ordering
The first parameter is the angle
in radians and the
last two parameters are the translation in each dimension.
Restrictions
Only defined for
dimensional input
output spaces.
twoand
Table 3.6: Characteristics of the Euler2DTransform class.
where θ is the rotation angle and (Tx , Ty ) are the components of the translation.
A challenging aspect of this transformation is the fact that translations and rotations do not form a
vector space and cannot be managed as linearly independent parameters. Typical optimizers make
the loose assumption that parameters exist in a vector space and rely on the step length to be small
enough for this assumption to hold approximately.
In addition to the non-linearity of the parameter space, the most common difficulty found when using
this transform is the difference in units used for rotations and translations. Rotations are measured
in radians; hence, their values are in the range [−π, π]. Translations are measured in millimeters and
their actual values vary depending on the image modality being considered. In practice, translations
have values on the order of 10 to 100. This scale difference between the rotation and translation
parameters is undesirable for gradient descent optimizers because they deviate from the trajectories
of descent and make optimization slower and more unstable. In order to compensate for these
differences, ITK optimizers accept an array of scale values that are used to normalize the parameter
space.
Registrations involving angles and translations should take advantage of the scale normalization
functionality in order to obtain the best performance out of the optimizers. The main characteristics
of the Euler2DTransform class are presented in Table 3.6.
3.9.8 CenteredRigid2DTransform
itk::CenteredRigid2DTransform implements a rigid transformation in 2D. The main difference
between this transform and the itk::Euler2DTransform is that here we can specify an arbitrary
center of rotation, while the Euler2DTransform always uses the origin of the coordinate system as
the center of rotation. This distinction is quite important in image registration since ITK images usually have their origin in the corner of the image rather than the middle. Rotational mis-registrations
usually exist, however, as rotations around the center of the image, or at least as rotations around a
266
Chapter 3. Registration
Behavior
Number of
Parameters
5
Represents a 2D rotation around a userprovided center followed by a 2D translation.
Parameter
Ordering
The first parameter is the angle in
radians. Second
and third are the
center of rotation coordinates
and the last two
parameters are
the translation in
each dimension.
Restrictions
Only defined for
dimensional input
output spaces.
twoand
Table 3.7: Characteristics of the CenteredRigid2DTransform class.
point in the middle of the anatomical structure captured by the image. Using gradient descent optimizers, it is almost impossible to solve non-origin rotations using a transform with origin rotations
since the deep basin of the real solution is usually located across a high ridge in the topography of
the cost function.
In practice, the user must supply the center of rotation in the input space, the angle of rotation
and a translation to be applied after the rotation. With these parameters, the transform initializes a
rotation matrix and a translation vector that together perform the equivalent of translating the center
of rotation to the origin of coordinates, rotating by the specified angle, translating back to the center
of rotation and finally translating by the user-specified vector.
As with the Euler2DTransform, this transform suffers from the difference in units used for rotations
and translations. Rotations are measured in radians; hence, their values are in the range [−π, π].
The center of rotation and the translations are measured in millimeters, and their actual values vary
depending on the image modality being considered. Registrations involving angles and translations
should take advantage of the scale normalization functionality of the optimizers in order to get the
best performance out of them.
The following equation illustrates the effect of the transform on an input point (x, y) that maps to the
output point (x′ , y′ ),
x′
y′
=
cos θ
sin θ
− sin θ
cosθ
x − Cx
Tx + Cx
·
+
y − Cy
Ty + Cy
(3.12)
where θ is the rotation angle, (Cx ,Cy ) are the coordinates of the rotation center and (Tx , Ty ) are the
components of the translation. Note that the center coordinates are subtracted before the rotation and
added back after the rotation. The main features of the CenteredRigid2DTransform are presented in
Table 3.7.
3.9. Transforms
267
Behavior
Number of
Parameters
4
Represents a 2D rotation, homogeneous
scaling and a 2D
translation. Note that
the translation component has no effect
on the transformation
of vectors and covariant vectors.
Parameter
Ordering
The first parameter is the
scaling factor for
all dimensions,
the second is the
angle in radians,
and
the
last
two parameters
are the translations in (x, y)
respectively.
Restrictions
Only defined for
dimensional input
output spaces.
twoand
Table 3.8: Characteristics of the Similarity2DTransform class.
3.9.9 Similarity2DTransform
The itk::Similarity2DTransform can be seen as a rigid transform combined with an isotropic
scaling factor. This transform preserves angles between lines. In its 2D implementation, the four
parameters of this transformation combine the characteristics of the itk::ScaleTransform and
itk::Euler2DTransform. In particular, those relating to the non-linearity of the parameter space
and the non-uniformity of the measurement units. Gradient descent optimizers should be used with
caution on such parameter spaces since the notions of gradient direction and step length are illdefined.
The following equation illustrates the effect of the transform on an input point (x, y) that maps to the
output point (x′ , y′ ),
x′
y′
=
λ
0
0
λ
cos θ − sin θ
x − Cx
Tx + Cx
·
·
+
sin θ cos θ
y − Cy
Ty + Cy
(3.13)
where λ is the scale factor, θ is the rotation angle, (Cx ,Cy ) are the coordinates of the rotation center
and (Tx , Ty ) are the components of the translation. Note that the center coordinates are subtracted
before the rotation and scaling, and they are added back afterwards. The main features of the Similarity2DTransform are presented in Table 3.8.
A possible approach for controlling optimization in the parameter space of this transform is to dynamically modify the array of scales passed to the optimizer. The effect produced by the parameter
scaling can be used to steer the walk in the parameter space (by giving preference to some of the
parameters over others). For example, perform some iterations updating only the rotation angle, then
balance the array of scale factors in the optimizer and perform another set of iterations updating only
the translations.
268
Chapter 3. Registration
Behavior
Represents a 3D rotation and
a 3D translation. The rotation is specified as a quaternion, defined by a set of four
numbers q. The relationship
between quaternion and rotation about vector n by angle θ is as follows:
Number of
Parameters
7
Parameter
Ordering
The first four parameters defines
the
quaternion
and the last three
parameters the
translation
in
each dimension.
Restrictions
Only defined for
three-dimensional
input and output
spaces.
q = (n sin(θ/2), cos(θ/2))
Note that if the quaternion
is not of unit length, scaling
will also result.
Table 3.9: Characteristics of the QuaternionRigidTransform class.
3.9.10 QuaternionRigidTransform
The itk::QuaternionRigidTransform class implements a rigid transformation in 3D space. The
rotational part of the transform is represented using a quaternion while the translation is represented
with a vector. Quaternions components do not form a vector space and hence raise the same concerns
as the itk::Similarity2DTransform when used with gradient descent optimizers.
The itk::QuaternionRigidTransformGradientDescentOptimizer was introduced into the
toolkit to address these concerns. This specialized optimizer implements a variation of a gradient descent algorithm adapted for a quaternion space. This class ensures that after advancing in
any direction on the parameter space, the resulting set of transform parameters is mapped back into
the permissible set of parameters. In practice, this comes down to normalizing the newly-computed
quaternion to make sure that the transformation remains rigid and no scaling is applied. The main
characteristics of the QuaternionRigidTransform are presented in Table 3.9.
The Quaternion rigid transform also accepts a user-defined center of rotation. In this way, the transform can easily be used for registering images where the rotation is mostly relative to the center of
the image instead of one of the corners. The coordinates of this rotation center are not subject to
optimization. They only participate in the computation of the mappings for Points and in the computation of the Jacobian. The transformations for Vectors and CovariantVector are not affected by
the selection of the rotation center.
3.9. Transforms
269
Behavior
Represents a 3D rotation. The rotation
is specified by a versor or unit quaternion. The rotation
is performed around
a user-specified center of rotation.
Number of
Parameters
3
Parameter
Ordering
The three parameters define the
versor.
Restrictions
Only defined for threedimensional input and
output spaces.
Table 3.10: Characteristics of the Versor Transform
3.9.11 VersorTransform
By definition, a Versor is the rotational part of a Quaternion. It can also be defined as a unitquaternion [24, 27]. Versors only have three independent components, since they are restricted to
reside in the space of unit-quaternions. The implementation of versors in the toolkit uses a set of
three numbers. These three numbers correspond to the first three components of a quaternion. The
fourth component of the quaternion is computed internally such that the quaternion is of unit length.
The main characteristics of the itk::VersorTransform are presented in Table 3.10.
This transform exclusively represents rotations in 3D. It is intended to rapidly solve the rotational
component of a more general misalignment. The efficiency of this transform comes from using a
parameter space of reduced dimensionality. Versors are the best possible representation for rotations
in 3D space. Sequences of versors allow the creation of smooth rotational trajectories; for this
reason, they behave stably under optimization methods.
The space formed by versor parameters is not a vector space. Standard gradient descent algorithms
are not appropriate for exploring this parameter space. An optimizer specialized for the versor space
is available in the toolkit under the name of itk::VersorTransformOptimizer. This optimizer
implements versor derivatives as originally defined by Hamilton [24].
The center of rotation can be specified by the user with the SetCenter() method. The center is not
part of the parameters to be optimized, therefore it remains the same during an optimization process.
Its value is used during the computations for transforming Points and when computing the Jacobian.
3.9.12 VersorRigid3DTransform
The itk::VersorRigid3DTransform implements a rigid transformation in 3D space. It is a variant of the itk::QuaternionRigidTransform and the itk::VersorTransform. It can be seen as
a itk::VersorTransform plus a translation defined by a vector. The advantage of this class with
270
Chapter 3. Registration
Behavior
Represents a 3D rotation and a 3D translation. The rotation
is specified by a versor or unit quaternion, while the translation is represented
by a vector. Users
can specify the coordinates of the center
of rotation.
Number of
Parameters
6
Parameter
Ordering
The first three
parameters define
the versor and
the last three
parameters the
translation
in
each dimension.
Restrictions
Only defined for threedimensional input and
output spaces.
Table 3.11: Characteristics of the VersorRigid3DTransform class.
respect to the QuaternionRigidTransform is that it exposes only six parameters, three for the versor
components and three for the translational components. This reduces the search space for the optimizer to six dimensions instead of the seven dimensional used by the QuaternionRigidTransform.
This transform also allows the users to set a specific center of rotation. The center coordinates are
not modified during the optimization performed in a registration process. The main features of this
transform are summarized in Table 3.11. This transform is probably the best option to use when
dealing with rigid transformations in 3D.
Given that the space of Versors is not a Vector space, typical gradient descent optimizers are not well suited for exploring the parametric space of this transform.
The
itk::VersorRigid3DTranformOptimizer has been introduced in the ITK toolkit with the purpose of providing an optimizer that is aware of the Versor space properties on the rotational part of
this transform, as well as the Vector space properties on the translational part of the transform.
3.9.13 Euler3DTransform
The itk::Euler3DTransform implements a rigid transformation in 3D space. It can be seen as
a rotation followed by a translation. This class exposes six parameters, three for the Euler angles
that represent the rotation and three for the translational components. This transform also allows the
users to set a specific center of rotation. The center coordinates are not modified during the optimization performed in a registration process. The main features of this transform are summarized in
Table 3.12.
Three rotational parameters are non-linear and do not behave like Vector spaces. This must be taken
into account when selecting an optimizer to work with this transform and when fine tuning the
parameters of the optimizer. It is strongly recommended to use this transform by introducing very
3.9. Transforms
271
Behavior
Represents a rigid rotation in 3D space.
That is, a rotation followed by a 3D translation. The rotation is
specified by three angles representing rotations to be applied
around the X, Y and
Z axes one after another. The translation
part is represented by
a Vector. Users can
also specify the coordinates of the center
of rotation.
Number of
Parameters
6
Parameter
Ordering
The first three
parameters are
the rotation angles around X, Y
and Z axes, and
the last three parameters are the
translations along
each dimension.
Restrictions
Only defined for threedimensional input and
output spaces.
Table 3.12: Characteristics of the Euler3DTransform class.
small variations on the rotational components. A small rotation will be in the range of 1 degree,
which in radians is approximately 0.01745.
You should not expect this transform to be able to compensate for large rotations just by being driven
with the optimizer. In practice you must provide a reasonable initialization of the transform angles
and only need to correct for residual rotations in the order of 10 or 20 degrees.
3.9.14 Similarity3DTransform
The itk::Similarity3DTransform implements a similarity transformation in 3D space. It can
be seen as an homogeneous scaling followed by a itk::VersorRigid3DTransform. This class
exposes seven parameters: one for the scaling factor, three for the versor components and three for
the translational components. This transform also allows the user to set a specific center of rotation.
The center coordinates are not modified during the optimization performed in a registration process.
Both the rotation and scaling operations are performed with respect to the center of rotation. The
main features of this transform are summarized in Table 3.13.
The scaling and rotational spaces are non-linear and do not behave like Vector spaces. This must be
taken into account when selecting an optimizer to work with this transform and when fine tuning the
parameters of the optimizer.
272
Chapter 3. Registration
Behavior
Represents a 3D rotation, a 3D translation and homogeneous scaling. The
scaling factor is specified by a scalar, the
rotation is specified
by a versor, and the
translation is represented by a vector.
Users can also specify the coordinates of
the center of rotation,
which is the same
center used for scaling.
Number of
Parameters
7
Parameter
Ordering
The first three
parameters define the Versor,
the next three
parameters the
translation
in
each dimension,
and the last parameter is the
isotropic scaling
factor.
Restrictions
Only defined for threedimensional input and
output spaces.
Table 3.13: Characteristics of the Similarity3DTransform class.
3.9.15 Rigid3DPerspectiveTransform
The itk::Rigid3DPerspectiveTransform implements a rigid transformation in 3D space followed by a perspective projection. This transform is intended to be used in 3D/2D registration
problems where a 3D object is projected onto a 2D plane. This is the case in Fluoroscopic images
used for image-guided intervention, and it is also the case for classical radiography. Users must
provide a value for the focal distance to be used during the computation of the perspective transform. This transform also allows users to set a specific center of rotation. The center coordinates
are not modified during the optimization performed in a registration process. The main features of
this transform are summarized in Table 3.14. This transform is also used when creating Digitally
Reconstructed Radiographs (DRRs).
The strategies for optimizing the parameters of this transform are the same ones used for optimizing the VersorRigid3DTransform. In particular, you can use the same VersorRigid3DTranformOptimizer in order to optimize the parameters of this class.
3.9.16 AffineTransform
The itk::AffineTransform is one of the most popular transformations used for image registration. Its main advantage comes from its representation as a linear transformation. The main features
3.9. Transforms
273
Behavior
Represents a rigid
3D transformation
followed by a perspective projection.
The rotation is specified by a Versor,
while the translation
is represented by a
Vector. Users can
specify the coordinates of the center
of rotation.
They
must specify a focal
distance to be used
for the perspective
projection.
The
rotation center and
the focal distance
parameters are not
modified during the
optimization process.
Number of
Parameters
6
Parameter
Ordering
The first three
parameters define
the Versor and
the last three
parameters the
Translation
in
each dimension.
Restrictions
Only defined for threedimensional input and
two-dimensional
output
spaces. This is one of the
few transforms where the
input space has a different
dimension from the output
space.
Table 3.14: Characteristics of the Rigid3DPerspectiveTransform class.
Behavior
Represents an affine
transform composed
of rotation, scaling,
shearing and translation. The transform
is specified by a N ×
N matrix and a N × 1
vector where N is the
space dimension.
Number of
Parameters
(N + 1) × N
Parameter
Ordering
The first N × N
parameters define
the matrix in
column-major
order
(where
the column index varies the
fastest). The last
N
parameters
define the translations for each
dimension.
Restrictions
Only defined when the input
and output space have the
same dimension.
Table 3.15: Characteristics of the AffineTransform class.
274
Chapter 3. Registration
of this transform are presented in Table 3.15.
The set of AffineTransform coefficients can actually be represented in a vector space of dimension
(N + 1) × N. This makes it possible for optimizers to be used appropriately on this search space.
However, the high dimensionality of the search space also implies a high computational complexity
of cost-function derivatives. The best compromise in the reduction of this computational time is to
use the transform’s Jacobian in combination with the image gradient for computing the cost-function
derivatives.
The coefficients of the N × N matrix can represent rotations, anisotropic scaling and shearing. These
coefficients are usually of a very different dynamic range compared to the translation coefficients.
Coefficients in the matrix tend to be in the range [−1 : 1], but are not restricted to this interval.
Translation coefficients, on the other hand, can be on the order of 10 to 100, and are basically
related to the image size and pixel spacing.
This difference in scale makes it necessary to take advantage of the functionality offered by the
optimizers for rescaling the parameter space. This is particularly relevant for optimizers based on
gradient descent approaches. This transform lets the user set an arbitrary center of rotation. The
coordinates of the rotation center do not make part of the parameters array passed to the optimizer.
Equation 3.14 illustrates the effect of applying the AffineTransform to a point in 3D space.
x′
M00
y′ = M10
z′
M20
M01
M11
M21
M02
x − Cx
Tx + Cx
M12 · y − Cy + Ty + Cy
M22
z − Cz
Tz + Cz
(3.14)
A registration based on the affine transform may be more effective when applied after simpler transformations have been used to remove the major components of misalignment. Otherwise it will
incur an overwhelming computational cost. For example, using an affine transform, the first set of
optimization iterations would typically focus on removing large translations. This task could instead
be accomplished by a translation transform in a parameter space of size N instead of the (N + 1) × N
associated with the affine transform.
Tracking the evolution of a registration process that uses AffineTransforms can be challenging, since
it is difficult to represent the coefficients in a meaningful way. A simple printout of the transform
coefficients generally does not offer a clear picture of the current behavior and trend of the optimization. A better implementation uses the affine transform to deform a wire-frame cube which is shown
in a 3D visualization display.
3.9.17 BSplineDeformableTransform
The itk::BSplineDeformableTransform is designed to be used for solving deformable registration problems. This transform is equivalent to generating a deformation field where a deformation
vector is assigned to every point in space. The deformation vectors are computed using BSpline interpolation from the deformation values of points located in a coarse grid, which is usually referred
to as the BSpline grid.
3.9. Transforms
275
Behavior
Represents a freeform
deformation
by providing a deformation field from
the interpolation of
deformations in a
coarse grid.
Number of
Parameters
M×N
Parameter
Ordering
Where M is the
number of nodes
in the BSpline
grid and N is the
dimension of the
space.
Restrictions
Only defined when the input and output space have
the same dimension. This
transform has the advantage
of being able to compute deformable registration. It also
has the disadvantage of a
very high-dimensional parametric space, and therefore
requiring long computation
times.
Table 3.16: Characteristics of the BSplineDeformableTransform class.
The BSplineDeformableTransform is not flexible enough to account for large rotations or shearing,
or scaling differences. In order to compensate for this limitation, it provides the functionality of
being composed with an arbitrary transform. This transform is known as the Bulk transform and it
applied to points before they are mapped with the displacement field.
This transform does not provide functionality for mapping Vectors nor CovariantVectors—only
Points can be mapped. This is because the variations of a vector under a deformable transform
actually depend on the location of the vector in space. In other words, Vectors only make sense as
the relative position between two points.
The BSplineDeformableTransform has a very large number of parameters and therefore is well
suited for the itk::LBFGSOptimizer and itk::LBFGSBOptimizer. The use of this transform
was proposed in the following papers [52, 39, 40].
3.9.18 KernelTransforms
Kernel Transforms are a set of Transforms that are also suitable for performing deformable registration. These transforms compute on-the-fly the displacements corresponding to a deformation field.
The displacement values corresponding to every point in space are computed by interpolation from
the vectors defined by a set of Source Landmarks and a set of Target Landmarks.
Several variations of these transforms are available in the toolkit. They differ in the type of interpolation kernel that is used when computing the deformation in a particular point of space. Note that
these transforms are computationally expensive and that their numerical complexity is proportional
to the number of landmarks and the space dimension.
The following is the list of Transforms based on the KernelTransform.
276
Chapter 3. Registration
• itk::ElasticBodySplineKernelTransform
• itk::ElasticBodyReciprocalSplineKernelTransform
• itk::ThinPlateSplineKernelTransform
• itk::ThinPlateR2LogRSplineKernelTransform
• itk::VolumeSplineKernelTransform
Details about the mathematical background of these transform can be found in the paper by Davis
et. al [14] and the papers by Rohr et. al [50, 51].
3.10. Interpolators
277
Transform T(x)
Y
Transform T(x)
Y
Iterator
Walk
X
Moving Image
X
Fixed Image
Moving Image
Fixed Image
Figure 3.40: The moving image is mapped into the fixed image space under some spatial transformation. An
iterator walks through the fixed image and its coordinates are mapped onto the moving image.
3.10 Interpolators
In the registration process, the metric typically compares intensity values in the fixed
image against the corresponding values in the
transformed moving image. When a point is
mapped from one space to another by a transform, it will in general be mapped to a non-grid
position. Therefore, interpolation is required to
evaluate the image intensity at the mapped position.
Figure 3.40 (left) illustrates the mapping of
the fixed image space onto the moving image
space. The transform maps points from the
fixed image coordinate system onto the moving image coordinate system. The figure highlights the region of overlap between the two
images after the mapping. The right side illus- Figure 3.41: Grid positions of the fixed image
trates how an iterator is used to walk through a map to non-grid positions of the moving image.
region of the fixed image. Each one of the iterator positions is mapped by the transform onto
the moving image space in order to find the homologous pixel.
Figure 3.41 presents a detailed view of the mapping from the fixed image to the moving image.
In general, the grid positions of the fixed image will not be mapped onto grid positions of the
moving image. Interpolation is needed for estimating the intensity of the moving image at these
non-grid positions. The service is provided in ITK by interpolator classes that can be plugged into
the registration method.
The following interpolators are available:
278
Chapter 3. Registration
• itk::NearestNeighborInterpolateImageFunction
• itk::LinearInterpolateImageFunction
• itk::BSplineInterpolateImageFunction
• itk::WindowedSincInterpolateImageFunction
In the context of registration, the interpolation method affects the smoothness of the optimization
search space and the overall computation time. On the other hand, interpolations are executed
thousands of times in a single optimization cycle. Hence, the user has to balance the simplicity of
computation with the smoothness of the optimization when selecting the interpolation scheme.
The basic input to an itk::InterpolateImageFunction is the image to be interpolated. Once
an image has been defined using SetInputImage(), a user can interpolate either at a point using
Evaluate() or an index using EvaluateAtContinuousIndex().
Interpolators provide the method IsInsideBuffer() that tests whether a particular image index or
a physical point falls inside the spatial domain for which image pixels exist.
3.10.1 Nearest Neighbor Interpolation
The itk::NearestNeighborInterpolateImageFunction simply uses the intensity of the nearest
grid position. That is, it assumes that the image intensity is piecewise constant with jumps mid-way
between grid positions. This interpolation scheme is cheap as it does not require any floating point
computations.
3.10.2 Linear Interpolation
The itk::LinearInterpolateImageFunction assumes that intensity varies linearly between
grid positions. Unlike nearest neighbor interpolation, the interpolated intensity is spatially continuous. However, the intensity gradient will be discontinuous at grid positions.
3.10.3 B-Spline Interpolation
The itk::BSplineInterpolateImageFunction represents the image intensity using B-spline
basis functions. When an input image is first connected to the interpolator, B-spline coefficients
are computed using recursive filtering (assuming mirror boundary conditions). Intensity at a nongrid position is computed by multiplying the B-spline coefficients with shifted B-spline kernels
within a small support region of the requested position. Figure 3.42 illustrates on the left how the
deformation values on the BSpline grid nodes are used for computing interpolated deformations in
the rest of space. Note for example that when a cubic BSpline is used, the grid must have one extra
node in one side of the image and two extra nodes on the other side, this along every dimension.
3.10. Interpolators
279
Figure 3.42: The left side illustrates the BSpline grid and the deformations that are known on those nodes.
The right side illustrates the region where interpolation is possible when the BSpline is of cubic order. The small
arrows represent deformation values that were interpolated from the grid deformations shown on the left side of
the diagram.
Currently, this interpolator supports splines of order 0 to 5. Using a spline of order 0 is almost
identical to nearest neighbor interpolation; a spline of order 1 is exactly identical to linear interpolation. For splines of order greater than 1, both the interpolated value and its derivative are spatially
continuous.
It is important to note that when using this scheme, the interpolated value may lie outside the range
of input image intensities. This is especially important when handling unsigned data, as it is possible
that the interpolated value is negative.
3.10.4 Windowed Sinc Interpolation
The itk::WindowedSincInterpolateImageFunction is the best possible interpolator for data
that have been digitized in a discrete grid. This interpolator has been developed based on Fourier
Analysis considerations. It is well known in signal processing that the process of sampling a spatial
function using a periodic discrete grid results in a replication of the spectrum of that signal in the
frequency domain.
The process of recovering the continuous signal from the discrete sampling is equivalent to the
removal of the replicated spectra in the frequency domain. This can be done by multiplying the
spectra with a box function that will set to zero all the frequencies above the highest frequency in
the original signal. Multiplying the spectrum with a box function is equivalent to convolving the
280
Chapter 3. Registration
spatial discrete signal with a sinc function
sinc(x) = sin (x)/x
(3.15)
The sinc function has infinite support, which of course in practice can not really be implemented.
Therefore, the sinc is usually truncated by multiplying it with a Window function. The Windowed
Sinc interpolator is the result of such an operation.
This interpolator presents a series of trade-offs in its utilization. Probably the most significant is
that the larger the window, the more precise will be the resulting interpolation. However, large
windows will also result in long computation times. Since the user can select the window size in this
interpolator, it is up to the user to determine how much interpolation quality is required in her/his
application and how much computation time can be justified. For details on the signal processing
theory behind this interpolator, please refer to Meijering et. al [41].
The region of the image used for computing the interpolator is determined by the window radius.
For example, in a 2D image where we want to interpolate the value at position (x, y) the following
computation will be performed.
I(x, y) =
⌊x⌋+m
∑
⌊y⌋+m
∑
i=⌊x⌋+1−m j=⌊y⌋+1−m
Ii, j K(x − i)K(y − j)
(3.16)
where m is the radius of the window. Typically, values such as 3 or 4 are reasonable for the window
radius. The function kernel K(t) is composed by the sinc function and one of the windows listed
above.
K(t) = w(t)sinc(t) = w(t)
sin(πt)
πt
(3.17)
Some of the windows that can be used with this interpolator are
Cosinus window
w(x) = cos(
πx
)
2m
Hamming window
w(x) = 0.54 + 0.46cos(
Welch window
πx
)
m
(3.19)
x2
)
m2
(3.20)
x
w(x) = sinc( )
m
(3.21)
w(x) = 1 − (
Lancos window
(3.18)
3.10. Interpolators
281
Blackman window
w(x) = 0.42 + 0.5cos(
πx
2πx
) + 0.08cos(
)
m
m
(3.22)
The window functions listed above are available inside the itk::Function namespace. The conclusions of the referenced paper suggest to use the Welch, Cosine, Kaiser, and Lancos windows for m
= 4,5. These are based on error in rotating medical images with respect to the linear interpolation
method. In some cases the results achieve a 20-fold improvement in accuracy.
This filter can be used in the same way you would use any ImageInterpolationFunction. For instance,
you can plug it into the ResampleImageFilter class. In order to instantiate the filter you must choose
several template parameters.
using InterpolatorType = WindowedSincInterpolateImageFunction<
TInputImage, VRadius, TWindowFunction,
TBoundaryCondition, TCoordRep >;
TInputImage is the image type, as for any other interpolator.
VRadius is the radius of the kernel, i.e., the m from the formula above.
TWindowFunction is the window function object, which you can choose from about five different
functions defined in this header. The default is the Hamming window, which is commonly used but
not optimal according to the cited paper.
TBoundaryCondition is the boundary condition class used to determine the values of pixels that fall off the image boundary. This class has the same meaning here as in the
itk::NeighborhoodIterator classes.
TCoordRep is again standard for interpolating functions, and should be float or double.
The WindowedSincInterpolateImageFunction is probably not the interpolator that you want to use
for performing registration. Its computation burden makes it too expensive for this purpose. The
best use of this interpolator is for the final resampling of the image, once the transform has been
found using another less expensive interpolator in the registration process.
282
Chapter 3. Registration
3.11 Metrics
Sigma
Gray levels
Figure 3.43: In Parzen windowing, a continuous
density function is constructed by superimposing
kernel functions (Gaussian function in this case)
centered on the intensity samples obtained from
the image.
In ITK, itk::ImageToImageMetricv4 objects quantitatively measure how well the
transformed moving image fits the fixed image by comparing the gray-scale intensity of
the images. These metrics are very flexible
and can work with any transform or interpolation method and do not require reduction of
the gray-scale images to sparse extracted information such as edges.
The metric component is perhaps the most
critical element of the registration framework.
The selection of which metric to use is highly
dependent on the registration problem to be
solved. For example, some metrics have a large
capture range while others require initialization
close to the optimal position. In addition, some
metrics are only suitable for comparing images obtained from the same imaging modality,
while others can handle inter-modality comparisons. Unfortunately, there are no clear-cut
rules as to how to choose a metric.
The matching Metric class controls most parts
of the registration process since it handles fixed, moving and virtual images as well as fixed and
moving transforms and interpolators. The method GetValue() can be used to evaluate the quantitative criterion at the transform parameters specified in the argument. Typically, the metric samples
points within a defined region of the virtual lattice. For each point, the corresponding fixed and moving image positions are computed using the fixed initial transform and the moving transform with
the specified parameters. Then, the fixed and moving interpolators are used to compute the fixed
and moving image’s intensities at the mapped positions. Details on this mapping are illustrated in
Figures 3.40 and 3.41 assuming that virtual lattice is the same as the fixed image lattice, which is
usually the case in practice.
The metrics also support region-based evaluation.
The SetFixedImageMask() and
SetMovingImageMask() methods may be used to restrict evaluation of the metric within a specified region. The masks may be of any type derived from itk::SpatialObject.
Besides the measure value, gradient-based optimization schemes also require derivatives of
the measure with respect to each transform parameter. The methods GetDerivatives() and
GetValueAndDerivatives() can be used to obtain the gradient information.
The following is the list of metrics currently available in ITKv4 registration framework:
3.11. Metrics
283
• Mean squares
itk::MeanSquaresImageToImageMetricv4
• Correlation
itk::CorrelationImageToImageMetricv4
• Mutual information by Mattes
itk::MattesMutualInformationImageToImageMetricv4
• Joint histogram mutual information
itk::JointHistogramMutualInformationHistogramImageToImageMetricv4
• Demons metric
itk::DemonsImageToImageMetricv4
• ANTS neighborhood correlation metric
itk::ANTSNeighborhoodCorrelationImageToImageMetricv4
Also, in case you are interested in using the legacy ITK registration framework, the following is the
list of metrics currently available in ITKv3:
• Mean squares
itk::MeanSquaresImageToImageMetric
• Normalized correlation
itk::NormalizedCorrelationImageToImageMetric
• Mean reciprocal squared difference
itk::MeanReciprocalSquareDifferenceImageToImageMetric
• Mutual information by Viola and Wells
itk::MutualInformationImageToImageMetric
• Mutual information by Mattes
itk::MattesMutualInformationImageToImageMetric
• Kullback Liebler distance metric by Kullback and Liebler
itk::KullbackLeiblerCompareHistogramImageToImageMetric
• Normalized mutual information
itk::NormalizedMutualInformationHistogramImageToImageMetric
• Mean squares histogram
itk::MeanSquaresHistogramImageToImageMetric
• Correlation coefficient histogram
itk::CorrelationCoefficientHistogramImageToImageMetric
284
Chapter 3. Registration
• Cardinality Match metric
itk::MatchCardinalityImageToImageMetric
• Kappa Statistics metric
itk::KappaStatisticImageToImageMetric
• Gradient Difference metric
itk::GradientDifferenceImageToImageMetric
In the following sections, we describe the ITKv4 metric types in detail. You can check ITK descriptions in doxygen for details about ITKv3 metric classes.
For ease of notation, we will refer to the fixed image f (X) and transformed moving image (m◦T (X))
as images A and B.
3.11.1 Mean Squares Metric
The itk::MeanSquaresImageToImageMetricv4 computes the mean squared pixel-wise difference in intensity between image A and B over a user defined region:
MS(A, B) =
1 N
∑ (Ai − Bi)2
N i=1
(3.23)
Ai is the i-th pixel of Image A
Bi is the i-th pixel of Image B
N is the number of pixels considered
The optimal value of the metric is zero. Poor matches between images A and B result in large values
of the metric. This metric is simple to compute and has a relatively large capture radius.
This metric relies on the assumption that intensity representing the same homologous point must be
the same in both images. Hence, its use is restricted to images of the same modality. Additionally,
any linear changes in the intensity result in a poor match value.
Exploring a Metric
Getting familiar with the characteristics of the Metric as a cost function is fundamental in order to
find the best way of setting up an optimization process that will use this metric for solving a registration problem. The following example illustrates a typical mechanism for studying the characteristics
of a Metric. Although the example is using the Mean Squares metric, the same methodology can be
applied to any of the other metrics available in the toolkit.
The source code for this section can be found in the file
MeanSquaresImageMetric1.cxx.
3.11. Metrics
285
This example illustrates how to explore the domain of an image metric. This is a useful exercise
before starting a registration process, since familiarity with the characteristics of the metric is fundamental for appropriate selection of the optimizer and its parameters used to drive the registration
process. This process helps identify how noisy a metric may be in a given range of parameters, and
it will also give an idea of the number of local minima or maxima in which an optimizer may get
trapped while exploring the parametric space.
We start by including the headers of the basic components: Metric, Transform and Interpolator.
#include "itkMeanSquaresImageToImageMetricv4.h"
#include "itkTranslationTransform.h"
#include "itkNearestNeighborInterpolateImageFunction.h"
We define the dimension and pixel type of the images to be used in the evaluation of the Metric.
constexpr unsigned int Dimension = 2;
using PixelType = float;
using ImageType = itk::Image< PixelType, Dimension >;
The type of the Metric is instantiated and one is constructed. In this case we decided to use the same
image type for both the fixed and the moving images.
using MetricType = itk::MeanSquaresImageToImageMetricv4<
ImageType, ImageType >;
MetricType::Pointer metric = MetricType::New();
We also instantiate the transform and interpolator types, and create objects of each class.
using TransformType = itk::TranslationTransform< double, Dimension >;
TransformType::Pointer transform = TransformType::New();
using InterpolatorType = itk::NearestNeighborInterpolateImageFunction<
ImageType, double >;
InterpolatorType::Pointer interpolator = InterpolatorType::New();
The classes required by the metric are connected to it. This includes the fixed and moving images,
the interpolator and the transform.
metric->SetTransform( transform );
metric->SetMovingInterpolator( interpolator );
286
Chapter 3. Registration
-60
Mean Squares Metric
-40
12000
10000
-20
8000
6000
4000
0
2000
0
20
60
40
20
-60
40
0
-40
-20
-20
0
-40
20
Translation in X (mm)
Translation in Y (mm)
40
60
60-60
-60
-40
-20
0
20
40
60
Figure 3.44: Plots of the Mean Squares Metric for an image compared to itself under multiple translations.
metric->SetFixedImage( fixedImage );
metric->SetMovingImage( movingImage );
Note that the SetTransform() method is equivalent to the SetMovingTransform() function. In
this example there is no need to use the SetFixedTransform(), since the virtual domain is assumed
to be the same as the fixed image domain set as following.
metric->SetVirtualDomainFromImage( fixedImage
);
Finally we select a region of the parametric space to explore. In this case we are using a translation
transform in 2D, so we simply select translations from a negative position to a positive position, in
both x and y. For each one of those positions we invoke the GetValue() method of the Metric.
MetricType::MovingTransformParametersType displacement( Dimension );
constexpr int rangex = 50;
constexpr int rangey = 50;
for( int dx = -rangex; dx <= rangex; dx++ )
{
for( int dy = -rangey; dy <= rangey; dy++ )
{
displacement[0] = dx;
displacement[1] = dy;
metric->SetParameters( displacement );
const double value = metric->GetValue();
std::cout << dx << "
" << dy << "
" << value << std::endl;
}
}
Running this code using the image BrainProtonDensitySlice.png as both the fixed and the moving
3.11. Metrics
287
images results in the plot shown in Figure 3.44. From this figure, it can be seen that a gradient-based
optimizer will be appropriate for finding the extrema of the Metric. It is also possible to estimate a
good value for the step length of a gradient-descent optimizer.
This exercise of plotting the Metric is probably the best thing to do when a registration process
is not converging and when it is unclear how to fine tune the different parameters involved in the
registration. This includes the optimizer parameters, the metric parameters and even options such as
preprocessing the image data with smoothing filters.
The shell and Gnuplot8 scripts used for generating the graphics in Figure 3.44 are available in the
directory
ITKSoftwareGuide/SoftwareGuide/Art
Of course, this plotting exercise becomes more challenging when the transform has more than three
parameters, and when those parameters have very different value ranges. In those cases it is necessary to select only a key subset of parameters from the transform and to study the behavior of the
metric when those parameters are varied.
3.11.2 Normalized Correlation Metric
The itk::CorrelationImageToImageMetricv4 computes pixel-wise cross-correlation and normalizes it by the square root of the autocorrelation of the images:
∑N (Ai · Bi )
NC(A, B) = −1 × q i=1
∑Ni=1 A2i · ∑Ni=1 B2i
(3.24)
Ai is the i-th pixel of Image A
Bi is the i-th pixel of Image B
N is the number of pixels considered
Note the −1 factor in the metric computation. This factor is used to make the metric be optimal when
its minimum is reached. The optimal value of the metric is then minus one. Misalignment between
the images results in small measure values. The use of this metric is limited to images obtained
using the same imaging modality. The metric is insensitive to multiplicative factors between the two
images. This metric produces a cost function with sharp peaks and well-defined minima. On the
other hand, it has a relatively small capture radius.
3.11.3 Mutual Information Metric
The itk::MattesMutualInformationImageToImageMetricv4 computes the mutual information
between image A and image B. Mutual information (MI) measures how much information one
8 http://www.gnuplot.info
288
Chapter 3. Registration
random variable (image intensity in one image) tells about another random variable (image intensity
in the other image). The major advantage of using MI is that the actual form of the dependency does
not have to be specified. Therefore, complex mapping between two images can be modeled. This
flexibility makes MI well suited as a criterion of multi-modality registration [46].
Mutual information is defined in terms of entropy. Let
H(A) = −
Z
pA (a) log pA (a) da
(3.25)
be the entropy of random variable A, H(B) the entropy of random variable B and
H(A, B) =
Z
pAB (a, b) log pAB (a, b) da db
(3.26)
be the joint entropy of A and B. If A and B are independent, then
pAB (a, b) = pA (a)pB (b)
(3.27)
H(A, B) = H(A) + H(B).
(3.28)
and
However, if there is any dependency, then
H(A, B) < H(A) + H(B).
(3.29)
The difference is called Mutual Information : I(A, B)
I(A, B) = H(A) + H(B) − H(A, B)
(3.30)
Parzen Windowing
In a typical registration problem, direct access to the marginal and joint probability densities is not
available and hence the densities must be estimated from the image data. Parzen windows (also
known as kernel density estimators) can be used for this purpose. In this scheme, the densities are
constructed by taking intensity samples S from the image and super-positioning kernel functions
K(·) centered on the elements of S as illustrated in Figure 3.43:
A variety of functions can be used as the smoothing kernel with the requirement that they are smooth,
symmetric, have zero mean and integrate to one. For example, boxcar, Gaussian and B-spline functions are suitable candidates. A smoothing parameter is used to scale the kernel function. The larger
the smoothing parameter, the wider the kernel function used and hence the smoother the density
estimate. If the parameter is too large, features such as modes in the density will get smoothed out.
On the other hand, if the smoothing parameter is too small, the resulting density may be too noisy.
The estimation is given by the following equation.
p(a) ≈ P∗ (a) =
1
K (a − s j )
N s∑
j ∈S
(3.31)
3.11. Metrics
289
Choosing the optimal smoothing parameter is a difficult research problem and beyond the scope of
this software guide. Typically, the optimal value of the smoothing parameter will depend on the data
and the number of samples used.
Mattes et al. Implementation
The implementation of mutual information metric available in ITKv4 follows
the method specified by Mattes et al.
in [39] and is implemented by the
itk::MattesMutualInformationImageToImageMetricv4 class.
In this implementation, only one set of intensity samples is drawn from the image. Using this set,
the marginal and joint probability density function (PDF) is evaluated at discrete positions or bins
uniformly spread within the dynamic range of the images. Entropy values are then computed by
summing over the bins.
The number of spatial samples used is a ratio of the total number of samples and is set using the SetMetricSamplingPercentage() method directly from the registration framework
itk::ImageRegistrationMethodv4. Also, The number of bins used to compute the entropy values is set in the metric class via the SetNumberOfHistogramBins() method.
Since the fixed image PDF does not contribute to the metric derivatives, it does not need to be
smooth. Hence, a zero-order (boxcar) B-spline kernel is used for computing the PDF. On the other
hand, to ensure smoothness, a third-order B-spline kernel is used to compute the moving image
intensity PDF. The advantage of using a B-spline kernel over a Gaussian kernel is that the B-spline
kernel has a finite support region. This is computationally attractive, as each intensity sample only
affects a small number of bins and hence does not require a N × N loop to compute the metric value.
During the PDF calculations, the image intensity values are linearly scaled to have a minimum of
zero and maximum of one. This rescaling means that a fixed B-spline kernel bandwidth of one can
be used to handle image data with arbitrary magnitude and dynamic range.
3.11.4 Normalized Mutual Information Metric
Given two images, A and B, the normalized mutual information may be computed as
NMI(A, B) = 1 +
I(A, B)
H(A) + H(B)
=
H(A, B)
H(A, B)
(3.32)
where the entropy of the images, H(A), H(B), the mutual information, I(A, B) and the joint entropy
H(A, B) are computed as mentioned in 3.11.3. Details of the implementation may be found in [23].
290
Chapter 3. Registration
3.11.5 Demons metric
The implementation of the itk::DemonsImageToImageMetricv4 metric is taken from
itk::DemonsRegistrationFunction.
The metric derivative can be calculated using image derivatives either from the fixed or moving
images. The default is to use fixed-image gradients. See ObjectToObjectMetric::SetGradientSource
to change this behavior.
An intensity threshold is used, below which image pixels are considered equal for
the purpose of derivative calculation.
The threshold can be changed by calling
SetIntensityDifferenceThreshold.
Note that this metric supports only moving transforms with local support and with a number of
local parameters that match the moving image dimension. In particular, it’s meant to be used with
itk::DisplacementFieldTransform and derived classes.
3.11.6 ANTS neighborhood correlation metric
The itk::ANTSNeighborhoodCorrelationImageToImageMetricv4 metric computes normalized cross correlation using a small neighborhood for each voxel between two images, with speed
optimizations for dense registration.
Around each voxel, the neighborhood is defined as a N-Dimensional rectangle centered at the voxel.
The size of the rectangle is 2*radius+1. Normalized correlation between neighborhoods of the fixed
image and the moving image are averaged over the whole image as the final metric. A radius less
than 2 can be unstable. 2 is the default.
3.12. Optimizers
291
3.12 Optimizers
Optimization algorithms are encapsulated as itk::ObjectToObjectOptimizer objects within
ITKv4. Optimizers are generic and can be used for applications other than registration. Within
the registration framework, subclasses of itk::SingleValuedNonLinearVnlOptimizerv4 are
implemented as a wrap around already implemented vnl classes.
The basic input to an optimizer is a cost function or metric object. In the context of
registration, itk::ImageToImageMetricv4 classes provide this functionality. The metric is set using SetInitialPosition() and the optimization algorithm is invoked by
StartOptimization(). Once the optimization has finished, the final parameters can be obtained
using GetCurrentPosition().
Some optimizers also allow rescaling of their individual parameters. This is convenient for normalizing parameter spaces where some parameters have different dynamic ranges. For example,
the first parameter of itk::Euler2DTransform represents an angle while the last two parameters
represent translations. A unit change in angle has a much greater impact on an image than a unit
change in translation. This difference in scale appears as long narrow valleys in the search space
making the optimization problem more difficult. Rescaling the translation parameters can help to fix
this problem. Scales are represented as an itk::Array of doubles and set using SetScales().
Estimating the scales parameters can also be done automatically using the
itk::OptimizerParameterScalesEstimatorTemplate and its subclasses. The scales estimator object is then set to the optimizer via SetScalesEstimator().
Despite the old version of ITK, there are only Single Valued types of optimizers available in ITKv4,
which are suitable for dealing with cost functions that return a single value. These are indeed the
most common type of cost functions, and are also known as Single Valued functions.
The types of single valued optimizers currently available in ITKv4 are:
• Amoeba: Nelder-Meade downhill simplex. This optimizer is actually implemented in the
vxl/vnl numerics toolkit. The ITK class itk::AmoebaOptimizerv4 is merely an adaptor
class.
• Gradient Descent: Advances parameters in the direction of the gradient where the step size
is governed by a learning rate ( itk::GradientDescentOptimizerv4).
• Gradient Descent Line Search: Gradient descent with a golden section line search.
itk::GradientDescentLineSearchOptimizerv4 implements a simple gradient descent
optimizer that is followed by a line search to find the best value for the learning rate.
• Conjugate Gradient Descent Line Search: Advances parameters in the direction of the
Polak-Ribiere conjugate gradient where a line search is used to find the best value for the
learning rate ( itk::ConjugateGradientLineSearchOptimizerv4).
• Quasi Newton: Implements a Quasi-Newton optimizer with BFGS Hessian estimation. Second order approximation of the cost function is usually more efficient since it estimates the
292
Chapter 3. Registration
Figure 3.45: Class diagram of the optimizersv4 hierarchy.
3.12. Optimizers
293
descent or ascent direction more precisely. However, computation of Hessian is usually expensive or unavailable. Alternatively Quasi-Newton methods can estimate a Hessian from the
gradients in previous steps. Here a specific Quasi-Newton method, BFGS, is used to compute
the Quasi-Newton steps ( itk::QuasiNewtonOptimizerv4).
• LBFGS: Limited memory Broyden, Fletcher, Goldfarb and Shannon minimization. It is an
adaptor to an optimizer in vnl ( itk::LBFGSOptimizerv4).
• LBFGSB: A modified version of the LBFGS optimizer that allows to specify bounds for the
parameters in the search space. It is an adaptor to an optimizer in netlib. Details on this
optimizer can be found in [10, 75] ( itk::LBFGSBOptimizerv4).
• One Plus One Evolutionary: Strategy that simulates the biological evolution of a set of
samples in the search space. This optimizer is mainly used in the process of bias correction
of MRI images ( itk::OnePlusOneEvolutionaryOptimizerv4). Details on this optimizer
can be found in [59].
• Regular Step Gradient Descent:
Advances parameters in the direction of
the gradient where a bipartition scheme is used to compute the step size (
itk::RegularStepGradientDescentOptimizerv4). This optimizer is also used for
Versor transforms parameters, where the current rotation is composed with the gradient
rotation to produce the new rotation versor. The translational part of the transform parameters
are updated as usually done in a vector space. It follows the definition of versor gradients
defined by Hamilton [24]
• Powell Optimizer: Powell optimization method. For an N-dimensional parameter space,
each iteration minimizes(maximizes) the function in N (initially orthogonal) directions. This
optimizer is described in [49]. ( itk::PowellOptimizerv4).
• Exhausive Optimizer: Fully samples a grid on the parameteric space. This optimizer is
equivalent to an exahaustive search in a discrete grid defined over the parametric space.
The grid is centered on the initial position. The subdivisions of the grid along each one
of the dimensions of the parametric space is defined by an array of number of steps (
itk::ExhaustiveOptimizerv4).
Figure 3.45 illustrates the full class hierarchy of optimizers in ITK. Optimizers in the lower right
corner are adaptor classes to optimizers existing in the vxl/vnl numerics toolkit. The optimizers
interact with the itk::CostFunction class. In the registration framework this cost function is
reimplemented in the form of ImageToImageMetric.
3.12.1 Registration using the One plus One Evolutionary Optimizer
The source code for this section can be found in the file
ImageRegistration11.cxx.
294
Chapter 3. Registration
This example illustrates how to combine the MutualInformation metric with an Evolutionary algorithm for optimization. Evolutionary algorithms are naturally well-suited for optimizing the Mutual
Information metric given its random and noisy behavior.
The structure of the example is almost identical to the one illustrated in ImageRegistration4. Therefore we focus here on the setup that is specifically required for the evolutionary optimizer.
#include
#include
#include
#include
#include
"itkImageRegistrationMethodv4.h"
"itkTranslationTransform.h"
"itkMattesMutualInformationImageToImageMetricv4.h"
"itkOnePlusOneEvolutionaryOptimizerv4.h"
"itkNormalVariateGenerator.h"
In this example the image types and all registration components, except the metric, are declared as
in Section 3.2. The Mattes mutual information metric type is instantiated using the image types.
using MetricType = itk::MattesMutualInformationImageToImageMetricv4<
FixedImageType,
MovingImageType >;
The histogram bins metric parameter is set as follows.
metric->SetNumberOfHistogramBins( 20 );
As our previous discussion in section 3.5.1, only a subsample of the virtual domain is needed to
evaluate the metric. The number of spatial samples to be used depends on the content of the image,
and the user can define the sampling percentage and the way that sampling operation is managed by
the registration framework as follows. Sampling startegy can can be defined as REGULAR or RANDOM,
while the default value is NONE.
registration->SetMetricSamplingPercentage( samplingPercentage );
RegistrationType::MetricSamplingStrategyType samplingStrategy =
RegistrationType::RANDOM;
registration->SetMetricSamplingStrategy( samplingStrategy );
Evolutionary algorithms are based on testing random variations of parameters. In order to support the computation of random values, ITK provides a family of random number generators. In
this example, we use the itk::NormalVariateGenerator which generates values with a normal
distribution.
using GeneratorType = itk::Statistics::NormalVariateGenerator;
GeneratorType::Pointer generator = GeneratorType::New();
3.12. Optimizers
295
The random number generator must be initialized with a seed.
generator->Initialize(12345);
Now we set the optimizer parameters.
optimizer->SetNormalVariateGenerator( generator );
optimizer->Initialize( 10 );
optimizer->SetEpsilon( 1.0 );
optimizer->SetMaximumIteration( 4000 );
This example is executed using the same multi-modality images as in the previous one. The registration converges after 24 iterations and produces the following results:
Translation X = 13.1719
Translation Y = 16.9006
These values are a very close match to the true misalignment introduced in the moving image.
3.12.2 Registration using masks constructed with Spatial objects
The source code for this section can be found in the file
ImageRegistration12.cxx.
This example illustrates the use of SpatialObjects as masks for selecting the pixels that should
contribute to the computation of Image Metrics. This example is almost identical to ImageRegistration6 with the exception that the SpatialObject masks are created and passed to the image
metric.
The most important header in this
itk::ImageMaskSpatialObject class.
example
is
the
one
corresponding
to
the
#include "itkImageMaskSpatialObject.h"
Here we instantiate the type of the itk::ImageMaskSpatialObject using the same dimension of
the images to be registered.
using MaskType = itk::ImageMaskSpatialObject< Dimension >;
Then we use the type for creating the spatial object mask that will restrict the registration to a reduced
region of the image.
296
Chapter 3. Registration
MaskType::Pointer spatialObjectMask = MaskType::New();
The mask in this case is read from a binary file using the ImageFileReader instantiated for an
unsigned char pixel type.
using ImageMaskType = itk::Image< unsigned char, Dimension >;
using MaskReaderType = itk::ImageFileReader< ImageMaskType >;
The reader is constructed and a filename is passed to it.
MaskReaderType::Pointer maskReader = MaskReaderType::New();
maskReader->SetFileName( argv[3] );
As usual, the reader is triggered by invoking its Update() method. Since this may eventually throw
an exception, the call must be placed in a try/catch block. Note that a full fledged application will
place this try/catch block at a much higher level, probably under the control of the GUI.
try
{
maskReader->Update();
}
catch( itk::ExceptionObject & err )
{
std::cerr << "ExceptionObject caught !" << std::endl;
std::cerr << err << std::endl;
return EXIT_FAILURE;
}
The output of the mask reader is connected as input to the ImageMaskSpatialObject.
spatialObjectMask->SetImage( maskReader->GetOutput() );
Finally, the spatial object mask is passed to the image metric.
metric->SetFixedImageMask( spatialObjectMask );
Let’s execute this example over some of the images provided in Examples/Data, for example:
• BrainProtonDensitySliceBorder20.png
• BrainProtonDensitySliceR10X13Y17.png
3.12. Optimizers
297
The second image is the result of intentionally rotating the first image by 10 degrees and shifting it
13mm in X and 17mm in Y . Both images have unit-spacing and are shown in Figure 3.14.
The registration converges after 20 iterations and produces the following results:
Angle (radians)
Angle (degrees)
Translation X =
Translation Y =
0.174712
10.0103
12.4521
16.0765
These values are a very close match to the true misalignments introduced in the moving image.
Now we resample the moving image using the transform resulting from the registration process.
TransformType::MatrixType matrix = transform->GetMatrix();
TransformType::OffsetType offset = transform->GetOffset();
std::cout << "Matrix = " << std::endl << matrix << std::endl;
std::cout << "Offset = " << std::endl << offset << std::endl;
3.12.3 Rigid registrations incorporating prior knowledge
The source code for this section can be found in the file
ImageRegistration13.cxx.
This example illustrates how to do registration with a 2D Rigid Transform and with MutualInformation metric.
#include "itkMattesMutualInformationImageToImageMetricv4.h"
The Euler2DTransform applies a rigid transform in 2D space.
using TransformType = itk::Euler2DTransform< double >;
using MetricType = itk::MattesMutualInformationImageToImageMetricv4<
FixedImageType,
MovingImageType >;
298
Chapter 3. Registration
metric->SetNumberOfHistogramBins( 20 );
double samplingPercentage = 0.20;
registration->SetMetricSamplingPercentage( samplingPercentage );
RegistrationType::MetricSamplingStrategyType samplingStrategy =
RegistrationType::RANDOM;
registration->SetMetricSamplingStrategy( samplingStrategy );
The itk::Euler2DTransform is initialized with 3 parameters, indicating the angle of rotation and the translation to be applied after rotation. The initialization is done by the
itk::CenteredTransformInitializer. The transform initializer can operate in two modes, the
first of which assumes that the anatomical objects to be registered are centered in their respective
images. Hence the best initial guess for the registration is the one that superimposes those two centers. This second approach assumes that the moments of the anatomical objects are similar for both
images and hence the best initial guess for registration is to superimpose both mass centers. The
center of mass is computed from the moments obtained from the gray level values. Here we adopt
the first approach. The GeometryOn() method toggles between the approaches.
using TransformInitializerType = itk::CenteredTransformInitializer<
TransformType,
FixedImageType,
MovingImageType >;
TransformInitializerType::Pointer initializer
= TransformInitializerType::New();
initializer->SetTransform( transform );
initializer->SetFixedImage( fixedImageReader->GetOutput() );
initializer->SetMovingImage( movingImageReader->GetOutput() );
initializer->GeometryOn();
initializer->InitializeTransform();
The optimizer scales the metrics (the gradient in this case) by the scales during each iteration. Here
we assume that the fixed and moving images are likely to be related by a translation.
using OptimizerScalesType = OptimizerType::ScalesType;
OptimizerScalesType optimizerScales( transform->GetNumberOfParameters() );
const double translationScale = 1.0 / 128.0;
optimizerScales[0] = 1.0;
optimizerScales[1] = translationScale;
optimizerScales[2] = translationScale;
optimizer->SetScales( optimizerScales );
optimizer->SetLearningRate( 0.5 );
3.12. Optimizers
299
optimizer->SetMinimumStepLength( 0.0001 );
optimizer->SetNumberOfIterations( 400 );
Let’s execute this example over some of the images provided in Examples/Data, for example:
• BrainProtonDensitySlice.png
• BrainProtonDensitySliceR10X13Y17.png
The second image is the result of intentionally rotating the first image by 10 degrees and shifting
it 13mm in X and 17mm in Y . Both images have unit-spacing and are shown in Figure 3.14. The
example yielded the following results.
Angle (radians)
Angle (degrees)
Translation X =
Translation Y =
0.174569
10.0021
13.0958
15.9156
These values match the true misalignment introduced in the moving image.
300
Chapter 3. Registration
Figure 3.46: Checkerboard comparisons before and after FEM-based deformable registration.
3.13 Deformable Registration
3.13.1 FEM-Based Image Registration
The source code for this section can be found in the file
DeformableRegistration1.cxx.
The finite element (FEM) library within the Insight Toolkit can be used to solve deformable image
registration problems. The first step in implementing a FEM-based registration is to include the
appropriate header files.
#include "itkFEMRegistrationFilter.h"
Next, we use using type alias to instantiate all necessary classes. We define the image and element
types we plan to use to solve a two-dimensional registration problem. We define multiple element
types so that they can be used without recompiling the code.
using
using
using
using
using
DiskImageType = itk::Image;
ImageType = itk::Image;
ElementType = itk::fem::Element2DC0LinearQuadrilateralMembrane;
ElementType2 = itk::fem::Element2DC0LinearTriangularMembrane;
FEMObjectType = itk::fem::FEMObject<2>;
Note that in order to solve a three-dimensional registration problem, we would simply define 3D
image and element types in lieu of those above. The following declarations could be used for a 3D
problem:
3.13. Deformable Registration
using
using
using
using
using
301
FileImage3DType = itk::Image;
Image3DType = itk::Image;
Element3DType = itk::fem::Element3DC0LinearHexahedronMembrane;
Element3DType2 = itk::fem::Element3DC0LinearTetrahedronMembrane;
FEMObject3DType = itk::fem::FEMObject<3>;
Once all the necessary components have been instantiated, we can instantiate the
itk::FEMRegistrationFilter, which depends on the image input and output types.
using RegistrationType = itk::fem::FEMRegistrationFilter<
ImageType,ImageType,FEMObjectType>;
In order to begin the registration, we declare an instance of the FEMRegistrationFilter and set
its parameters. For simplicity, we will call it registrationFilter.
RegistrationType::Pointer registrationFilter = RegistrationType::New();
registrationFilter->SetMaxLevel(1);
registrationFilter->SetUseNormalizedGradient( true );
registrationFilter->ChooseMetric( 0 );
unsigned int maxiters = 20;
float
E = 100;
float
p = 1;
registrationFilter->SetElasticity(E, 0);
registrationFilter->SetRho(p, 0);
registrationFilter->SetGamma(1., 0);
registrationFilter->SetAlpha(1.);
registrationFilter->SetMaximumIterations( maxiters, 0 );
registrationFilter->SetMeshPixelsPerElementAtEachResolution(4, 0);
registrationFilter->SetWidthOfMetricRegion(1, 0);
registrationFilter->SetNumberOfIntegrationPoints(2, 0);
registrationFilter->SetDoLineSearchOnImageEnergy( 0 );
registrationFilter->SetTimeStep(1.);
registrationFilter->SetEmployRegridding(false);
registrationFilter->SetUseLandmarks(false);
In order to initialize the mesh of elements, we must first create “dummy” material and element
objects and assign them to the registration filter. These objects are subsequently used to either read
a predefined mesh from a file or generate a mesh using the software. The values assigned to the
fields within the material object are arbitrary since they will be replaced with those specified earlier.
Similarly, the element object will be replaced with those from the desired mesh.
// Create the material properties
itk::fem::MaterialLinearElasticity::Pointer m;
m = itk::fem::MaterialLinearElasticity::New();
m->SetGlobalNumber(0);
// Young's modulus of the membrane
m->SetYoungsModulus(registrationFilter->GetElasticity());
302
m->SetCrossSectionalArea(1.0);
m->SetThickness(1.0);
m->SetMomentOfInertia(1.0);
m->SetPoissonsRatio(0.);
m->SetDensityHeatProduct(1.0);
Chapter 3. Registration
//
//
//
//
//
Cross-sectional area
Thickness
Moment of inertia
Poisson's ratio -- DONT CHOOSE 1.0!!
Density-Heat capacity product
// Create the element type
ElementType::Pointer e1=ElementType::New();
e1->SetMaterial(m);
registrationFilter->SetElement(e1);
registrationFilter->SetMaterial(m);
Now we are ready to run the registration:
registrationFilter->RunRegistration();
To output the image resulting from the registration, we can call GetWarpedImage(). The image is
written in floating point format.
itk::ImageFileWriter::Pointer warpedImageWriter;
warpedImageWriter = itk::ImageFileWriter::New();
warpedImageWriter->SetInput( registrationFilter->GetWarpedImage() );
warpedImageWriter->SetFileName("warpedMovingImage.mha");
try
{
warpedImageWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
We can also output the displacement field resulting from the registration; we can call
GetDisplacementField() to get the multi-component image.
using DispWriterType = itk::ImageFileWriter;
DispWriterType::Pointer dispWriter = DispWriterType::New();
dispWriter->SetInput( registrationFilter->GetDisplacementField() );
dispWriter->SetFileName("displacement.mha");
try
{
dispWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
3.13. Deformable Registration
303
Figure 3.46 presents the results of the FEM-based deformable registration applied to two timeseparated slices of a living rat dataset. Checkerboard comparisons of the two images are shown
before registration (left) and after registration (right). Both images were acquired from the same
living rat, the first after inspiration of air into the lungs and the second after exhalation. Deformation
occurs due to the relaxation of the diaphragm and the intercostal muscles, both of which exert force
on the lung tissue and cause air to be expelled.
The following is a documented sample parameter file that can be used with this deformable registration example. This example demonstrates the setup of a basic registration problem that does not use
multi-resolution strategies. As a result, only one value for the parameters between (# of pixels
per element) and (maximum iterations) is necessary. In order to use a multi-resolution strategy, you would have to specify values for those parameters at each level of the pyramid.
3.13.2 BSplines Image Registration
The source code for this section can be found in the file
DeformableRegistration4.cxx.
This example illustrates the use of the itk::BSplineTransform class for performing registration
of two 2D images in an ITKv4 registration framework. Due to the large number of parameters of the
BSpline transform, we will use a itk::LBFGSOptimizerv4 instead of a simple steepest descent or
a conjugate gradient descent optimizer.
The following are the most relevant headers to this example.
#include "itkBSplineTransform.h"
#include "itkLBFGSOptimizerv4.h"
The parameter space of the BSplineTransform is composed by the set of all the deformations
associated with the nodes of the BSpline grid. This large number of parameters makes it possible to
represent a wide variety of deformations, at the cost of requiring a significant amount of computation
time.
We instantiate now the type of the BSplineTransform using as template parameters the type for
coordinates representation, the dimension of the space, and the order of the BSpline.
const unsigned int SpaceDimension = ImageDimension;
constexpr unsigned int SplineOrder = 3;
using CoordinateRepType = double;
using TransformType = itk::BSplineTransform<
CoordinateRepType,
SpaceDimension,
SplineOrder >;
304
Chapter 3. Registration
The transform object is constructed below.
TransformType::Pointer
transform
= TransformType::New();
Fixed parameters of the BSpline transform should be defined before the registration. These parameters define origin, dimension, direction and mesh size of the transform grid and are set based on
specifications of the fixed image space lattice. We can use itk::BSplineTransformInitializer
to initialize fixed parameters of a BSpline transform.
using InitializerType = itk::BSplineTransformInitializer<
TransformType,
FixedImageType>;
InitializerType::Pointer transformInitializer = InitializerType::New();
unsigned int numberOfGridNodesInOneDimension = 8;
TransformType::MeshSizeType
meshSize;
meshSize.Fill( numberOfGridNodesInOneDimension - SplineOrder );
transformInitializer->SetTransform( transform );
transformInitializer->SetImage( fixedImage );
transformInitializer->SetTransformDomainMeshSize( meshSize );
transformInitializer->InitializeTransform();
After setting the fixed parameters of the transform, we set the initial transform to be an identity
transform. It is like setting all the transform parameters to zero in created parameter space.
transform->SetIdentity();
Then, the initialized transform is connected to the registration object and is set to be optimized
directly during the registration process.
Calling InPlaceOn() means that the current initialized transform will optimized directly and is
grafted to the output, so it can be considered as the output transform object. Otherwise, the initial
transform will be copied or “cloned” to the output transform object, and the copied object will be
optimized during the registration process.
registration->SetInitialTransform( transform );
registration->InPlaceOn();
The itk::RegistrationParameterScalesFromPhysicalShift class is used to estimate the parameters scales before we set the optimizer.
3.13. Deformable Registration
305
using ScalesEstimatorType =
itk::RegistrationParameterScalesFromPhysicalShift;
ScalesEstimatorType::Pointer scalesEstimator = ScalesEstimatorType::New();
scalesEstimator->SetMetric( metric );
scalesEstimator->SetTransformForward( true );
scalesEstimator->SetSmallParameterVariation( 1.0 );
Now the scale estimator is passed to the itk::LBFGSOptimizerv4, and we set other parameters of
the optimizer as well.
optimizer->SetGradientConvergenceTolerance( 5e-2 );
optimizer->SetLineSearchAccuracy( 1.2 );
optimizer->SetDefaultStepLength( 1.5 );
optimizer->TraceOn();
optimizer->SetMaximumNumberOfFunctionEvaluations( 1000 );
optimizer->SetScalesEstimator( scalesEstimator );
Let’s execute this example using the rat lung images from the previous examples.
• RatLungSlice1.mha
• RatLungSlice2.mha
The transform object is updated during the registration process and is passed to the resampler to
map the moving image space onto the fixed image space.
OptimizerType::ParametersType finalParameters = transform->GetParameters();
3.13.3 Level Set Motion for Deformable Registration
The source code for this section can be found in the file
DeformableRegistration5.cxx.
This example demonstrates how to use the level set motion to deformably register two images. The
first step is to include the header files.
#include
#include
#include
#include
"itkLevelSetMotionRegistrationFilter.h"
"itkHistogramMatchingImageFilter.h"
"itkCastImageFilter.h"
"itkWarpImageFilter.h"
Second, we declare the types of the images.
306
Chapter 3. Registration
constexpr unsigned int Dimension = 2;
using PixelType = unsigned short;
using FixedImageType = itk::Image< PixelType, Dimension >;
using MovingImageType = itk::Image< PixelType, Dimension >;
Image file readers are set up in a similar fashion to previous examples. To support the re-mapping
of the moving image intensity, we declare an internal image type with a floating point pixel type and
cast the input images to the internal image type.
using InternalPixelType = float;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
using FixedImageCasterType = itk::CastImageFilter< FixedImageType,
InternalImageType >;
using MovingImageCasterType = itk::CastImageFilter< MovingImageType,
InternalImageType >;
FixedImageCasterType::Pointer fixedImageCaster = FixedImageCasterType::New();
MovingImageCasterType::Pointer movingImageCaster
= MovingImageCasterType::New();
fixedImageCaster->SetInput( fixedImageReader->GetOutput() );
movingImageCaster->SetInput( movingImageReader->GetOutput() );
The level set motion algorithm relies on the assumption that pixels representing the same homologous point on an object have the same intensity on both the fixed and moving images to be registered.
In this example, we will preprocess the moving image to match the intensity between the images
using the itk::HistogramMatchingImageFilter.
The basic idea is to match the histograms of the two images at a user-specified number of quantile
values. For robustness, the histograms are matched so that the background pixels are excluded from
both histograms. For MR images, a simple procedure is to exclude all gray values smaller than the
mean gray value of the image.
using MatchingFilterType = itk::HistogramMatchingImageFilter<
InternalImageType,
InternalImageType >;
MatchingFilterType::Pointer matcher = MatchingFilterType::New();
For this example, we set the moving image as the source or input image and the fixed image as the
reference image.
matcher->SetInput( movingImageCaster->GetOutput() );
matcher->SetReferenceImage( fixedImageCaster->GetOutput() );
We then select the number of bins to represent the histograms and the number of points or quantile
values where the histogram is to be matched.
3.13. Deformable Registration
307
matcher->SetNumberOfHistogramLevels( 1024 );
matcher->SetNumberOfMatchPoints( 7 );
Simple background extraction is done by thresholding at the mean intensity.
matcher->ThresholdAtMeanIntensityOn();
In the itk::LevelSetMotionRegistrationFilter, the deformation field is represented as an
image whose pixels are floating point vectors.
using VectorPixelType = itk::Vector< float, Dimension >;
using DisplacementFieldType = itk::Image< VectorPixelType, Dimension >;
using RegistrationFilterType = itk::LevelSetMotionRegistrationFilter<
InternalImageType,
InternalImageType,
DisplacementFieldType>;
RegistrationFilterType::Pointer filter = RegistrationFilterType::New();
The input fixed image is simply the output of the fixed image casting filter. The input moving image
is the output of the histogram matching filter.
filter->SetFixedImage( fixedImageCaster->GetOutput() );
filter->SetMovingImage( matcher->GetOutput() );
The level set motion registration filter has two parameters: the number of iterations to be performed
and the standard deviation of the Gaussian smoothing kernel to be applied to the image prior to
calculating gradients.
filter->SetNumberOfIterations( 50 );
filter->SetGradientSmoothingStandardDeviations(4);
The registration algorithm is triggered by updating the filter. The filter output is the computed
deformation field.
filter->Update();
The itk::WarpImageFilter can be used to warp the moving image with the output deformation
field. Like the itk::ResampleImageFilter, the WarpImageFilter requires the specification of the
input image to be resampled, an input image interpolator, and the output image spacing and origin.
using WarperType = itk::WarpImageFilter<
MovingImageType,
MovingImageType,
308
Chapter 3. Registration
DisplacementFieldType >;
using InterpolatorType = itk::LinearInterpolateImageFunction<
MovingImageType,
double
>;
WarperType::Pointer warper = WarperType::New();
InterpolatorType::Pointer interpolator = InterpolatorType::New();
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
warper->SetInput( movingImageReader->GetOutput() );
warper->SetInterpolator( interpolator );
warper->SetOutputSpacing( fixedImage->GetSpacing() );
warper->SetOutputOrigin( fixedImage->GetOrigin() );
warper->SetOutputDirection( fixedImage->GetDirection() );
Unlike the ResampleImageFilter, the WarpImageFilter warps or transforms the input image with respect to the deformation field represented by an image of vectors. The resulting warped or resampled
image is written to file as per previous examples.
warper->SetDisplacementField( filter->GetOutput() );
Let’s execute this example using the rat lung data from the previous example. The associated data
files can be found in Examples/Data:
• RatLungSlice1.mha
• RatLungSlice2.mha
The result of the demons-based deformable registration is presented in Figure 3.47. The checkerboard comparison shows that the algorithm was able to recover the misalignment due to expiration.
It may be also desirable to write the deformation field as an image of vectors. This can be done with
the following code.
using FieldWriterType = itk::ImageFileWriter< DisplacementFieldType >;
FieldWriterType::Pointer fieldWriter = FieldWriterType::New();
fieldWriter->SetFileName( argv[4] );
fieldWriter->SetInput( filter->GetOutput() );
fieldWriter->Update();
Note that the file format used for writing the deformation field must be capable of representing
multiple components per pixel. This is the case for the MetaImage and VTK file formats.
3.13. Deformable Registration
309
Figure 3.47: Checkerboard comparisons before and after demons-based deformable registration.
3.13.4 BSplines Multi-Grid Image Registration
The source code for this section can be found in the file
DeformableRegistration6.cxx.
This example illustrates the use of the itk::BSplineTransform class in a multi-resolution scheme.
Here we run 3 levels of resolutions. The first level of registration is performed with the spline grid
of low resolution. Then, a common practice is to increase the resolution of the B-spline mesh (or,
analogously, the control point grid size) at each level.
For this purpose, we introduce the concept of transform adaptors. Each level of each
stage is defined by a transform adaptor which describes how to adapt the transform to
the current level by increasing the resolution from the previous level.
Here, we used
itk::BSplineTransformParametersAdaptor class to adapt the BSpline transform parameters
at each resolution level. Note that for many transforms, such as affine, the concept of an adaptor
may be nonsensical since the number of transform parameters does not change between resolution
levels.
This examples use the itk::LBFGS2Optimizerv4, which is the new implementation of the quasiNewtown unbounded limited-memory Broyden Fletcher Goldfarb Shannon (LBFGS) optimizer.
The unbounded version does not require specification of the bounds of the parameters space, since
the number of parameters change at each B-Spline resolution this implementation is preferred.
Since this example is quite similar to the previous example on the use of the BSplineTransform
we omit most of the details already discussed and will focus on the aspects related to the multiresolution approach.
We include the header files for the transform, optimizer and adaptor.
310
Chapter 3. Registration
#include "itkBSplineTransform.h"
#include "itkLBFGS2Optimizerv4.h"
#include "itkBSplineTransformParametersAdaptor.h"
We instantiate the type of the BSplineTransform using as template parameters the type for coordinates representation, the dimension of the space, and the order of the BSpline.
const unsigned int SpaceDimension = ImageDimension;
constexpr unsigned int SplineOrder = 3;
using CoordinateRepType = double;
using TransformType = itk::BSplineTransform<
CoordinateRepType,
SpaceDimension,
SplineOrder >;
We construct the transform object, initialize its parameters and connect that to the registration object.
TransformType::Pointer outputBSplineTransform = TransformType::New();
// Initialize the fixed parameters of transform (grid size, etc).
//
using InitializerType = itk::BSplineTransformInitializer<
TransformType,
FixedImageType>;
InitializerType::Pointer transformInitializer = InitializerType::New();
unsigned int numberOfGridNodesInOneDimension = 8;
TransformType::MeshSizeType
meshSize;
meshSize.Fill( numberOfGridNodesInOneDimension - SplineOrder );
transformInitializer->SetTransform( outputBSplineTransform );
transformInitializer->SetImage( fixedImage );
transformInitializer->SetTransformDomainMeshSize( meshSize );
transformInitializer->InitializeTransform();
// Set transform to identity
//
using ParametersType = TransformType::ParametersType;
const unsigned int numberOfParameters =
outputBSplineTransform->GetNumberOfParameters();
ParametersType parameters( numberOfParameters );
parameters.Fill( 0.0 );
outputBSplineTransform->SetParameters( parameters );
registration->SetInitialTransform( outputBSplineTransform );
registration->InPlaceOn();
3.13. Deformable Registration
311
The registration process is run in three levels. The shrink factors and smoothing sigmas are set for
each level.
constexpr unsigned int numberOfLevels = 3;
RegistrationType::ShrinkFactorsArrayType shrinkFactorsPerLevel;
shrinkFactorsPerLevel.SetSize( numberOfLevels );
shrinkFactorsPerLevel[0] = 3;
shrinkFactorsPerLevel[1] = 2;
shrinkFactorsPerLevel[2] = 1;
RegistrationType::SmoothingSigmasArrayType smoothingSigmasPerLevel;
smoothingSigmasPerLevel.SetSize( numberOfLevels );
smoothingSigmasPerLevel[0] = 2;
smoothingSigmasPerLevel[1] = 1;
smoothingSigmasPerLevel[2] = 0;
registration->SetNumberOfLevels( numberOfLevels );
registration->SetSmoothingSigmasPerLevel( smoothingSigmasPerLevel );
registration->SetShrinkFactorsPerLevel( shrinkFactorsPerLevel );
Create the transform adaptors to modify the flexibility of the deformable transform for each level of
this multi-resolution scheme.
RegistrationType::TransformParametersAdaptorsContainerType adaptors;
// First, get fixed image physical dimensions
TransformType::PhysicalDimensionsType
fixedPhysicalDimensions;
for( unsigned int i=0; i< SpaceDimension; i++ )
{
fixedPhysicalDimensions[i] = fixedImage->GetSpacing()[i] *
static_cast(
fixedImage->GetLargestPossibleRegion().GetSize()[i] - 1 );
}
// Create the transform adaptors specific to B-splines
for( unsigned int level = 0; level < numberOfLevels; level++ )
{
using ShrinkFilterType = itk::ShrinkImageFilter<
FixedImageType,
FixedImageType>;
ShrinkFilterType::Pointer shrinkFilter = ShrinkFilterType::New();
shrinkFilter->SetShrinkFactors( shrinkFactorsPerLevel[level] );
shrinkFilter->SetInput( fixedImage );
shrinkFilter->Update();
// A good heuristic is to double the b-spline mesh resolution at each level
//
TransformType::MeshSizeType requiredMeshSize;
for( unsigned int d = 0; d < ImageDimension; d++ )
{
requiredMeshSize[d] = meshSize[d] << level;
312
Chapter 3. Registration
}
using BSplineAdaptorType =
itk::BSplineTransformParametersAdaptor;
BSplineAdaptorType::Pointer bsplineAdaptor = BSplineAdaptorType::New();
bsplineAdaptor->SetTransform( outputBSplineTransform );
bsplineAdaptor->SetRequiredTransformDomainMeshSize( requiredMeshSize );
bsplineAdaptor->SetRequiredTransformDomainOrigin(
shrinkFilter->GetOutput()->GetOrigin() );
bsplineAdaptor->SetRequiredTransformDomainDirection(
shrinkFilter->GetOutput()->GetDirection() );
bsplineAdaptor->SetRequiredTransformDomainPhysicalDimensions(
fixedPhysicalDimensions );
adaptors.push_back( bsplineAdaptor );
}
registration->SetTransformParametersAdaptorsPerLevel( adaptors );
3.13.5 BSplines Multi-Grid Image Registration in 3D
The source code for this section can be found in the file
DeformableRegistration7.cxx.
This example illustrates the use of the itk::BSplineTransform class for performing registration
of two 3D images. The example code is for the most part identical to the code presented in Section 3.13.4. The major difference is that in this example we set the image dimension to 3 and replace
the itk::LBFGSOptimizerv4 optimizer with the itk::LBFGSBOptimizerv4. We made the modification because we found that LBFGS does not behave well when the starting position is at or close
to optimal; instead we used LBFGSB in unconstrained mode.
The following are the most relevant headers to this example.
#include "itkBSplineTransform.h"
#include "itkLBFGSBOptimizerv4.h"
The parameter space of the BSplineTransform is composed by the set of all the deformations associated with the nodes of the BSpline grid. This large number of parameters enables it to represent
a wide variety of deformations, at the cost of requiring a significant amount of computation time.
We instantiate now the type of the BSplineTransform using as template parameters the type for
coordinates representation, the dimension of the space, and the order of the BSpline.
const unsigned int SpaceDimension = ImageDimension;
constexpr unsigned int SplineOrder = 3;
using CoordinateRepType = double;
3.13. Deformable Registration
313
using TransformType = itk::BSplineTransform<
CoordinateRepType,
SpaceDimension,
SplineOrder >;
The transform object is constructed, initialized like previous examples and passed to the registration
method.
TransformType::Pointer outputBSplineTransform = TransformType::New();
registration->SetInitialTransform( outputBSplineTransform );
registration->InPlaceOn();
Next we set the parameters of the LBFGSB Optimizer. Note that this optimizer does not support scales estimator and sets all the parameters scales to one. Also, we should set the boundary
condition for each variable, where boundSelect[i] can be set as: UNBOUNDED, LOWERBOUNDED,
BOTHBOUNDED, UPPERBOUNDED.
const unsigned int numParameters =
outputBSplineTransform->GetNumberOfParameters();
OptimizerType::BoundSelectionType boundSelect( numParameters );
OptimizerType::BoundValueType upperBound( numParameters );
OptimizerType::BoundValueType lowerBound( numParameters );
boundSelect.Fill( OptimizerType::UNBOUNDED );
upperBound.Fill( 0.0 );
lowerBound.Fill( 0.0 );
optimizer->SetBoundSelection( boundSelect );
optimizer->SetUpperBound( upperBound );
optimizer->SetLowerBound( lowerBound );
optimizer->SetCostFunctionConvergenceFactor( 1e+12 );
optimizer->SetGradientConvergenceTolerance( 1.0e-35 );
optimizer->SetNumberOfIterations( 500 );
optimizer->SetMaximumNumberOfFunctionEvaluations( 500 );
optimizer->SetMaximumNumberOfCorrections( 5 );
3.13.6 Image Warping with Kernel Splines
The source code for this section can be found in the file
LandmarkWarping2.cxx.
314
Chapter 3. Registration
This example illustrates how to deform an image using a KernelBase spline and two sets of landmarks.
In addition to standard headers included in previous examples, this example requires the following
includes:
#include "itkVector.h"
#include "itkLandmarkDisplacementFieldSource.h"
#include
After reading in the fixed and moving images, the deformer object is instantiated from the
itk::LandmarkDisplacementFieldSource class, and parameters of the image space and orientation are set.
using DisplacementSourceType =
itk::LandmarkDisplacementFieldSource;
DisplacementSourceType::Pointer deformer = DisplacementSourceType::New();
deformer->SetOutputSpacing( fixedImage->GetSpacing() );
deformer->SetOutputOrigin( fixedImage->GetOrigin() );
deformer->SetOutputRegion( fixedImage->GetLargestPossibleRegion() );
deformer->SetOutputDirection( fixedImage->GetDirection() );
Source and target landmarks are then created, and the points themselves are read in from a file
stream.
using LandmarkContainerType = DisplacementSourceType::LandmarkContainer;
using LandmarkPointType = DisplacementSourceType::LandmarkPointType;
LandmarkContainerType::Pointer sourceLandmarks =
LandmarkContainerType::New();
LandmarkContainerType::Pointer targetLandmarks =
LandmarkContainerType::New();
LandmarkPointType sourcePoint;
LandmarkPointType targetPoint;
std::ifstream pointsFile;
pointsFile.open( argv[1] );
unsigned int pointId = 0;
pointsFile >> sourcePoint;
pointsFile >> targetPoint;
while( !pointsFile.fail() )
{
sourceLandmarks->InsertElement( pointId, sourcePoint );
targetLandmarks->InsertElement( pointId, targetPoint );
3.13. Deformable Registration
315
++pointId;
pointsFile >> sourcePoint;
pointsFile >> targetPoint;
}
pointsFile.close();
The source and target landmark objects are then assigned to deformer.
deformer->SetSourceLandmarks( sourceLandmarks );
deformer->SetTargetLandmarks( targetLandmarks );
After calling UpdateLargestPossibleRegion() on the deformer, the displacement field may be
obtained via the GetOutput() method.
3.13.7 Image Warping with BSplines
The source code for this section can be found in the file
BSplineWarping1.cxx.
This example illustrates how to deform a 2D image using a BSplineTransform.
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkResampleImageFilter.h"
#include "itkBSplineTransform.h"
#include "itkTransformFileWriter.h"
First, we define the necessary types for the fixed and moving images and image readers.
constexpr unsigned int ImageDimension = 2;
using PixelType = unsigned char;
using FixedImageType = itk::Image< PixelType, ImageDimension >;
using MovingImageType = itk::Image< PixelType, ImageDimension >;
using FixedReaderType = itk::ImageFileReader< FixedImageType >;
using MovingReaderType = itk::ImageFileReader< MovingImageType >;
using MovingWriterType = itk::ImageFileWriter< MovingImageType >;
Use the values from the fixed image to set the corresponding values in the resampler.
316
Chapter 3. Registration
FixedImageType::SpacingType fixedSpacing
= fixedImage->GetSpacing();
FixedImageType::PointType
fixedOrigin
= fixedImage->GetOrigin();
FixedImageType::DirectionType fixedDirection = fixedImage->GetDirection();
resampler->SetOutputSpacing( fixedSpacing );
resampler->SetOutputOrigin( fixedOrigin );
resampler->SetOutputDirection( fixedDirection );
FixedImageType::RegionType fixedRegion = fixedImage->GetBufferedRegion();
FixedImageType::SizeType fixedSize = fixedRegion.GetSize();
resampler->SetSize( fixedSize );
resampler->SetOutputStartIndex( fixedRegion.GetIndex() );
resampler->SetInput( movingReader->GetOutput() );
movingWriter->SetInput( resampler->GetOutput() );
We instantiate now the type of the BSplineTransform using as template parameters the type for
coordinates representation, the dimension of the space, and the order of the B-spline.
const unsigned int SpaceDimension = ImageDimension;
constexpr unsigned int SplineOrder = 3;
using CoordinateRepType = double;
using TransformType = itk::BSplineTransform<
CoordinateRepType,
SpaceDimension,
SplineOrder >;
TransformType::Pointer bsplineTransform = TransformType::New();
Next, fill the parameters of the B-spline transform using values from the fixed image and mesh.
constexpr unsigned int numberOfGridNodes = 7;
TransformType::PhysicalDimensionsType
TransformType::MeshSizeType
fixedPhysicalDimensions;
meshSize;
for( unsigned int i=0; i< SpaceDimension; i++ )
{
fixedPhysicalDimensions[i] = fixedSpacing[i] * static_cast(
fixedSize[i] - 1 );
}
meshSize.Fill( numberOfGridNodes - SplineOrder );
bsplineTransform->SetTransformDomainOrigin( fixedOrigin );
bsplineTransform->SetTransformDomainPhysicalDimensions(
fixedPhysicalDimensions );
bsplineTransform->SetTransformDomainMeshSize( meshSize );
3.13. Deformable Registration
317
bsplineTransform->SetTransformDomainDirection( fixedDirection );
using ParametersType = TransformType::ParametersType;
const unsigned int numberOfParameters =
bsplineTransform->GetNumberOfParameters();
const unsigned int numberOfNodes = numberOfParameters / SpaceDimension;
ParametersType parameters( numberOfParameters );
The B-spline grid should now be fed with coefficients at each node. Since this is a two-dimensional
grid, each node should receive two coefficients. Each coefficient pair is representing a displacement
vector at this node. The coefficients can be passed to the B-spline in the form of an array where the
first set of elements are the first component of the displacements for all the nodes, and the second
set of elements is formed by the second component of the displacements for all the nodes.
In this example we read such displacements from a file, but for convenience we have written this
file using the pairs of (x, y) displacement for every node. The elements read from the file should
therefore be reorganized when assigned to the elements of the array. We do this by storing all the
odd elements from the file in the first block of the array, and all the even elements from the file
in the second block of the array. Finally the array is passed to the B-spline transform using the
SetParameters() method.
std::ifstream infile;
infile.open( argv[1] );
for( unsigned int n=0; n < numberOfNodes; ++n )
{
infile >> parameters[n];
infile >> parameters[n+numberOfNodes];
}
infile.close();
Finally the array is passed to the B-spline transform using the SetParameters().
bsplineTransform->SetParameters( parameters );
At this point we are ready to use the transform as part of the resample filter. We trigger the execution
of the pipeline by invoking Update() on the last filter of the pipeline, in this case writer.
resampler->SetTransform( bsplineTransform );
try
{
318
movingWriter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception thrown " << std::endl;
std::cerr << excp << std::endl;
return EXIT_FAILURE;
}
Chapter 3. Registration
3.14. Demons Deformable Registration
319
3.14 Demons Deformable Registration
For the problem of intra-modality deformable registration, the Insight Toolkit provides an implementation of Thirion’s “demons” algorithm [61, 62]. In this implementation, each image is viewed
as a set of iso-intensity contours. The main idea is that a regular grid of forces deform an image by
pushing the contours in the normal direction. The orientation and magnitude of the displacement is
derived from the instantaneous optical flow equation:
D(X) · ∇f(X) = − (m(X) − f(X))
(3.33)
In the above equation, f (X) is the fixed image, m(X) is the moving image to be registered, and D(X)
is the displacement or optical flow between the images. It is well known in optical flow literature that
Equation 3.33 is insufficient to specify D(X) locally and is usually determined using some form of
regularization. For registration, the projection of the vector on the direction of the intensity gradient
is used:
D(X) = −
(m(X) − f(X)) ∇f(X)
(3.34)
k∇fk2
However, this equation becomes unstable for small values of the image gradient, resulting in large
displacement values. To overcome this problem, Thirion re-normalizes the equation such that:
D(X) = −
(m(X) − f(X)) ∇f(X)
(3.35)
k∇fk2 + (m(X) − f(X))2 /K
Where K is a normalization factor that accounts for the units imbalance between intensities and
gradients. This factor is computed as the mean squared value of the pixel spacings. The inclusion
of K ensures the force computation is invariant to the pixel scaling of the images.
Starting with an initial deformation field D0 (X), the demons algorithm iteratively updates the field
using Equation 3.35 such that the field at the N-th iteration is given by:
N
D (X) = D
N−1
(X) −
m(X + DN−1 (X)) − f(X) ∇f(X)
2
k∇fk2 + (m(X + DN−1 (X)) − f(X))
(3.36)
Reconstruction of the deformation field is an ill-posed problem where matching the fixed and moving
images has many solutions. For example, since each image pixel is free to move independently, it is
possible that all pixels of one particular value in m(X) could map to a single image pixel in f (X) of
the same value. The resulting deformation field may be unrealistic for real-world applications. An
option to solve for the field uniquely is to enforce an elastic-like behavior, smoothing the deformation
field with a Gaussian filter between iterations.
In ITK, the demons algorithm is implemented as part of the finite difference solver (FDS) framework
and its use is demonstrated in the following example.
320
Chapter 3. Registration
3.14.1 Asymmetrical Demons Deformable Registration
The source code for this section can be found in the file
DeformableRegistration2.cxx.
This example demonstrates how to use the “demons” algorithm to deformably register two images.
The first step is to include the header files.
#include
#include
#include
#include
"itkDemonsRegistrationFilter.h"
"itkHistogramMatchingImageFilter.h"
"itkCastImageFilter.h"
"itkWarpImageFilter.h"
Second, we declare the types of the images.
constexpr unsigned int Dimension = 2;
using PixelType = unsigned short;
using FixedImageType = itk::Image< PixelType, Dimension >;
using MovingImageType = itk::Image< PixelType, Dimension >;
Image file readers are set up in a similar fashion to previous examples. To support the re-mapping
of the moving image intensity, we declare an internal image type with a floating point pixel type and
cast the input images to the internal image type.
using InternalPixelType = float;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
using FixedImageCasterType = itk::CastImageFilter< FixedImageType,
InternalImageType >;
using MovingImageCasterType = itk::CastImageFilter< MovingImageType,
InternalImageType >;
FixedImageCasterType::Pointer fixedImageCaster = FixedImageCasterType::New();
MovingImageCasterType::Pointer movingImageCaster
= MovingImageCasterType::New();
fixedImageCaster->SetInput( fixedImageReader->GetOutput() );
movingImageCaster->SetInput( movingImageReader->GetOutput() );
The demons algorithm relies on the assumption that pixels representing the same homologous point
on an object have the same intensity on both the fixed and moving images to be registered. In this
example, we will preprocess the moving image to match the intensity between the images using the
itk::HistogramMatchingImageFilter.
The basic idea is to match the histograms of the two images at a user-specified number of quantile
values. For robustness, the histograms are matched so that the background pixels are excluded from
both histograms. For MR images, a simple procedure is to exclude all gray values that are smaller
than the mean gray value of the image.
3.14. Demons Deformable Registration
321
using MatchingFilterType = itk::HistogramMatchingImageFilter<
InternalImageType,
InternalImageType >;
MatchingFilterType::Pointer matcher = MatchingFilterType::New();
For this example, we set the moving image as the source or input image and the fixed image as the
reference image.
matcher->SetInput( movingImageCaster->GetOutput() );
matcher->SetReferenceImage( fixedImageCaster->GetOutput() );
We then select the number of bins to represent the histograms and the number of points or quantile
values where the histogram is to be matched.
matcher->SetNumberOfHistogramLevels( 1024 );
matcher->SetNumberOfMatchPoints( 7 );
Simple background extraction is done by thresholding at the mean intensity.
matcher->ThresholdAtMeanIntensityOn();
In the itk::DemonsRegistrationFilter, the deformation field is represented as an image whose
pixels are floating point vectors.
using VectorPixelType = itk::Vector< float, Dimension >;
using DisplacementFieldType = itk::Image< VectorPixelType, Dimension >;
using RegistrationFilterType = itk::DemonsRegistrationFilter<
InternalImageType,
InternalImageType,
DisplacementFieldType>;
RegistrationFilterType::Pointer filter = RegistrationFilterType::New();
The input fixed image is simply the output of the fixed image casting filter. The input moving image
is the output of the histogram matching filter.
filter->SetFixedImage( fixedImageCaster->GetOutput() );
filter->SetMovingImage( matcher->GetOutput() );
The demons registration filter has two parameters: the number of iterations to be performed and the
standard deviation of the Gaussian smoothing kernel to be applied to the deformation field after each
iteration.
322
Chapter 3. Registration
filter->SetNumberOfIterations( 50 );
filter->SetStandardDeviations( 1.0 );
The registration algorithm is triggered by updating the filter. The filter output is the computed
deformation field.
filter->Update();
The itk::WarpImageFilter can be used to warp the moving image with the output deformation
field. Like the itk::ResampleImageFilter, the WarpImageFilter requires the specification of
the input image to be resampled, an input image interpolator, and the output image spacing and
origin.
using WarperType = itk::WarpImageFilter<
MovingImageType,
MovingImageType,
DisplacementFieldType >;
using InterpolatorType = itk::LinearInterpolateImageFunction<
MovingImageType,
double
>;
WarperType::Pointer warper = WarperType::New();
InterpolatorType::Pointer interpolator = InterpolatorType::New();
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
warper->SetInput( movingImageReader->GetOutput() );
warper->SetInterpolator( interpolator );
warper->SetOutputSpacing( fixedImage->GetSpacing() );
warper->SetOutputOrigin( fixedImage->GetOrigin() );
warper->SetOutputDirection( fixedImage->GetDirection() );
Unlike ResampleImageFilter, WarpImageFilter warps or transforms the input image with respect to the deformation field represented by an image of vectors. The resulting warped or resampled
image is written to file as per previous examples.
warper->SetDisplacementField( filter->GetOutput() );
Let’s execute this example using the rat lung data from the previous example. The associated data
files can be found in Examples/Data:
• RatLungSlice1.mha
• RatLungSlice2.mha
The result of the demons-based deformable registration is presented in Figure 3.48. The checkerboard comparison shows that the algorithm was able to recover the misalignment due to expiration.
3.14. Demons Deformable Registration
323
Figure 3.48: Checkerboard comparisons before and after demons-based deformable registration.
It may be also desirable to write the deformation field as an image of vectors. This can be done with
the following code.
using FieldWriterType = itk::ImageFileWriter< DisplacementFieldType >;
FieldWriterType::Pointer fieldWriter = FieldWriterType::New();
fieldWriter->SetFileName( argv[4] );
fieldWriter->SetInput( filter->GetOutput() );
fieldWriter->Update();
Note that the file format used for writing the deformation field must be capable of representing multiple components per pixel. This is the case for the MetaImage and VTK file formats for example.
A variant of the force computation is also implemented in which the gradient of the deformed moving image is also involved. This provides a level of symmetry in the force calculation during one
iteration of the PDE update. The equation used in this case is
D(X) = −
2 (m(X) − f(X)) (∇f(X) + ∇g(X))
k∇f + ∇gk2 + (m(X) − f(X))2 /K
The following example illustrates the use of this deformable registration method.
3.14.2 Symmetrical Demons Deformable Registration
The source code for this section can be found in the file
DeformableRegistration3.cxx.
(3.37)
324
Chapter 3. Registration
This example demonstrates how to use a variant of the “demons” algorithm to deformably register
two images. This variant uses a different formulation for computing the forces to be applied to
the image in order to compute the deformation fields. The variant uses both the gradient of the
fixed image and the gradient of the deformed moving image in order to compute the forces. This
mechanism for computing the forces introduces a symmetry with respect to the choice of the fixed
and moving images. This symmetry only holds during the computation of one iteration of the
PDE updates. It is unlikely that total symmetry may be achieved by this mechanism for the entire
registration process.
The first step for using this filter is to include the following header files.
#include
#include
#include
#include
"itkSymmetricForcesDemonsRegistrationFilter.h"
"itkHistogramMatchingImageFilter.h"
"itkCastImageFilter.h"
"itkWarpImageFilter.h"
Second, we declare the types of the images.
constexpr unsigned int Dimension = 2;
using PixelType = unsigned short;
using FixedImageType = itk::Image< PixelType, Dimension >;
using MovingImageType = itk::Image< PixelType, Dimension >;
Image file readers are set up in a similar fashion to previous examples. To support the re-mapping
of the moving image intensity, we declare an internal image type with a floating point pixel type and
cast the input images to the internal image type.
using InternalPixelType = float;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
using FixedImageCasterType = itk::CastImageFilter< FixedImageType,
InternalImageType >;
using MovingImageCasterType = itk::CastImageFilter< MovingImageType,
InternalImageType >;
FixedImageCasterType::Pointer fixedImageCaster = FixedImageCasterType::New();
MovingImageCasterType::Pointer movingImageCaster
= MovingImageCasterType::New();
fixedImageCaster->SetInput( fixedImageReader->GetOutput() );
movingImageCaster->SetInput( movingImageReader->GetOutput() );
The demons algorithm relies on the assumption that pixels representing the same homologous point
on an object have the same intensity on both the fixed and moving images to be registered. In this
example, we will preprocess the moving image to match the intensity between the images using the
itk::HistogramMatchingImageFilter.
3.14. Demons Deformable Registration
325
The basic idea is to match the histograms of the two images at a user-specified number of quantile
values. For robustness, the histograms are matched so that the background pixels are excluded from
both histograms. For MR images, a simple procedure is to exclude all gray values that are smaller
than the mean gray value of the image.
using MatchingFilterType = itk::HistogramMatchingImageFilter<
InternalImageType,
InternalImageType >;
MatchingFilterType::Pointer matcher = MatchingFilterType::New();
For this example, we set the moving image as the source or input image and the fixed image as the
reference image.
matcher->SetInput( movingImageCaster->GetOutput() );
matcher->SetReferenceImage( fixedImageCaster->GetOutput() );
We then select the number of bins to represent the histograms and the number of points or quantile
values where the histogram is to be matched.
matcher->SetNumberOfHistogramLevels( 1024 );
matcher->SetNumberOfMatchPoints( 7 );
Simple background extraction is done by thresholding at the mean intensity.
matcher->ThresholdAtMeanIntensityOn();
In the itk::SymmetricForcesDemonsRegistrationFilter, the deformation field is represented
as an image whose pixels are floating point vectors.
using VectorPixelType = itk::Vector< float, Dimension >;
using DisplacementFieldType = itk::Image< VectorPixelType, Dimension >;
using RegistrationFilterType = itk::SymmetricForcesDemonsRegistrationFilter<
InternalImageType,
InternalImageType,
DisplacementFieldType>;
RegistrationFilterType::Pointer filter = RegistrationFilterType::New();
The input fixed image is simply the output of the fixed image casting filter. The input moving image
is the output of the histogram matching filter.
filter->SetFixedImage( fixedImageCaster->GetOutput() );
filter->SetMovingImage( matcher->GetOutput() );
326
Chapter 3. Registration
The demons registration filter has two parameters: the number of iterations to be performed and the
standard deviation of the Gaussian smoothing kernel to be applied to the deformation field after each
iteration.
filter->SetNumberOfIterations( 50 );
filter->SetStandardDeviations( 1.0 );
The registration algorithm is triggered by updating the filter. The filter output is the computed
deformation field.
filter->Update();
The itk::WarpImageFilter can be used to warp the moving image with the output deformation
field. Like the itk::ResampleImageFilter, the WarpImageFilter requires the specification of the
input image to be resampled, an input image interpolator, and the output image spacing and origin.
using WarperType = itk::WarpImageFilter<
MovingImageType,
MovingImageType,
DisplacementFieldType >;
using InterpolatorType = itk::LinearInterpolateImageFunction<
MovingImageType,
double
>;
WarperType::Pointer warper = WarperType::New();
InterpolatorType::Pointer interpolator = InterpolatorType::New();
FixedImageType::Pointer fixedImage = fixedImageReader->GetOutput();
warper->SetInput( movingImageReader->GetOutput() );
warper->SetInterpolator( interpolator );
warper->SetOutputSpacing( fixedImage->GetSpacing() );
warper->SetOutputOrigin( fixedImage->GetOrigin() );
warper->SetOutputDirection( fixedImage->GetDirection() );
Unlike the ResampleImageFilter, the WarpImageFilter warps or transforms the input image with respect to the deformation field represented by an image of vectors. The resulting warped or resampled
image is written to file as per previous examples.
warper->SetDisplacementField( filter->GetOutput() );
Let’s execute this example using the rat lung data from the previous example. The associated data
files can be found in Examples/Data:
• RatLungSlice1.mha
• RatLungSlice2.mha
3.15. Visualizing Deformation fields
327
Figure 3.49: Checkerboard comparisons before and after demons-based deformable registration.
The result of the demons-based deformable registration is presented in Figure 3.49. The checkerboard comparison shows that the algorithm was able to recover the misalignment due to expiration.
It may be also desirable to write the deformation field as an image of vectors. This can be done with
the following code.
using FieldWriterType = itk::ImageFileWriter< DisplacementFieldType >;
FieldWriterType::Pointer fieldWriter = FieldWriterType::New();
fieldWriter->SetFileName( argv[4] );
fieldWriter->SetInput( filter->GetOutput() );
fieldWriter->Update();
Note that the file format used for writing the deformation field must be capable of representing multiple components per pixel. This is the case for the MetaImage and VTK file formats for example.
3.15 Visualizing Deformation fields
Vector deformation fields may be visualized using ParaView. ParaView [25] is an open-source,
multi-platform visualization application and uses the Visualization Toolkit as the data processing
and rendering engine and has a user interface written using a unique blend of Tcl/Tk and C++. You
may download it from http://paraview.org.
328
Chapter 3. Registration
3.15.1 Visualizing 2D deformation fields
Let us visualize the deformation field obtained from Demons Registration algorithm generated from
ITK/Examples/RegistrationITKv4/DeformableRegistration2.cxx.
Load the Deformation field in Paraview. (The deformation field must be capable of handling vector
data, such as MetaImages). Paraview shows a color map of the magnitudes of the deformation fields
as shown in 3.50.
Covert the deformation field to 3D vector data using a Calculator. The Calculator may be found in
the Filter pull down menu. A screenshot of the calculator tab is shown in Figure 3.51. Although
the deformation field is a 2D vector, we will generate a 3D vector with the third component set to 0
since Paraview generates glyphs only for 3D vectors. You may now apply a glyph of arrows to the
resulting 3D vector field by using Glyph on the menu bar. The glyphs obtained will be very dense
since a glyph is generated for each point in the data set. To better visualize the deformation field,
you may adopt one of the following approaches.
Reduce the number of glyphs by reducing the number in Max. Number of Glyphs to a reasonable amount. This uniformly downsamples the number of glyphs. Alternatively, you may apply a
Threshold filter to the Magnitude of the vector dataset and then glyph the vector data that lie above
the threshold. This eliminates the smaller deformation fields that clutter the display. You may now
reduce the number of glyphs to a reasonable value.
Figure 3.52 shows the vector field visualized using Paraview by thresholding the vector magnitudes
by 2.1 and restricting the number of glyphs to 100.
3.15.2 Visualizing 3D deformation fields
Let us create a 3D deformation field. We will use Thin Plate Splines to warp a 3D dataset and create a
deformation field. We will pick a set of point landmarks and translate them to provide a specification
of correspondences at point landmarks. Note that the landmarks have been picked randomly for
purposes of illustration and are not intended to portray a true deformation. The landmarks may be
used to produce a deformation field in several ways. Most techniques minimize some regularizing
functional representing the irregularity of the deformation field, which is usually some function of
the spatial derivatives of the field. Here will we use thin plate splines. Thin plate splines minimize
the regularizing functional
I[ f (x, y)] =
ZZ
2
2
2
+ fyy
)dxdy
( fxx
+ 2 fxy
(3.38)
where the subscripts denote partial derivatives of f.
We may now proceed as before to visualize the deformation field using Paraview as shown in Figure
3.53.
Let us register the deformed volumes generated by Thin plate warping in the previous example
3.15. Visualizing Deformation fields
329
Figure 3.50: Deformation field magnitudes displayed using Paraview
Figure 3.51: Calculators and filters may be used to compute the vector magnitude, compose vectors etc.
330
Chapter 3. Registration
Figure 3.52: Deformation field visualized using Paraview after thresholding and subsampling.
3.15. Visualizing Deformation fields
Figure 3.53: 3D Deformation field visualized using Paraview.
331
332
Chapter 3. Registration
Iteration
1
2
3
4
5
6
7
8
Function
value
156.981
68.956
38.146
26.690
23.295
21.454
20.322
19.751
kGk
Step length
14.911
11.774
4.802
2.515
1.106
1.032
1.557
0.594
0.202
1.500
1.500
1.500
1.500
1.500
1.500
1.500
Table 3.17: LBFGS Optimizer trace.
using DeformableRegistration4.cxx. Since ITK is in general N-dimensional, the only change in the
example is to replace the ImageDimension by 3.
The registration method uses B-splines and an LBFGS optimizer. The trace in Table. 3.17 prints the
trace of the optimizer through the search space.
Here kGk is the norm of the gradient at the current estimate of the minimum, x. “Function Value” is
the current value of the function, f(x).
The resulting deformation field that maps the moving to the fixed image is shown in 3.54. A difference image of two slices before and after registration is shown in 3.55. As can be seen from the
figures, the deformation field is in close agreement to the one generated from the Thin plate spline
warping.
3.15. Visualizing Deformation fields
Figure 3.54: Resulting deformation field that maps the moving image to the fixed image.
Figure 3.55: Difference image from a slice before and after registration.
333
334
Chapter 3. Registration
3.16 Model Based Registration
This section introduces the concept of
registering a geometrical model with
points
fitness value
an image. We refer to this concept as
SpatialObject
Metric
model based registration but this may
pixels
Optimizer
not be the most widespread terminolInterpolator
ogy. In this approach, a geometrical
Parameters
Moving Image
Transform
model is built first and a number of
pixels
points
parameters are identified in the model.
Figure 3.56: The basic components of model based regis- Variations of these parameters make it
tration are an image, a spatial object, a transform, a met- possible to adapt the model to the morric, an interpolator and an optimizer.
phology of a particular patient. The task
of registration is then to find the optimal combination of model parameters that will make this model
a good representation of the anatomical structures contained in an image.
For example, let’s say that in the axial view of a brain image we can roughly approximate the skull
with an ellipse. The ellipse becomes our simplified geometrical model, and registration is the task of
finding the best center for the ellipse, the measures of its axis lengths and its orientation in the plane.
This is illustrated in Figure 3.57. If we compare this approach with the image-to-image registration
problem, we can see that the main difference here is that in addition to mapping the spatial position
of the model, we can also customize internal parameters that change its shape.
Figure 3.56 illustrates the major components of the registration framework in ITK when a modelbased registration problem is configured. The basic input data for the registration is provided by
pixel data in an itk::Image and by geometrical data stored in a itk::SpatialObject. A metric
has to be defined in order to evaluate the fitness between the model and the image. This fitness value
can be improved by introducing variations in the spatial positioning of the SpatialObject and/or by
changing its internal parameters. The search space for the optimizer is now the composition of the
transform parameter and the shape internal parameters.
This same approach can be considered a segmentation technique, since once the model has been
optimally superimposed on the image we could label pixels according to their associations with specific parts of the model. The applications of model to image registration/segmentation are endless.
The main advantage of this approach is probably that, as opposed to image-to-image registration, it
actually provides Insight into the anatomical structure contained in the image. The adapted model
becomes a condensed representation of the essential elements of the anatomical structure.
ITK provides a hierarchy of classes intended to support the construction of shape models. This
hierarchy has the SpatialObject as its base class. A number of basic functionalities are defined at
this level, including the capacity to evaluate whether a given point is inside or outside of the model,
form complex shapes by creating hierarchical conglomerates of basic shapes, and support basic
spatial parameterizations like scale, orientation and position.
The following sections present examples of the typical uses of these powerful elements of the toolkit.
3.16. Model Based Registration
Model and Image Before Registration
335
Model and Image After Registration
Figure 3.57: Basic concept of Model-to-Image registration. A simplified geometrical model (ellipse) is registered against an anatomical structure (skull) by applying a spatial transform and modifying the model internal
parameters. This image is not the result of an actual registration, it is shown here only with the purpose of
illustrating the concept of model to image registration.
The source code for this section can be found in the file
ModelToImageRegistration1.cxx.
This example illustrates the use of the itk::SpatialObject as a component of the registration
framework in order to perform model based registration. The current example creates a geometrical
model composed of several ellipses. Then, it uses the model to produce a synthetic binary image of
the ellipses. Next, it introduces perturbations on the position and shape of the model, and finally it
uses the perturbed version as the input to a registration problem. A metric is defined to evaluate the
fitness between the geometric model and the image.
Let’s look first at the classes required to support SpatialObject.
In this example we
use the itk::EllipseSpatialObject as the basic shape components and we use the
itk::GroupSpatialObject to group them together as a representation of a more complex shape.
Their respective headers are included below.
#include "itkEllipseSpatialObject.h"
#include "itkGroupSpatialObject.h"
In order to generate the initial synthetic image of the ellipses, we use the
itk::SpatialObjectToImageFilter that tests—for every pixel in the image—whether the
pixel (and hence the spatial object) is inside or outside the geometric model.
336
Chapter 3. Registration
#include "itkSpatialObjectToImageFilter.h"
A metric is defined to evaluate the fitness between the SpatialObject and the Image. The base class
for this type of metric is the itk::ImageToSpatialObjectMetric, whose header is included
below.
#include "itkImageToSpatialObjectMetric.h"
As in previous registration problems, we have to evaluate the image intensity in non-grid positions.
The itk::LinearInterpolateImageFunction is used here for this purpose.
#include "itkLinearInterpolateImageFunction.h"
The SpatialObject is mapped from its own space into the image space by using a itk::Transform.
In this example, we use the itk::Euler2DTransform.
#include "itkEuler2DTransform.h"
Registration is fundamentally an optimization problem. Here we include the optimizer used to search
the parameter space and identify the best transformation that will map the shape model on top of
the image. The optimizer used in this example is the itk::OnePlusOneEvolutionaryOptimizer
that implements an evolutionary algorithm.
#include "itkOnePlusOneEvolutionaryOptimizer.h"
As in previous registration examples, it is important to track the evolution of the optimizer as it
progresses through the parameter space. This is done by using the Command/Observer paradigm.
The following lines of code implement the itk::Command observer that monitors the progress of
the registration. The code is quite similar to what we have used in previous registration examples.
#include "itkCommand.h"
template < class TOptimizer >
class IterationCallback : public itk::Command
{
public:
using Self = IterationCallback;
using Superclass = itk::Command;
using Pointer = itk::SmartPointer;
using ConstPointer = itk::SmartPointer;
itkTypeMacro( IterationCallback, Superclass );
itkNewMacro( Self );
/** Type defining the optimizer. */
3.16. Model Based Registration
337
using OptimizerType = TOptimizer;
/** Method to specify the optimizer. */
void SetOptimizer( OptimizerType * optimizer )
{
m_Optimizer = optimizer;
m_Optimizer->AddObserver( itk::IterationEvent(), this );
}
/** Execute method will print data at each iteration */
void Execute(itk::Object *caller,
const itk::EventObject & event) override
{
Execute( (const itk::Object *)caller, event);
}
void Execute(const itk::Object *,
const itk::EventObject & event) override
{
if( typeid( event ) == typeid( itk::StartEvent ) )
{
std::cout << std::endl << "Position
Value";
std::cout << std::endl << std::endl;
}
else if( typeid( event ) == typeid( itk::IterationEvent ) )
{
std::cout << m_Optimizer->GetCurrentIteration() << "
";
std::cout << m_Optimizer->GetValue() << "
";
std::cout << m_Optimizer->GetCurrentPosition() << std::endl;
}
else if( typeid( event ) == typeid( itk::EndEvent ) )
{
std::cout << std::endl << std::endl;
std::cout << "After " << m_Optimizer->GetCurrentIteration();
std::cout << " iterations " << std::endl;
std::cout << "Solution is
= " << m_Optimizer->GetCurrentPosition();
std::cout << std::endl;
}
}
This command will be invoked at every iteration of the optimizer and will print out the current
combination of transform parameters.
Consider now the most critical component of this new registration approach: the metric. This component evaluates the match between the SpatialObject and the Image. The smoothness and regularity
of the metric determine the difficulty of the task assigned to the optimizer. In this case, we use a very
robust optimizer that should be able to find its way even in the most discontinuous cost functions.
The metric to be implemented should derive from the ImageToSpatialObjectMetric class.
The following code implements a simple metric that computes the sum of the pixels that are inside
the spatial object. In fact, the metric maximum is obtained when the model and the image are
aligned. The metric is templated over the type of the SpatialObject and the type of the Image.
338
Chapter 3. Registration
template
class SimpleImageToSpatialObjectMetric :
public itk::ImageToSpatialObjectMetric
{
The fundamental operation of the metric is its GetValue() method. It is in this method that the
fitness value is computed. In our current example, the fitness is computed over the points of the
SpatialObject. For each point, its coordinates are mapped through the transform into image space.
The resulting point is used to evaluate the image and the resulting value is accumulated in a sum.
Since we are not allowing scale changes, the optimal value of the sum will result when all the
SpatialObject points are mapped on the white regions of the image. Note that the argument for the
GetValue() method is the array of parameters of the transform.
MeasureType GetValue( const ParametersType & parameters ) const override
{
double value;
this->m_Transform->SetParameters( parameters );
value = 0;
for(auto it = m_PointList.begin(); it != m_PointList.end(); ++it)
{
PointType transformedPoint = this->m_Transform->TransformPoint(*it);
if( this->m_Interpolator->IsInsideBuffer( transformedPoint ) )
{
value += this->m_Interpolator->Evaluate( transformedPoint );
}
}
return value;
}
Having defined all the registration components we are ready to put the pieces together and implement
the registration process.
First we instantiate the GroupSpatialObject and EllipseSpatialObject. These two objects are parameterized by the dimension of the space. In our current example a 2D instantiation is created.
using GroupType = itk::GroupSpatialObject< 2 >;
using EllipseType = itk::EllipseSpatialObject< 2 >;
The image is instantiated in the following lines using the pixel type and the space dimension. This
image uses a float pixel type since we plan to blur it in order to increase the capture radius of the
optimizer. Images of real pixel type behave better under blurring than those of integer pixel type.
using ImageType = itk::Image< float, 2 >;
Here is where the fun begins! In the following lines we create the EllipseSpatialObjects using their
New() methods, and assigning the results to SmartPointers. These lines will create three ellipses.
3.16. Model Based Registration
339
EllipseType::Pointer ellipse1 = EllipseType::New();
EllipseType::Pointer ellipse2 = EllipseType::New();
EllipseType::Pointer ellipse3 = EllipseType::New();
Every class deriving from SpatialObject has particular parameters enabling the user to tailor its
shape. In the case of the EllipseSpatialObject, SetRadius() is used to define the ellipse size. An
additional SetRadius(Array) method allows the user to define the ellipse axes independently.
ellipse1->SetRadius( 10.0
ellipse2->SetRadius( 10.0
ellipse3->SetRadius( 10.0
);
);
);
The ellipses are created centered in space by default. We use the following lines of code to arrange the ellipses in a triangle. The spatial transform intrinsically associated with the object is
accessed by the GetTransform() method. This transform can define a translation in space with the
SetOffset() method. We take advantage of this feature to place the ellipses at particular points in
space.
EllipseType::TransformType::OffsetType offset;
offset[ 0 ] = 100.0;
offset[ 1 ] = 40.0;
ellipse1->GetObjectToParentTransform()->SetOffset(offset);
ellipse1->ComputeObjectToWorldTransform();
offset[ 0 ] = 40.0;
offset[ 1 ] = 150.0;
ellipse2->GetObjectToParentTransform()->SetOffset(offset);
ellipse2->ComputeObjectToWorldTransform();
offset[ 0 ] = 150.0;
offset[ 1 ] = 150.0;
ellipse3->GetObjectToParentTransform()->SetOffset(offset);
ellipse3->ComputeObjectToWorldTransform();
Note that after a change has been made in the transform, the SpatialObject invokes the method
ComputeGlobalTransform() in order to update its global transform. The reason for doing this is
that SpatialObjects can be arranged in hierarchies. It is then possible to change the position of a set
of spatial objects by moving the parent of the group.
Now we add the three EllipseSpatialObjects to a GroupSpatialObject that will be subsequently
passed on to the registration method. The GroupSpatialObject facilitates the management of the
three ellipses as a higher level structure representing a complex shape. Groups can be nested any
number of levels in order to represent shapes with higher detail.
340
GroupType::Pointer group
group->AddSpatialObject(
group->AddSpatialObject(
group->AddSpatialObject(
Chapter 3. Registration
= GroupType::New();
ellipse1 );
ellipse2 );
ellipse3 );
Having the geometric model ready, we proceed to generate the binary image representing the imprint
of the space occupied by the ellipses. The SpatialObjectToImageFilter is used to that end. Note that
this filter is instantiated over the spatial object used and the image type to be generated.
using SpatialObjectToImageFilterType =
itk::SpatialObjectToImageFilter< GroupType, ImageType >;
With the defined type, we construct a filter using the New() method. The newly created filter is
assigned to a SmartPointer.
SpatialObjectToImageFilterType::Pointer imageFilter =
SpatialObjectToImageFilterType::New();
The GroupSpatialObject is passed as input to the filter.
imageFilter->SetInput( group
);
The itk::SpatialObjectToImageFilter acts as a resampling filter. Therefore it requires the
user to define the size of the desired output image. This is specified with the SetSize() method.
ImageType::SizeType size;
size[ 0 ] = 200;
size[ 1 ] = 200;
imageFilter->SetSize( size );
Finally we trigger the execution of the filter by calling the Update() method.
imageFilter->Update();
In order to obtain a smoother metric,
we blur the image using a
itk::DiscreteGaussianImageFilter. This extends the capture radius of the metric and
produce a more continuous cost function to optimize. The following lines instantiate the Gaussian
filter and create one object of this type using the New() method.
using GaussianFilterType =
itk::DiscreteGaussianImageFilter< ImageType, ImageType >;
GaussianFilterType::Pointer gaussianFilter =
GaussianFilterType::New();
3.16. Model Based Registration
341
The output of the SpatialObjectToImageFilter is connected as input to the DiscreteGaussianImageFilter.
gaussianFilter->SetInput( imageFilter->GetOutput() );
The variance of the filter is defined as a large value in order to increase the capture radius. Finally
the execution of the filter is triggered using the Update() method.
constexpr double variance = 20;
gaussianFilter->SetVariance(variance);
gaussianFilter->Update();
Below we instantiate the type of the itk::ImageToSpatialObjectRegistrationMethod method
and instantiate a registration object with the New() method. Note that the registration type is templated over the Image and the SpatialObject types. The spatial object in this case is the group of
spatial objects.
using RegistrationType =
itk::ImageToSpatialObjectRegistrationMethod< ImageType, GroupType >;
RegistrationType::Pointer registration = RegistrationType::New();
Now we instantiate the metric that is templated over the image type and the spatial object type. As
usual, the New() method is used to create an object.
using MetricType = SimpleImageToSpatialObjectMetric< ImageType, GroupType >;
MetricType::Pointer metric = MetricType::New();
An interpolator will be needed to evaluate the image at non-grid positions. Here we instantiate a
linear interpolator type.
using InterpolatorType =
itk::LinearInterpolateImageFunction< ImageType, double >;
InterpolatorType::Pointer interpolator = InterpolatorType::New();
The following lines instantiate the evolutionary optimizer.
using OptimizerType = itk::OnePlusOneEvolutionaryOptimizer;
OptimizerType::Pointer optimizer = OptimizerType::New();
Next, we instantiate the transform class. In this case we use the Euler2DTransform that implements
a rigid transform in 2D space.
342
Chapter 3. Registration
using TransformType = itk::Euler2DTransform<>;
TransformType::Pointer transform = TransformType::New();
Evolutionary algorithms are based on testing random variations of parameters. In order to support the computation of random values, ITK provides a family of random number generators. In
this example, we use the itk::NormalVariateGenerator which generates values with a normal
distribution.
itk::Statistics::NormalVariateGenerator::Pointer generator
= itk::Statistics::NormalVariateGenerator::New();
The random number generator must be initialized with a seed.
generator->Initialize(12345);
The OnePlusOneEvolutionaryOptimizer is initialized by specifying the random number generator,
the number of samples for the initial population and the maximum number of iterations.
optimizer->SetNormalVariateGenerator( generator );
optimizer->Initialize( 10 );
optimizer->SetMaximumIteration( 400 );
As in previous registration examples, we take care to normalize the dynamic range of the different
transform parameters. In particular, the we must compensate for the ranges of the angle and translations of the Euler2DTransform. In order to achieve this goal, we provide an array of scales to the
optimizer.
TransformType::ParametersType parametersScale;
parametersScale.set_size(3);
parametersScale[0] = 1000; // angle scale
for( unsigned int i=1; i<3; i++ )
{
parametersScale[i] = 2; // offset scale
}
optimizer->SetScales( parametersScale );
Here we instantiate the Command object that will act as an observer of the registration method and
print out parameters at each iteration. Earlier, we defined this command as a class templated over the
optimizer type. Once it is created with the New() method, we connect the optimizer to the command.
using IterationCallbackType = IterationCallback< OptimizerType >;
IterationCallbackType::Pointer callback = IterationCallbackType::New();
callback->SetOptimizer( optimizer );
3.16. Model Based Registration
343
All the components are plugged into the ImageToSpatialObjectRegistrationMethod object. The typical Set() methods are used here. Note the use of the SetMovingSpatialObject() method for
connecting the spatial object. We provide the blurred version of the original synthetic binary image
as the input image.
registration->SetFixedImage( gaussianFilter->GetOutput() );
registration->SetMovingSpatialObject( group );
registration->SetTransform( transform );
registration->SetInterpolator( interpolator );
registration->SetOptimizer( optimizer );
registration->SetMetric( metric );
The initial set of transform parameters is passed to the registration method using the
SetInitialTransformParameters() method. Note that since our original model is already registered with the synthetic image, we introduce an artificial mis-registration in order to initialize the
optimization at some point away from the optimal value.
TransformType::ParametersType initialParameters(
transform->GetNumberOfParameters() );
initialParameters[0] = 0.2;
// Angle
initialParameters[1] = 7.0;
// Offset X
initialParameters[2] = 6.0;
// Offset Y
registration->SetInitialTransformParameters(initialParameters);
Due to the character of the metric used to evaluate the fitness between the spatial object and the
image, we must tell the optimizer that we are interested in finding the maximum value of the metric.
Some metrics associate low numeric values with good matching, while others associate high numeric
values with good matching. The MaximizeOn() and MaximizeOff() methods allow the user to deal
with both types of metrics.
optimizer->MaximizeOn();
Finally, we trigger the execution of the registration process with the Update() method. We place
this call in a try/catch block in case any exception is thrown during the process.
try
{
registration->Update();
std::cout << "Optimizer stop condition: "
<< registration->GetOptimizer()->GetStopConditionDescription()
<< std::endl;
}
catch( itk::ExceptionObject & exp )
{
std::cerr << "Exception caught ! " << std::endl;
344
Chapter 3. Registration
std::cerr << exp << std::endl;
}
The set of transform parameters resulting from the registration can be recovered with the
GetLastTransformParameters() method. This method returns the array of transform parameters that should be interpreted according to the implementation of each transform. In our current
example, the Euler2DTransform has three parameters: the rotation angle, the translation in x and the
translation in y.
RegistrationType::ParametersType finalParameters
= registration->GetLastTransformParameters();
std::cout << "Final Solution is : " << finalParameters << std::endl;
The results are presented in Figure 3.58. The left side shows the evolution of the angle parameter as
a function of iteration numbers, while the right side shows the (x, y) translation.
3.17 Point Set Registration
PointSet-to-PointSet registration is a common problem in medical image analysis. It usually arises
in cases where landmarks are extracted from images and are used for establishing the spatial correspondence between the images. This type of registration can be considered to be the simplest
case of feature-based registration. In general terms, feature-based registration is more efficient than
the intensity based method that we have presented so far. However, feature-base registration brings
the new problem of identifying and extracting the features from the images, which is not a minor
challenge.
The two most common scenarios in PointSet to PointSet registration are
• Two PointSets with the same number of points, and where each point in one set has a known
correspondence to exactly one point in the second set.
• Two PointSets without known correspondences between the points of one set and the points
of the other. In this case the PointSets may have different numbers of points.
The first case can be solved with a closed form solution when we are dealing
with a Rigid or an Affine Transform [26].
This is done in ITK with the class
itk::LandmarkBasedTransformInitializer. If we are interested in a deformable Transformation then the problem can be solved with the itk::KernelTransform family of classes, which
includes Thin Plate Splines among others [51]. In both circumstances, the availability o f correspondences between the points make possible to apply a straight forward solution to the problem.
3.17. Point Set Registration
345
14
Rotation Angle (degrees)
12
10
8
6
4
2
0
0
50
100
150
200
Iteration No.
250
300
350
400
5
Translation Y (mm)
0
−5
−10
−15
−20
−25
0
5
10
15
20
Translation X (mm)
25
30
Figure 3.58: Plots of the angle and translation parameters for a registration process between an spatial object
and an image.
346
Chapter 3. Registration
The classical algorithm for performing PointSet to PointSet registration is the Iterative Closest Point
(ICP) algorithm. The following examples illustrate how this can be used in ITK.
3.17.1 Point Set Registration in 2D
The source code for this section can be found in the file
IterativeClosestPoint1.cxx.
This example illustrates how to perform Iterative Closest Point (ICP) registration in ITK. The main
class featured in this section is the itk::EuclideanDistancePointMetric.
The first step is to include the relevant headers.
#include
#include
#include
#include
"itkTranslationTransform.h"
"itkEuclideanDistancePointMetric.h"
"itkLevenbergMarquardtOptimizer.h"
"itkPointSetToPointSetRegistrationMethod.h"
Next, define the necessary types for the fixed and moving pointsets and point containers.
constexpr unsigned int Dimension = 2;
using PointSetType = itk::PointSet< float, Dimension >;
PointSetType::Pointer fixedPointSet = PointSetType::New();
PointSetType::Pointer movingPointSet = PointSetType::New();
using PointType = PointSetType::PointType;
using PointsContainer = PointSetType::PointsContainer;
PointsContainer::Pointer fixedPointContainer = PointsContainer::New();
PointsContainer::Pointer movingPointContainer = PointsContainer::New();
PointType fixedPoint;
PointType movingPoint;
After the points are read in from files, set up the metric type.
using MetricType = itk::EuclideanDistancePointMetric<
PointSetType, PointSetType>;
MetricType::Pointer metric = MetricType::New();
Now, setup the transform, optimizers, and registration method using the point set types defined
earlier.
3.17. Point Set Registration
347
using TransformType = itk::TranslationTransform< double, Dimension >;
TransformType::Pointer transform = TransformType::New();
// Optimizer Type
using OptimizerType = itk::LevenbergMarquardtOptimizer;
OptimizerType::Pointer
optimizer
= OptimizerType::New();
optimizer->SetUseCostFunctionGradient(false);
// Registration Method
using RegistrationType = itk::PointSetToPointSetRegistrationMethod<
PointSetType, PointSetType >;
RegistrationType::Pointer
registration = RegistrationType::New();
// Scale the translation components of the Transform in the Optimizer
OptimizerType::ScalesType scales( transform->GetNumberOfParameters() );
scales.Fill( 0.01 );
Next we setup the convergence criteria, and other properties required by the optimizer.
unsigned long
numberOfIterations = 100;
double
gradientTolerance = 1e-5;
double
valueTolerance
= 1e-5;
double
epsilonFunction
= 1e-6;
// convergence criterion
// convergence criterion
// convergence criterion
optimizer->SetScales( scales );
optimizer->SetNumberOfIterations( numberOfIterations );
optimizer->SetValueTolerance( valueTolerance );
optimizer->SetGradientTolerance( gradientTolerance );
optimizer->SetEpsilonFunction( epsilonFunction );
In this case we start from an identity transform, but in reality the user will usually be able to provide
a better guess than this.
transform->SetIdentity();
Finally, connect all the components required for the registration, and an observer.
registration->SetMetric(
metric
);
registration->SetOptimizer(
optimizer
);
registration->SetTransform(
transform
);
registration->SetFixedPointSet( fixedPointSet );
registration->SetMovingPointSet( movingPointSet
// Connect an observer
);
348
Chapter 3. Registration
CommandIterationUpdate::Pointer observer = CommandIterationUpdate::New();
optimizer->AddObserver( itk::IterationEvent(), observer );
3.17.2 Point Set Registration in 3D
The source code for this section can be found in the file
IterativeClosestPoint2.cxx.
This example illustrates how to perform Iterative Closest Point (ICP) registration in ITK using sets
of 3D points.
The first step is to include the relevant headers.
#include
#include
#include
#include
#include
#include
"itkEuler3DTransform.h"
"itkEuclideanDistancePointMetric.h"
"itkLevenbergMarquardtOptimizer.h"
"itkPointSetToPointSetRegistrationMethod.h"
First, define the necessary types for the moving and fixed point sets.
using PointSetType = itk::PointSet< float, Dimension >;
PointSetType::Pointer fixedPointSet = PointSetType::New();
PointSetType::Pointer movingPointSet = PointSetType::New();
using PointType = PointSetType::PointType;
using PointsContainer = PointSetType::PointsContainer;
PointsContainer::Pointer fixedPointContainer = PointsContainer::New();
PointsContainer::Pointer movingPointContainer = PointsContainer::New();
PointType fixedPoint;
PointType movingPoint;
After the points are read in from files, setup the metric to be used later by the registration.
using MetricType = itk::EuclideanDistancePointMetric<
PointSetType, PointSetType >;
MetricType::Pointer metric = MetricType::New();
Next, setup the tranform, optimizers, and registration.
3.17. Point Set Registration
349
using TransformType = itk::Euler3DTransform< double >;
TransformType::Pointer transform = TransformType::New();
// Optimizer Type
using OptimizerType = itk::LevenbergMarquardtOptimizer;
OptimizerType::Pointer
optimizer
= OptimizerType::New();
optimizer->SetUseCostFunctionGradient(false);
// Registration Method
using RegistrationType = itk::PointSetToPointSetRegistrationMethod<
PointSetType, PointSetType >;
RegistrationType::Pointer
registration = RegistrationType::New();
Scale the translation components of the Transform in the Optimizer
OptimizerType::ScalesType scales( transform->GetNumberOfParameters() );
Next, set the scales and ranges for translations and rotations in the transform. Also, set the convergence criteria and number of iterations to be used by the optimizer.
constexpr double translationScale = 1000.0; // dynamic range of translations
constexpr double rotationScale = 1.0;
// dynamic range of rotations
scales[0]
scales[1]
scales[2]
scales[3]
scales[4]
scales[5]
=
=
=
=
=
=
1.0
1.0
1.0
1.0
1.0
1.0
/
/
/
/
/
/
rotationScale;
rotationScale;
rotationScale;
translationScale;
translationScale;
translationScale;
unsigned long
numberOfIterations = 2000;
double
gradientTolerance = 1e-4;
// convergence criterion
double
valueTolerance
= 1e-4;
// convergence criterion
double
epsilonFunction
= 1e-5;
// convergence criterion
optimizer->SetScales( scales );
optimizer->SetNumberOfIterations( numberOfIterations );
optimizer->SetValueTolerance( valueTolerance );
optimizer->SetGradientTolerance( gradientTolerance );
optimizer->SetEpsilonFunction( epsilonFunction );
Here we start with an identity transform, although the user will usually be able to provide a better
guess than this.
350
Chapter 3. Registration
transform->SetIdentity();
Connect all the components required for the registration.
registration->SetMetric(
metric
);
registration->SetOptimizer(
optimizer
);
registration->SetTransform(
transform
);
registration->SetFixedPointSet( fixedPointSet );
registration->SetMovingPointSet( movingPointSet
);
3.17.3 Point Set to Distance Map Metric
The source code for this section can be found in the file
IterativeClosestPoint3.cxx.
This example illustrates how to perform Iterative Closest Point (ICP) registration in ITK using a
DistanceMap in order to increase the performance. There is of course a trade-off between the time
needed for computing the DistanceMap and the time saved by its repeated use during the iterative
computation of the point-to-point distances. It is then necessary in practice to ponder both factors.
itk::EuclideanDistancePointMetric.
The first step is to include the relevant headers.
#include
#include
#include
#include
#include
#include
#include
#include
"itkTranslationTransform.h"
"itkEuclideanDistancePointMetric.h"
"itkLevenbergMarquardtOptimizer.h"
"itkPointSetToPointSetRegistrationMethod.h"
"itkDanielssonDistanceMapImageFilter.h"
"itkPointSetToImageFilter.h"
Next, define the necessary types for the fixed and moving point sets.
using PointSetType = itk::PointSet< float, Dimension >;
PointSetType::Pointer fixedPointSet = PointSetType::New();
PointSetType::Pointer movingPointSet = PointSetType::New();
using PointType = PointSetType::PointType;
using PointsContainer = PointSetType::PointsContainer;
PointsContainer::Pointer fixedPointContainer = PointsContainer::New();
3.17. Point Set Registration
351
PointsContainer::Pointer movingPointContainer = PointsContainer::New();
PointType fixedPoint;
PointType movingPoint;
Setup the metric, transform, optimizers and registration in a manner similar to the previous two
examples.
In the preparation of the distance map, we first need to map the fixed points into a binary image.
using BinaryImageType = itk::Image< unsigned char,
Dimension >;
using PointsToImageFilterType = itk::PointSetToImageFilter<
PointSetType,
BinaryImageType>;
PointsToImageFilterType::Pointer
pointsToImageFilter = PointsToImageFilterType::New();
pointsToImageFilter->SetInput( fixedPointSet );
BinaryImageType::SpacingType spacing;
spacing.Fill( 1.0 );
BinaryImageType::PointType origin;
origin.Fill( 0.0 );
Continue to prepare the distance map, in order to accelerate the distance computations.
pointsToImageFilter->SetSpacing( spacing );
pointsToImageFilter->SetOrigin( origin
);
pointsToImageFilter->Update();
BinaryImageType::Pointer binaryImage = pointsToImageFilter->GetOutput();
using DistanceImageType = itk::Image< unsigned short, Dimension >;
using DistanceFilterType = itk::DanielssonDistanceMapImageFilter<
BinaryImageType, DistanceImageType>;
DistanceFilterType::Pointer distanceFilter = DistanceFilterType::New();
distanceFilter->SetInput( binaryImage );
distanceFilter->Update();
metric->SetDistanceMap( distanceFilter->GetOutput() );
352
Chapter 3. Registration
3.18 Registration Troubleshooting
So you read the previous sections, you wrote the code, it compiles and links fine, but when you run
it the registration results are not what you were expecting. In that case, this section is for you. This
is a compilation of the most common problems that users face when performing image registration.
It provides explanations on the potential sources of the problems, and advice on how to deal with
those problems.
Most of the material in this section has been taken from frequently asked questions of the ITK users
list.
3.18.1 Too many samples outside moving image buffer
http://public.kitware.com/pipermail/insight-users/2007-March/021442.html
This is a common error message in image registration.
It means that at the current iteration of the optimization, the two images as so off-registration that
their spatial overlap is not large enough for bringing them back into registration.
The common causes of this problem are:
• Poor initialization: You must initialize the transform properly. Please familiarize yourself
with the itk::CenteredTransformInitializer class.
• Optimizer steps too large. If you optimizer takes steps that are too large, it risks to become
unstable and to send the images too far apart. You may want to start the optimizer with a
maximum step length of 1.0, and only increase it once you have managed to fine tune all other
registration parameters.
Increasing the step length makes your program faster, but it also makes it more unstable.
• Poor set up of the transform parameters scaling. This is extremely critical in registration.
You must make sure that you balance the relative difference of scale between the rotation
parameters and the translation parameters.
In typical medical datasets such as CT and MR, translations are measured in millimeters, and
therefore are in the range of -100:100, while rotations are measured in radians, and therefore
they tend to be in the range of -1:1.
A rotation of 3 radians is catastrophic, while a translation of 3 millimeters is rather inoffensive.
That difference in scale is the one that must be accounted for.
3.18.2 General heuristics for parameter fine-tunning
http://public.kitware.com/pipermail/insight-users/2007-March/021435.html
3.18. Registration Troubleshooting
353
Here is some advice on how to fine tune the parameters of the registration process.
1) Set Maximum step length to 0.1 and do not change it until all other parameters are stable.
2) Set Minimum step length to 0.001 and do not change it.
You could interpret these two parameters as if their units were radians. So, 0.1 radian = 5.7 degrees.
3) Number of histogram bins:
First plot the histogram of your image using the example program in
Insight/Examples/Statistics/ImageHistogram2.cxx
In that program use first a large number of bins (for example 2000) and identify the different populations of intensity level and to what anatomical structures they correspond.
Once you identify the anatomical structures in the histogram, then rerun that same program with less
and less number of bins, until you reach the minimun number of bins for which all the tissues that
are important for your application, are still distinctly differentiated in the histogram. At that point,
take that number of bins and us it for your Mutual Information metric.
4) Number of Samples: The trade-off with the number of samples is the following:
a) computation time of registration is linearly proportional to the number of samples b) the samples
must be enough to significantly populate the joint histogram. c) Once the histogram is populated,
there is not much use in adding more samples. Therefore do the following:
Plot the joint histogram of both images, using the number of bins that you selected in item (3). You
can do this by modifying the code of the example:
Insight/Examples/Statistics/ ImageMutualInformation1.cxx you have to change the code to print out
the values of the bins. Then use a plotting program such as gnuplot, or Matlab, or even Excel and
look at the distribution. The number of samples to take must be enough for producing the same
”appearance” of the joint histogram. As an arbitrary rule of thumb you may want to start using
a high number of samples (80% - 100%). And do not change it until you have mastered the other
parameters of the registration. Once you get your registration to converge you can revisit the number
of samples and reduce it in order to make the registration run faster. You can simply reduce it until
you find that the registration becomes unstable. That’s your critical bound for the minimum number
of samples. Take that number and multiply it by the magic number 1.5, to send it back to a stable
region, or if your application is really critical, then use an even higher magic number x2.0.
This is just engineering: you figure out what is the minimal size of a piece of steel that will support
a bridge, and then you enlarge it to keep it away from the critical value.
5) The MOST critical values of the registration process are the scaling parameters that define the
proportions between the parameters of the transform. In your case, for an Affine Transform in 2D,
you have 6 parameters. The first four are the ones of the Matrix, and the last two are the translation.
The rotation matrix value must be in the ranges of radians which is typically [ -1 to 1 ], while the
translation values are in the ranges of millimeters (your image size units). You want to start by
setting the scaling of the matrix parameters to 1.0, and the scaling of the Translation parameters to
354
Chapter 3. Registration
the holy esoteric values:
1.0 / ( 10.0 * pixelspacing[0] * imagesize[0] ) 1.0 / ( 10.0 * pixelspacing[1] * imagesize[1] )
This is telling the optimizer that you consider that rotating the image by 57 degrees is as ”significant”
as translating the image by half its physical extent.
Note that esoteric value has included the arbitrary number 10.0 in the denominator, for no other
reason that we have been lucky when using that factor. This of course is just a superstition, so you
should feel free to experiment with different values of this number.
Just keep in mind that what the optimizer will do is to “jump” in a parametric space of 6 dimensions,
and that the component of the jump on every dimension will be proportional to 1/scaling factor *
OptimizerStepLength. Since you set the optimizer Step Length to 0.1, the optimizer will start by
exploring the rotations at jumps of about 5 degrees, which is a conservative rotation for most medical
applications.
If you have reasons to think that your rotations are larger or smaller, then you should modify the
scaling factor of the matrix parameters accordingly.
In the same way, if you think that 1/10 of the image size is too large as the first step for exploring
the translations, then you should modify the scaling of translation parameters accordingly.
In order to drive all these you need to analyze the feedback that the observer is providing you. For
example, plot the metric values, and plot the translation coordinates so that you can get a feeling of
how the registration is behaving.
Note also that image registration is not a science. It is a pure engineerig practice, and therefore,
there are no correct answers, nor “truths” to be found. It is all about how much quality you want,
and how must computation time, and development time you are willing to pay for that quality. The
“satisfying” answer for your specific application must be found by exploring the trade-offs between
the different parameters that regulate the image registration process.
If you are proficient in VTK you may want to consider attaching some visualization to the Event
observer, so that you can have a visual feedback on the progress of the registration. This is a lot
more productive than trying to interpret the values printed out on the console by the observer.
CHAPTER
FOUR
SEGMENTATION
Segmentation of medical images is a challenging task. A myriad of different methods have been
proposed and implemented in recent years. In spite of the huge effort invested in this problem, there
is no single approach that can generally solve the problem of segmentation for the large variety of
image modalities existing today.
The most effective segmentation algorithms are obtained by carefully customizing combinations of
components. The parameters of these components are tuned for the characteristics of the image
modality used as input and the features of the anatomical structure to be segmented.
The Insight Toolkit provides a basic set of algorithms that can be used to develop and customize
a full segmentation application. Some of the most commonly used segmentation components are
described in the following sections.
4.1 Region Growing
Region growing algorithms have proven to be an effective approach for image segmentation. The
basic approach of a region growing algorithm is to start from a seed region (typically one or more
pixels) that are considered to be inside the object to be segmented. The pixels neighboring this
region are evaluated to determine if they should also be considered part of the object. If so, they are
added to the region and the process continues as long as new pixels are added to the region. Region
growing algorithms vary depending on the criteria used to decide whether a pixel should be included
in the region or not, the type connectivity used to determine neighbors, and the strategy used to visit
neighboring pixels.
Several implementations of region growing are available in ITK. This section describes some of the
most commonly used.
356
Chapter 4. Segmentation
4.1.1 Connected Threshold
A simple criterion for including pixels in a growing region is to evaluate intensity value inside a
specific interval.
The source code for this section can be found in the file
ConnectedThresholdImageFilter.cxx.
The following example illustrates the use of the itk::ConnectedThresholdImageFilter. This
filter uses the flood fill iterator. Most of the algorithmic complexity of a region growing method
comes from visiting neighboring pixels. The flood fill iterator assumes this responsibility and greatly
simplifies the implementation of the region growing algorithm. Thus the algorithm is left to establish
a criterion to decide whether a particular pixel should be included in the current region or not.
The criterion used by the ConnectedThresholdImageFilter is based on an interval of intensity
values provided by the user. Lower and upper threshold values should be provided. The regiongrowing algorithm includes those pixels whose intensities are inside the interval.
I(X) ∈ [lower, upper]
(4.1)
Let’s look at the minimal code required to use this algorithm. First, the following header defining
the ConnectedThresholdImageFilter class must be included.
#include "itkConnectedThresholdImageFilter.h"
Noise present in the image can reduce the capacity of this filter to grow large regions. When faced
with noisy images, it is usually convenient to pre-process the image by using an edge-preserving
smoothing filter. Any of the filters discussed in Section 2.7.3 could be used to this end. In this
particular example we use the itk::CurvatureFlowImageFilter, so we need to include its header
file.
#include "itkCurvatureFlowImageFilter.h"
We declare the image type based on a particular pixel type and dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The smoothing filter is instantiated using the image type as a template parameter.
4.1. Region Growing
357
using CurvatureFlowImageFilterType =
itk::CurvatureFlowImageFilter< InternalImageType, InternalImageType >;
Then the filter is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
CurvatureFlowImageFilterType::Pointer smoothing =
CurvatureFlowImageFilterType::New();
We now declare the type of the region growing filter.
ConnectedThresholdImageFilter.
In this case it is the
using ConnectedFilterType =
itk::ConnectedThresholdImageFilter< InternalImageType,
InternalImageType >;
Then we construct one filter of this class using the New() method.
ConnectedFilterType::Pointer connectedThreshold = ConnectedFilterType::New();
Now it is time to connect a simple, linear pipeline. A file reader is added at the beginning of the
pipeline and a cast filter and writer are added at the end. The cast filter is required to convert float
pixel types to integer types since only a few image file formats support float types.
smoothing->SetInput( reader->GetOutput() );
connectedThreshold->SetInput( smoothing->GetOutput() );
caster->SetInput( connectedThreshold->GetOutput() );
writer->SetInput( caster->GetOutput() );
CurvatureFlowImageFilter requires a couple of parameters. The following are typical values
for 2D images. However, these values may have to be adjusted depending on the amount of noise
present in the input image.
smoothing->SetNumberOfIterations( 5 );
smoothing->SetTimeStep( 0.125 );
We now set the lower and upper threshold values. Any pixel whose value is between
lowerThreshold and upperThreshold will be included in the region, and any pixel whose value is
outside will be excluded. Setting these values too close together will be too restrictive for the region
to grow; setting them too far apart will cause the region to engulf the image.
358
Chapter 4. Segmentation
Structure
White matter
Ventricle
Gray matter
Seed Index
(60, 116)
(81, 112)
(107, 69)
Lower
150
210
180
Upper
180
250
210
Output Image
Second from left in Figure 4.1
Third from left in Figure 4.1
Fourth from left in Figure 4.1
Table 4.1: Parameters used for segmenting some brain structures shown in Figure 4.1 with the filter
itk::ConnectedThresholdImageFilter .
connectedThreshold->SetLower( lowerThreshold );
connectedThreshold->SetUpper( upperThreshold );
The output of this filter is a binary image with zero-value pixels everywhere except on the extracted
region. The intensity value set inside the region is selected with the method SetReplaceValue().
connectedThreshold->SetReplaceValue( 255 );
The algorithm must be initialized by setting a seed point (i.e., the itk::Index of the pixel from
which the region will grow) using the SetSeed() method. It is convenient to initialize with a point
in a typical region of the anatomical structure to be segmented.
connectedThreshold->SetSeed( index );
Invocation of the Update() method on the writer triggers execution of the pipeline. It is usually
wise to put update calls in a try/catch block in case errors occur and exceptions are thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
Let’s run this example using as input the image BrainProtonDensitySlice.png provided in the
directory Examples/Data. We can easily segment the major anatomical structures by providing
seeds in the appropriate locations and defining values for the lower and upper thresholds. Figure 4.1
illustrates several examples of segmentation. The parameters used are presented in Table 4.1.
Notice that the gray matter is not being completely segmented. This illustrates the vulnerability of
the region-growing methods when the anatomical structures to be segmented do not have a homo-
4.1. Region Growing
359
Figure 4.1: Segmentation results for the ConnectedThreshold filter for various seed points.
geneous statistical distribution over the image space. You may want to experiment with different
values of the lower and upper thresholds to verify how the accepted region will extend.
Another option for segmenting regions is to take advantage of the functionality provided by the
ConnectedThresholdImageFilter for managing multiple seeds. The seeds can be passed oneby-one to the filter using the AddSeed() method. You could imagine a user interface in which an
operator clicks on multiple points of the object to be segmented and each selected point is passed as
a seed to this filter.
4.1.2 Otsu Segmentation
Another criterion for classifying pixels is to minimize the error of misclassification. The goal is to
find a threshold that classifies the image into two clusters such that we minimize the area under the
histogram for one cluster that lies on the other cluster’s side of the threshold. This is equivalent to
minimizing the within class variance or equivalently maximizing the between class variance.
The source code for this section can be found in the file
OtsuThresholdImageFilter.cxx.
This example illustrates how to use the itk::OtsuThresholdImageFilter.
#include "itkOtsuThresholdImageFilter.h"
The next step is to decide which pixel types to use for the input and output images, and to define the
image dimension.
using InputPixelType = unsigned char;
using OutputPixelType = unsigned char;
constexpr unsigned int Dimension = 2;
The input and output image types are now defined using their respective pixel types and dimensions.
360
Chapter 4. Segmentation
using InputImageType = itk::Image< InputPixelType, Dimension >;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
The filter type can be instantiated using the input and output image types defined above.
using FilterType = itk::OtsuThresholdImageFilter<
InputImageType, OutputImageType >;
An itk::ImageFileReader class is also instantiated in order to read image data from a file. (See
Section 1 on page 1 for more information about reading and writing data.)
using ReaderType = itk::ImageFileReader< InputImageType >;
An itk::ImageFileWriter is instantiated in order to write the output image to a file.
using WriterType = itk::ImageFileWriter< OutputImageType >;
Both the filter and the reader are created by invoking their New() methods and assigning the result
to itk::SmartPointers.
ReaderType::Pointer reader = ReaderType::New();
FilterType::Pointer filter = FilterType::New();
The image obtained with the reader is passed as input to the OtsuThresholdImageFilter.
filter->SetInput( reader->GetOutput() );
The method SetOutsideValue() defines the intensity value to be assigned to those pixels
whose intensities are outside the range defined by the lower and upper thresholds. The method
SetInsideValue() defines the intensity value to be assigned to pixels with intensities falling inside the threshold range.
filter->SetOutsideValue( outsideValue );
filter->SetInsideValue( insideValue );
Execution of the filter is triggered by invoking the Update() method, which we wrap in a
try/catch block. If the filter’s output has been passed as input to subsequent filters, the Update()
call on any downstream filters in the pipeline will indirectly trigger the update of this filter.
4.1. Region Growing
361
Figure 4.2: Effect of the OtsuThresholdImageFilter on a slice from a MRI proton density image of the brain.
try
{
filter->Update();
}
catch( itk::ExceptionObject & excp )
{
std::cerr << "Exception thrown " << excp << std::endl;
}
We can now retrieve the internally-computed threshold value with the GetThreshold() method and
print it to the console.
int threshold = filter->GetThreshold();
std::cout << "Threshold = " << threshold << std::endl;
Figure 4.2 illustrates the effect of this filter on a MRI proton density image of the brain. This
figure shows the limitations of this filter for performing segmentation by itself. These limitations
are particularly noticeable in noisy images and in images lacking spatial uniformity as is the case
with MRI due to field bias.
The following classes provide similar functionality:
• itk::ThresholdImageFilter
362
Chapter 4. Segmentation
The source code for this section can be found in the file
OtsuMultipleThresholdImageFilter.cxx.
This example illustrates how to use the itk::OtsuMultipleThresholdsCalculator.
#include "itkOtsuMultipleThresholdsCalculator.h"
OtsuMultipleThresholdsCalculator calculates thresholds for a given histogram so as to maximize the between-class variance. We use ScalarImageToHistogramGenerator to generate histograms. The histogram type defined by the generator is then used to instantiate the type of the Otsu
threshold calculator.
using ScalarImageToHistogramGeneratorType =
itk::Statistics::ScalarImageToHistogramGenerator;
using HistogramType = ScalarImageToHistogramGeneratorType::HistogramType;
using CalculatorType = itk::OtsuMultipleThresholdsCalculator;
Once thresholds are computed we will use BinaryThresholdImageFilter to segment the input
image.
using FilterType = itk::BinaryThresholdImageFilter<
InputImageType, OutputImageType >;
Create a histogram generator and calculator using the standard New() method.
ScalarImageToHistogramGeneratorType::Pointer scalarImageToHistogramGenerator
= ScalarImageToHistogramGeneratorType::New();
CalculatorType::Pointer calculator = CalculatorType::New();
FilterType::Pointer filter = FilterType::New();
Set the following properties for the histogram generator and the calculators, in this case grabbing
the number of thresholds from the command line.
scalarImageToHistogramGenerator->SetNumberOfBins( 128 );
calculator->SetNumberOfThresholds( std::stoi( argv[4] ) );
The pipeline will look as follows:
scalarImageToHistogramGenerator->SetInput( reader->GetOutput() );
calculator->SetInputHistogram(
scalarImageToHistogramGenerator->GetOutput() );
filter->SetInput( reader->GetOutput() );
writer->SetInput( filter->GetOutput() );
4.1. Region Growing
363
Here we obtain a const reference to the thresholds by calling the GetOutput() method.
const CalculatorType::OutputType &thresholdVector = calculator->GetOutput();
We now iterate through thresholdVector, printing each value to the console and writing an image
thresholded with adjacent values from the container. (In the edge cases, the minimum and maximum
values of the InternalPixelType are used).
for( auto itNum = thresholdVector.begin();
itNum != thresholdVector.end();
++itNum )
{
std::cout << "OtsuThreshold["
<< (int)(itNum - thresholdVector.begin())
<< "] = "
<< static_cast::PrintType>(*itNum)
<< std::endl;
Also write out the image thresholded between the upper threshold and the max intensity.
upperThreshold = itk::NumericTraits::max();
filter->SetLowerThreshold( lowerThreshold );
filter->SetUpperThreshold( upperThreshold );
4.1.3 Neighborhood Connected
The source code for this section can be found in the file
NeighborhoodConnectedImageFilter.cxx.
The following example illustrates the use of the itk::NeighborhoodConnectedImageFilter.
This filter is a close variant of the itk::ConnectedThresholdImageFilter. On one hand, the
ConnectedThresholdImageFilter considers only the value of the pixel itself when determining
whether it belongs to the region: if its value is within the interval [lowerThreshold,upperThreshold]
it is included, otherwise it is excluded. NeighborhoodConnectedImageFilter, on the other hand,
considers a user-defined neighborhood surrounding the pixel, requiring that the intensity of each
neighbor be within the interval for it to be included.
The reason for considering the neighborhood intensities instead of only the current pixel intensity
is that small structures are less likely to be accepted in the region. The operation of this filter
is equivalent to applying ConnectedThresholdImageFilter followed by mathematical morphology erosion using a structuring element of the same shape as the neighborhood provided to the
NeighborhoodConnectedImageFilter.
364
Chapter 4. Segmentation
#include "itkNeighborhoodConnectedImageFilter.h"
The itk::CurvatureFlowImageFilter is used here to smooth the image while preserving edges.
#include "itkCurvatureFlowImageFilter.h"
We now define the image type using a particular pixel type and image dimension. In this case the
float type is used for the pixels due to the requirements of the smoothing filter.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The smoothing filter type is instantiated using the image type as a template parameter.
using CurvatureFlowImageFilterType =
itk::CurvatureFlowImageFilter;
Then, the filter is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
CurvatureFlowImageFilterType::Pointer smoothing =
CurvatureFlowImageFilterType::New();
We now declare the type of the region growing filter. In this case it is the NeighborhoodConnectedImageFilter.
using ConnectedFilterType =
itk::NeighborhoodConnectedImageFilter;
One filter of this class is constructed using the New() method.
ConnectedFilterType::Pointer neighborhoodConnected
= ConnectedFilterType::New();
Now it is time to create a simple, linear data processing pipeline. A file reader is added at the
beginning of the pipeline and a cast filter and writer are added at the end. The cast filter is required
to convert float pixel types to integer types since only a few image file formats support float
types.
4.1. Region Growing
365
smoothing->SetInput( reader->GetOutput() );
neighborhoodConnected->SetInput( smoothing->GetOutput() );
caster->SetInput( neighborhoodConnected->GetOutput() );
writer->SetInput( caster->GetOutput() );
CurvatureFlowImageFilter requires a couple of parameters. The following are typical values for
2D images. However, they may have to be adjusted depending on the amount of noise present in the
input image.
smoothing->SetNumberOfIterations( 5 );
smoothing->SetTimeStep( 0.125 );
NeighborhoodConnectedImageFilter requires that two main parameters are specified. They are
the lower and upper thresholds of the interval in which intensity values must fall to be included in
the region. Setting these two values too close will not allow enough flexibility for the region to grow.
Setting them too far apart will result in a region that engulfs the image.
neighborhoodConnected->SetLower( lowerThreshold );
neighborhoodConnected->SetUpper( upperThreshold );
Here, we add the crucial parameter that defines the neighborhood size used to determine whether a
pixel lies in the region. The larger the neighborhood, the more stable this filter will be against noise
in the input image, but also the longer the computing time will be. Here we select a filter of radius
2 along each dimension. This results in a neighborhood of 5 × 5 pixels.
InternalImageType::SizeType radius;
radius[0] = 2;
radius[1] = 2;
// two pixels along X
// two pixels along Y
neighborhoodConnected->SetRadius( radius );
As in the ConnectedThresholdImageFilter example, we must provide the intensity value to be
used for the output pixels accepted in the region and at least one seed point to define the starting
point.
neighborhoodConnected->SetSeed( index );
neighborhoodConnected->SetReplaceValue( 255 );
Invocation of the Update() method on the writer triggers the execution of the pipeline. It is usually
wise to put update calls in a try/catch block in case errors occur and exceptions are thrown.
366
Chapter 4. Segmentation
Figure 4.3: Segmentation results of the NeighborhoodConnectedImageFilter for various seed points.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
Now we’ll run this example using the image BrainProtonDensitySlice.png as input available
from the directory Examples/Data. We can easily segment the major anatomical structures by
providing seeds in the appropriate locations and defining values for the lower and upper thresholds.
For example
Structure
White matter
Ventricle
Gray matter
Seed Index
(60, 116)
(81, 112)
(107, 69)
Lower
150
210
180
Upper
180
250
210
Output Image
Second from left in Figure 4.3
Third from left in Figure 4.3
Fourth from left in Figure 4.3
As with the ConnectedThresholdImageFilter example, several seeds could be provided to
the filter by repetedly calling the AddSeed() method with different indices. Compare Figures
4.3 and 4.1, demonstrating the outputs of NeighborhoodConnectedThresholdImageFilter and
ConnectedThresholdImageFilter, respectively. It is instructive to adjust the neighborhood radii
and observe its effect on the smoothness of segmented object borders, size of the segmented region,
and computing time.
4.1.4 Confidence Connected
The source code for this section can be found in the file
ConfidenceConnected.cxx.
4.1. Region Growing
367
The following example illustrates the use of the itk::ConfidenceConnectedImageFilter. The
criterion used by the ConfidenceConnectedImageFilter is based on simple statistics of the current
region. First, the algorithm computes the mean and standard deviation of intensity values for all
the pixels currently included in the region. A user-provided factor is used to multiply the standard
deviation and define a range around the mean. Neighbor pixels whose intensity values fall inside the
range are accepted and included in the region. When no more neighbor pixels are found that satisfy
the criterion, the algorithm is considered to have finished its first iteration. At that point, the mean
and standard deviation of the intensity levels are recomputed using all the pixels currently included
in the region. This mean and standard deviation defines a new intensity range that is used to visit
current region neighbors and evaluate whether their intensity falls inside the range. This iterative
process is repeated until no more pixels are added or the maximum number of iterations is reached.
The following equation illustrates the inclusion criterion used by this filter,
I(X) ∈ [m − f σ, m + f σ]
(4.2)
where m and σ are the mean and standard deviation of the region intensities, f is a factor defined by
the user, I() is the image and X is the position of the particular neighbor pixel being considered for
inclusion in the region.
Let’s look at the minimal code required to use this algorithm. First, the following header defining
the itk::ConfidenceConnectedImageFilter class must be included.
#include "itkConfidenceConnectedImageFilter.h"
Noise present in the image can reduce the capacity of this filter to grow large regions. When faced
with noisy images, it is usually convenient to pre-process the image by using an edge-preserving
smoothing filter. Any of the filters discussed in Section 2.7.3 can be used to this end. In this
particular example we use the itk::CurvatureFlowImageFilter, hence we need to include its
header file.
#include "itkCurvatureFlowImageFilter.h"
We now define the image type using a pixel type and a particular dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The smoothing filter type is instantiated using the image type as a template parameter.
368
Chapter 4. Segmentation
using CurvatureFlowImageFilterType =
itk::CurvatureFlowImageFilter< InternalImageType, InternalImageType >;
Next the filter is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
CurvatureFlowImageFilterType::Pointer smoothing =
CurvatureFlowImageFilterType::New();
We now declare the type of the region growing filter.
ConfidenceConnectedImageFilter.
In this case it is the
using ConnectedFilterType = itk::ConfidenceConnectedImageFilter<
InternalImageType, InternalImageType>;
Then, we construct one filter of this class using the New() method.
ConnectedFilterType::Pointer confidenceConnected
= ConnectedFilterType::New();
Now it is time to create a simple, linear pipeline. A file reader is added at the beginning of the
pipeline and a cast filter and writer are added at the end. The cast filter is required here to convert
float pixel types to integer types since only a few image file formats support float types.
smoothing->SetInput( reader->GetOutput() );
confidenceConnected->SetInput( smoothing->GetOutput() );
caster->SetInput( confidenceConnected->GetOutput() );
writer->SetInput( caster->GetOutput() );
CurvatureFlowImageFilter requires two parameters. The following are typical values for 2D
images. However they may have to be adjusted depending on the amount of noise present in the
input image.
smoothing->SetNumberOfIterations( 5 );
smoothing->SetTimeStep( 0.125 );
ConfidenceConnectedImageFilter also requires two parameters. First, the factor f defines how
large the range of intensities will be. Small values of the multiplier will restrict the inclusion of
pixels to those having very similar intensities to those in the current region. Larger values of the
multiplier will relax the accepting condition and will result in more generous growth of the region.
Values that are too large will cause the region to grow into neighboring regions which may belong
to separate anatomical structures. This is not desirable behavior.
4.1. Region Growing
369
confidenceConnected->SetMultiplier( 2.5 );
The number of iterations is specified based on the homogeneity of the intensities of the anatomical
structure to be segmented. Highly homogeneous regions may only require a couple of iterations. Regions with ramp effects, like MRI images with inhomogeneous fields, may require more iterations.
In practice, it seems to be more important to carefully select the multiplier factor than the number
of iterations. However, keep in mind that there is no guarantee that this algorithm will converge on
a stable region. It is possible that by letting the algorithm run for more iterations the region will end
up engulfing the entire image.
confidenceConnected->SetNumberOfIterations( 5 );
The output of this filter is a binary image with zero-value pixels everywhere except on the extracted region. The intensity value to be set inside the region is selected with the method
SetReplaceValue().
confidenceConnected->SetReplaceValue( 255 );
The initialization of the algorithm requires the user to provide a seed point. It is convenient to select
this point to be placed in a typical region of the anatomical structure to be segmented. A small
neighborhood around the seed point will be used to compute the initial mean and standard deviation
for the inclusion criterion. The seed is passed in the form of an itk::Index to the SetSeed()
method.
confidenceConnected->SetSeed( index );
The size of the initial neighborhood around the seed is defined with the method
SetInitialNeighborhoodRadius(). The neighborhood will be defined as an N-dimensional rectangular region with 2r + 1 pixels on the side, where r is the value passed as initial neighborhood
radius.
confidenceConnected->SetInitialNeighborhoodRadius( 2 );
The invocation of the Update() method on the writer triggers the execution of the pipeline. It is
recommended to place update calls in a try/catch block in case errors occur and exceptions are
thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
370
Chapter 4. Segmentation
Figure 4.4: Segmentation results for the ConfidenceConnected filter for various seed points.
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
Let’s now run this example using as input the image BrainProtonDensitySlice.png provided in
the directory Examples/Data. We can easily segment the major anatomical structures by providing
seeds in the appropriate locations. For example
Structure
White matter
Ventricle
Gray matter
Seed Index
(60, 116)
(81, 112)
(107, 69)
Output Image
Second from left in Figure 4.4
Third from left in Figure 4.4
Fourth from left in Figure 4.4
Note that the gray matter is not being completely segmented. This illustrates the vulnerability of the
region growing methods when the anatomical structures to be segmented do not have a homogeneous
statistical distribution over the image space. You may want to experiment with different numbers of
iterations to verify how the accepted region will extend.
Application of the Confidence Connected filter on the Brain Web Data
This section shows some results obtained by applying the Confidence Connected filter on the BrainWeb database. The filter was applied on a 181 × 217 × 181 crosssection of the brainweb165a10f17
dataset. The data is a MR T1 acquisition, with an intensity non-uniformity of 20% and a slice thickness 1mm. The dataset may be obtained from http://www.bic.mni.mcgill.ca/brainweb/ or
https://data.kitware.com/#folder/5882712d8d777f4f3f3072df
The previous code was used in this example replacing the image dimension by 3. Gradient
Anistropic diffusion was applied to smooth the image. The filter used 2 iterations, a time step of
0.05 and a conductance value of 3. The smoothed volume was then segmented using the Confidence
4.1. Region Growing
371
Figure 4.5: White matter segmented using Confidence Connected region growing.
Connected approach. Five seed points were used at coordinate locations (118,85,92), (63,87,94),
(63,157,90), (111,188,90), (111,50,88). The ConfidenceConnnected filter used the parameters, a
neighborhood radius of 2, 5 iterations and an f of 2.5 (the same as in the previous example). The
results were then rendered using VolView.
Figure 4.5 shows the rendered volume. Figure 4.6 shows an axial, saggital and a coronal slice of the
volume.
Figure 4.6: Axial, sagittal and coronal slice segmented using Confidence Connected region growing.
372
Chapter 4. Segmentation
4.1.5 Isolated Connected
The source code for this section can be found in the file
IsolatedConnectedImageFilter.cxx.
The following example illustrates the use of the itk::IsolatedConnectedImageFilter. This
filter is a close variant of the itk::ConnectedThresholdImageFilter. In this filter two seeds
and a lower threshold are provided by the user. The filter will grow a region connected to the first
seed and not connected to the second one. In order to do this, the filter finds an intensity value that
could be used as upper threshold for the first seed. A binary search is used to find the value that
separates both seeds.
This example closely follows the previous ones. Only the relevant pieces of code are highlighted
here.
The header of the IsolatedConnectedImageFilter is included below.
#include "itkIsolatedConnectedImageFilter.h"
We define the image type using a pixel type and a particular dimension.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The IsolatedConnectedImageFilter is instantiated in the lines below.
using ConnectedFilterType =
itk::IsolatedConnectedImageFilter< InternalImageType,
InternalImageType >;
One filter of this class is constructed using the New() method.
ConnectedFilterType::Pointer isolatedConnected = ConnectedFilterType::New();
Now it is time to connect the pipeline.
smoothing->SetInput( reader->GetOutput() );
isolatedConnected->SetInput( smoothing->GetOutput() );
caster->SetInput( isolatedConnected->GetOutput() );
writer->SetInput( caster->GetOutput() );
The IsolatedConnectedImageFilter expects the user to specify a threshold and two seeds. In
this example, we take all of them from the command line arguments.
4.1. Region Growing
373
isolatedConnected->SetLower( lowerThreshold );
isolatedConnected->AddSeed1( indexSeed1 );
isolatedConnected->AddSeed2( indexSeed2 );
As in the itk::ConnectedThresholdImageFilter we must now specify the intensity value to be
set on the output pixels and at least one seed point to define the initial region.
isolatedConnected->SetReplaceValue( 255 );
The invocation of the Update() method on the writer triggers the execution of the pipeline.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
The intensity value allowing us to separate both regions can be recovered with the method
GetIsolatedValue().
std::cout << "Isolated Value Found = ";
std::cout << isolatedConnected->GetIsolatedValue() << std::endl;
Let’s now run this example using the image BrainProtonDensitySlice.png provided in the directory Examples/Data. We can easily segment the major anatomical structures by providing seed
pairs in the appropriate locations and defining values for the lower threshold. It is important to
keep in mind in this and the previous examples that the segmentation is being performed using the
smoothed version of the image. The selection of threshold values should therefore be performed
in the smoothed image since the distribution of intensities could be quite different from that of the
input image. As a reminder of this fact, Figure 4.7 presents, from left to right, the input image and
the result of smoothing with the itk::CurvatureFlowImageFilter followed by segmentation
results.
This filter is intended to be used in cases where adjacent anatomical structures are difficult to separate. Selecting one seed in one structure and the other seed in the adjacent structure creates the
appropriate setup for computing the threshold that will separate both structures. Table 4.2 presents
the parameters used to obtain the images shown in Figure 4.7.
374
Chapter 4. Segmentation
Adjacent Structures
Gray matter vs White matter
Seed1
(61, 140)
Seed2
(63, 43)
Lower
150
Isolated value found
183.31
Table 4.2: Parameters used for separating white matter from gray matter in Figure 4.7 using the IsolatedConnectedImageFilter.
Figure 4.7: Segmentation results of the IsolatedConnectedImageFilter.
4.1.6 Confidence Connected in Vector Images
The source code for this section can be found in the file
VectorConfidenceConnected.cxx.
This example illustrates the use of the confidence connected concept applied to images with
vector pixel types. The confidence connected algorithm is implemented for vector images
in the class itk::VectorConfidenceConnected. The basic difference between the scalar
and vector version is that the vector version uses the covariance matrix instead of a variance, and a vector mean instead of a scalar mean. The membership of a vector pixel
value to the region is measured using the Mahalanobis distance as implemented in the class
itk::Statistics::MahalanobisDistanceThresholdImageFunction.
#include "itkVectorConfidenceConnectedImageFilter.h"
We now define the image type using a particular pixel type and dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
constexpr unsigned int Dimension = 2;
using PixelComponentType = unsigned char;
4.1. Region Growing
375
using InputPixelType = itk::RGBPixel< PixelComponentType >;
using InputImageType = itk::Image< InputPixelType, Dimension >;
We now declare the type of the region-growing filter.
itk::VectorConfidenceConnectedImageFilter.
In this case it is the
using ConnectedFilterType =
itk::VectorConfidenceConnectedImageFilter< InputImageType,
OutputImageType >;
Then, we construct one filter of this class using the New() method.
ConnectedFilterType::Pointer confidenceConnected
= ConnectedFilterType::New();
Next we create a simple, linear data processing pipeline.
confidenceConnected->SetInput( reader->GetOutput() );
writer->SetInput( confidenceConnected->GetOutput() );
VectorConfidenceConnectedImageFilter requires two parameters. First, the multiplier factor
f defines how large the range of intensities will be. Small values of the multiplier will restrict the
inclusion of pixels to those having similar intensities to those already in the current region. Larger
values of the multiplier relax the accepting condition and result in more generous growth of the
region. Values that are too large will cause the region to grow into neighboring regions which may
actually belong to separate anatomical structures.
confidenceConnected->SetMultiplier( multiplier );
The number of iterations is typically determined based on the homogeneity of the image intensity
representing the anatomical structure to be segmented. Highly homogeneous regions may only
require a couple of iterations. Regions with ramp effects, like MRI images with inhomogeneous
fields, may require more iterations. In practice, it seems to be more relevant to carefully select
the multiplier factor than the number of iterations. However, keep in mind that there is no reason
to assume that this algorithm should converge to a stable region. It is possible that by letting the
algorithm run for more iterations the region will end up engulfing the entire image.
confidenceConnected->SetNumberOfIterations( iterations );
The output of this filter is a binary image with zero-value pixels everywhere except on the extracted region. The intensity value to be put inside the region is selected with the method
SetReplaceValue().
376
Chapter 4. Segmentation
confidenceConnected->SetReplaceValue( 255 );
The initialization of the algorithm requires the user to provide a seed point. This point should be
placed in a typical region of the anatomical structure to be segmented. A small neighborhood around
the seed point will be used to compute the initial mean and standard deviation for the inclusion
criterion. The seed is passed in the form of an itk::Index to the SetSeed() method.
confidenceConnected->SetSeed( index );
The size of the initial neighborhood around the seed is defined with the method
SetInitialNeighborhoodRadius(). The neighborhood will be defined as an N-Dimensional rectangular region with 2r + 1 pixels on the side, where r is the value passed as initial neighborhood
radius.
confidenceConnected->SetInitialNeighborhoodRadius( 3 );
The invocation of the Update() method on the writer triggers the execution of the pipeline. It is
usually wise to put update calls in a try/catch block in case errors occur and exceptions are thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
}
Now let’s run this example using as input the image VisibleWomanEyeSlice.png provided in the
directory Examples/Data. We can easily segment the major anatomical structures by providing
seeds in the appropriate locations. For example,
Structure
Rectum
Rectum
Vitreo
Seed Index
(70, 120)
(23, 93)
(66, 66)
Multiplier
7
7
3
Iterations
1
1
1
Output Image
Second from left in Figure 4.8
Third from left in Figure 4.8
Fourth from left in Figure 4.8
The coloration of muscular tissue makes it easy to distinguish them from the surrounding anatomical
structures. The optic vitrea on the other hand has a coloration that is not very homogeneous inside
the eyeball and does not facilitate a full segmentation based only on color.
The values of the final mean vector and covariance matrix used for the last iteration can be queried
using the methods GetMean() and GetCovariance().
4.2. Segmentation Based on Watersheds
377
Figure 4.8: Segmentation results of the VectorConfidenceConnected filter for various seed points.
using MeanVectorType = ConnectedFilterType::MeanVectorType;
using CovarianceMatrixType = ConnectedFilterType::CovarianceMatrixType;
const MeanVectorType & mean = confidenceConnected->GetMean();
const CovarianceMatrixType & covariance
= confidenceConnected->GetCovariance();
std::cout << "Mean vector = "
<< mean
<< std::endl;
std::cout << "Covariance matrix = " << covariance << std::endl;
4.2 Segmentation Based on Watersheds
4.2.1 Overview
Watershed segmentation classifies pixels into regions using gradient descent on image features and
analysis of weak points along region boundaries. Imagine water raining onto a landscape topology
and flowing with gravity to collect in low basins. The size of those basins will grow with increasing
amounts of precipitation until they spill into one another, causing small basins to merge together into
larger basins. Regions (catchment basins) are formed by using local geometric structure to associate
points in the image domain with local extrema in some feature measurement such as curvature or
gradient magnitude. This technique is less sensitive to user-defined thresholds than classic regiongrowing methods, and may be better suited for fusing different types of features from different data
sets. The watersheds technique is also more flexible in that it does not produce a single image
segmentation, but rather a hierarchy of segmentations from which a single region or set of regions
can be extracted a-priori, using a threshold, or interactively, with the help of a graphical user interface
[73, 74].
The strategy of watershed segmentation is to treat an image f as a height function, i.e., the surface
Chapter 4. Segmentation
Watershed Depth
378
Intensity profile of input image
Intensity profile of filtered image
Watershed Segmentation
Figure 4.9: A fuzzy-valued boundary map, from an image or set of images, is segmented using local minima
and catchment basins.
formed by graphing f as a function of its independent parameters, ~x ∈ U. The image f is often not
the original input data, but is derived from that data through some filtering, graded (or fuzzy) feature
extraction, or fusion of feature maps from different sources. The assumption is that higher values
of f (or − f ) indicate the presence of boundaries in the original data. Watersheds may therefore
be considered as a final or intermediate step in a hybrid segmentation method, where the initial
segmentation is the generation of the edge feature map.
Gradient descent associates regions with local minima of f (clearly interior points) using the watersheds of the graph of f , as in Figure 4.9. That is, a segment consists of all points in U whose paths
of steepest descent on the graph of f terminate at the same minimum in f . Thus, there are as many
segments in an image as there are minima in f . The segment boundaries are “ridges” [29, 30, 19]
in the graph of f . In the 1D case (U ⊂ ℜ), the watershed boundaries are the local maxima of f ,
and the results of the watershed segmentation is trivial. For higher-dimensional image domains,
the watershed boundaries are not simply local phenomena; they depend on the shape of the entire
watershed.
The drawback of watershed segmentation is that it produces a region for each local minimum—in
practice too many regions—and an over segmentation results. To alleviate this, we can establish a
minimum watershed depth. The watershed depth is the difference in height between the watershed
minimum and the lowest boundary point. In other words, it is the maximum depth of water a region
could hold without flowing into any of its neighbors. Thus, a watershed segmentation algorithm can
sequentially combine watersheds whose depths fall below the minimum until all of the watersheds
are of sufficient depth. This depth measurement can be combined with other saliency measurements,
such as size. The result is a segmentation containing regions whose boundaries and size are significant. Because the merging process is sequential, it produces a hierarchy of regions, as shown in
Figure 4.10. Previous work has shown the benefit of a user-assisted approach that provides a graphical interface to this hierarchy, so that a technician can quickly move from the small regions that lie
within an area of interest to the union of regions that correspond to the anatomical structure [74].
There are two different algorithms commonly used to implement watersheds: top-down and bottomup. The top-down, gradient descent strategy was chosen for ITK because we want to consider the
output of multi-scale differential operators, and the f in question will therefore have floating point
values. The bottom-up strategy starts with seeds at the local minima in the image and grows regions
outward and upward at discrete intensity levels (equivalent to a sequence of morphological operations and sometimes called morphological watersheds [55].) This limits the accuracy by enforcing
4.2. Segmentation Based on Watersheds
379
Image
Boolean Operations
on Sub−Trees
(e.g. User Interaction)
Node
Node
Node
Node
Threshold of
Watershed depth
Node
Node
Node
Node
Node
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Leaf
Figure 4.10: A watershed segmentation combined with a saliency measure (watershed depth) produces a
hierarchy of regions. Structures can be derived from images by either thresholding the saliency measure or
combining subtrees within the hierarchy.
a set of discrete gray levels on the image.
Figure 4.11 shows how the ITK image-to-image watersheds filter is constructed. The filter is actually
a collection of smaller filters that modularize the several steps of the algorithm in a mini-pipeline.
The segmenter object creates the initial segmentation via steepest descent from each pixel to local
minima. Shallow background regions are removed (flattened) before segmentation using a simple
minimum value threshold (this helps to minimize oversegmentation of the image). The initial segmentation is passed to a second sub-filter that generates a hierarchy of basins to a user-specified
maximum watershed depth. The relabeler object at the end of the mini-pipeline uses the hierarchy
and the initial segmentation to produce an output image at any scale below the user-specified maximum. Data objects are cached in the mini-pipeline so that changing watershed depths only requires
a (fast) relabeling of the basic segmentation. The three parameters that control the filter are shown
in Figure 4.11 connected to their relevant processing stages.
4.2.2 Using the ITK Watershed Filter
The source code for this section can be found in the file
WatershedSegmentation1.cxx.
The following example illustrates how to preprocess and segment images using the
itk::WatershedImageFilter. Note that the care with which the data are preprocessed will greatly
affect the quality of your result. Typically, the best results are obtained by preprocessing the original image with an edge-preserving diffusion filter, such as one of the anisotropic diffusion filters,
380
Chapter 4. Segmentation
Output Flood Level
Watershed Image Filter
Height
Image
Threshold
Segmenter
Basic
Segmentation
Tree
Generator
Image
Relabeler
Merge
Tree
Labeled
Image
Data Object
Process Object
Maximum Flood Level
Parameter
Figure 4.11: The construction of the Insight watersheds filter.
or the bilateral image filter. As noted in Section 4.2.1, the height function used as input should be
created such that higher positive values correspond to object boundaries. A suitable height function
for many applications can be generated as the gradient magnitude of the image to be segmented.
The itk::VectorGradientMagnitudeAnisotropicDiffusionImageFilter class is used to
smooth the image and the itk::VectorGradientMagnitudeImageFilter is used to generate
the height function. We begin by including all preprocessing filter header files and the header file
for the WatershedImageFilter. We use the vector versions of these filters because the input dataset is
a color image.
#include "itkVectorGradientAnisotropicDiffusionImageFilter.h"
#include "itkVectorGradientMagnitudeImageFilter.h"
#include "itkWatershedImageFilter.h"
We now declare the image and pixel types to use for instantiation of the filters. All of these filters expect real-valued pixel types in order to work properly. The preprocessing stages are applied directly
to the vector-valued data and the segmentation uses floating point scalar data. Images are converted
from RGB pixel type to numerical vector type using itk::VectorCastImageFilter.
using
using
using
using
using
using
RGBPixelType = itk::RGBPixel< unsigned char >;
RGBImageType = itk::Image< RGBPixelType, 2 >;
VectorPixelType = itk::Vector< float, 3 >;
VectorImageType = itk::Image< VectorPixelType, 2 >;
LabeledImageType = itk::Image< itk::IdentifierType, 2 >;
ScalarImageType = itk::Image< float, 2 >;
The various image processing filters are declared using the types created above and eventually used
in the pipeline.
using FileReaderType = itk::ImageFileReader< RGBImageType >;
using CastFilterType =
itk::VectorCastImageFilter< RGBImageType, VectorImageType >;
4.2. Segmentation Based on Watersheds
381
using DiffusionFilterType =
itk::VectorGradientAnisotropicDiffusionImageFilter<
VectorImageType, VectorImageType >;
using GradientMagnitudeFilterType =
itk::VectorGradientMagnitudeImageFilter;
using WatershedFilterType = itk::WatershedImageFilter;
Next we instantiate the filters and set their parameters. The first step in the image processing pipeline
is diffusion of the color input image using an anisotropic diffusion filter. For this class of filters, the
CFL condition requires that the time step be no more than 0.25 for two-dimensional images, and no
more than 0.125 for three-dimensional images. The number of iterations and the conductance term
will be taken from the command line. See Section 2.7.3 for more information on the ITK anisotropic
diffusion filters.
DiffusionFilterType::Pointer diffusion = DiffusionFilterType::New();
diffusion->SetNumberOfIterations( std::stoi(argv[4]) );
diffusion->SetConductanceParameter( std::stod(argv[3]) );
diffusion->SetTimeStep(0.125);
The ITK gradient magnitude filter for vector-valued images can optionally take several parameters.
Here we allow only enabling or disabling of principal component analysis.
GradientMagnitudeFilterType::Pointer
gradient = GradientMagnitudeFilterType::New();
gradient->SetUsePrincipleComponents(std::stoi(argv[7]));
Finally we set up the watershed filter. There are two parameters. Level controls watershed depth,
and Threshold controls the lower thresholding of the input. Both parameters are set as a percentage
(0.0 - 1.0) of the maximum depth in the input image.
WatershedFilterType::Pointer watershed = WatershedFilterType::New();
watershed->SetLevel( std::stod(argv[6]) );
watershed->SetThreshold( std::stod(argv[5]) );
The output of WatershedImageFilter is an image of unsigned long integer labels, where a label
denotes membership of a pixel in a particular segmented region. This format is not practical for
visualization, so for the purposes of this example, we will convert it to RGB pixels. RGB images
have the advantage that they can be saved as a simple png file and viewed using any standard image
viewer software. The itk::Functor::ScalarToRGBPixelFunctor class is a special function
object designed to hash a scalar value into an itk::RGBPixel. Plugging this functor into the
itk::UnaryFunctorImageFilter creates an image filter which converts scalar images to RGB
images.
382
Chapter 4. Segmentation
Figure 4.12: Segmented section of Visible Human female head and neck cryosection data. At left is the
original image. The image in the middle was generated with parameters: conductance = 2.0, iterations = 10,
threshold = 0.0, level = 0.05, principal components = on. The image on the right was generated with parameters:
conductance = 2.0, iterations = 10, threshold = 0.001, level = 0.15, principal components = off.
using ColorMapFunctorType =
itk::Functor::ScalarToRGBPixelFunctor;
using ColorMapFilterType =
itk::UnaryFunctorImageFilter;
ColorMapFilterType::Pointer colormapper = ColorMapFilterType::New();
The filters are connected into a single pipeline, with readers and writers at each end.
caster->SetInput(reader->GetOutput());
diffusion->SetInput(caster->GetOutput());
gradient->SetInput(diffusion->GetOutput());
watershed->SetInput(gradient->GetOutput());
colormapper->SetInput(watershed->GetOutput());
writer->SetInput(colormapper->GetOutput());
Tuning the filter parameters for any particular application is a process of trial and error. The threshold parameter can be used to great effect in controlling oversegmentation of the image. Raising the
threshold will generally reduce computation time and produce output with fewer and larger regions.
The trick in tuning parameters is to consider the scale level of the objects that you are trying to
segment in the image. The best time/quality trade-off will be achieved when the image is smoothed
and thresholded to eliminate features just below the desired scale.
Figure 4.12 shows output from the example code. The input image is taken from the Visible Human
female data around the right eye. The images on the right are colorized watershed segmentations
4.2. Segmentation Based on Watersheds
383
with parameters set to capture objects such as the optic nerve and lateral rectus muscles, which can
be seen just above and to the left and right of the eyeball. Note that a critical difference between the
two segmentations is the mode of the gradient magnitude calculation.
A note on the computational complexity of the watershed algorithm is warranted. Most of the
complexity of the ITK implementation lies in generating the hierarchy. Processing times for this
stage are non-linear with respect to the number of catchment basins in the initial segmentation. This
means that the amount of information contained in an image is more significant than the number of
pixels in the image. A very large, but very flat input take less time to segment than a very small, but
very detailed input.
384
Chapter 4. Segmentation
4.3 Level Set Segmentation
Zero Set f(x,y)=0
Interior
f(x,y) > 0
Exterior f(x,y) < 0
Figure 4.13: Concept of zero set in a level set.
The paradigm of the level set is that it is a numerical method for tracking the evolution of
contours and surfaces. Instead of manipulating
the contour directly, the contour is embedded
as the zero level set of a higher dimensional
function called the level-set function, ψ(X, t).
The level-set function is then evolved under the
control of a differential equation. At any time,
the evolving contour can be obtained by extracting the zero level-set Γ(X, t) = {ψ(X, t) =
0} from the output. The main advantages of using level sets is that arbitrarily complex shapes
can be modeled and topological changes such
as merging and splitting are handled implicitly.
Level sets can be used for image segmentation by using image-based features such as mean intensity, gradient and edges in the governing differential equation. In a typical approach, a contour is
initialized by a user and is then evolved until it fits the form of an anatomical structure in the image. Many different implementations and variants of this basic concept have been published in the
literature. An overview of the field has been made by Sethian [56].
The following sections introduce practical examples of some of the level set segmentation methods
available in ITK. The remainder of this section describes features common to all of these filters
except the itk::FastMarchingImageFilter, which is derived from a different code framework.
Understanding these features will aid in using the filters more effectively.
Each filter makes use of a generic level-set equation to compute the update to the solution ψ of the
partial differential equation.
d
ψ = −αA(x) · ∇ψ − βP(x) | ∇ψ | +γZ(x)κ | ∇ψ |
dt
(4.3)
where A is an advection term, P is a propagation (expansion) term, and Z is a spatial modifier term
for the mean curvature κ. The scalar constants α, β, and γ weight the relative influence of each of
the terms on the movement of the interface. A segmentation filter may use all of these terms in its
calculations, or it may omit one or more terms. If a term is left out of the equation, then setting the
corresponding scalar constant weighting will have no effect.
All of the level-set based segmentation filters must operate with floating point precision to produce
valid results. The third, optional template parameter is the numerical type used for calculations
and as the output image pixel type. The numerical type is float by default, but can be changed
to double for extra precision. A user-defined, signed floating point type that defines all of the
necessary arithmetic operators and has sufficient precision is also a valid choice. You should not use
4.3. Level Set Segmentation
385
Ψ(x, t)
−2.4 −1.3 −0.6 −0.7 −0.8 −1.8
−2.4 −1.4 −0.3 0.4 0.3 0.2 −0.8 −1.8
−2.4 −1.4 −0.4 0.6 1.6 1.3 1.2 0.2 −0.8 −1.8
−1.2 −0.2 0.8 1.8
2.3 1.3 0.3 −0.7
−1.1 −0.1 0.9 0.7 1.7
1.2 0.2 −0.8
−2.5 −1.5 −0.5 −0.3 0.7 2.4 1.4 0.4 −0.6
−2.5 −1.5 −1.3 −0.4 1.3 0.3 0.4 −0.6
−1.6 −0.6 0.4 −0.7 −0.6 −1.6
−1.6 −0.6 −1.7
Figure 4.14: The implicit level set surface Γ is the black line superimposed over the image grid. The location
of the surface is interpolated by the image pixel values. The grid pixels closest to the implicit surface are shown
in gray.
types such as int or unsigned char for the numerical parameter. If the input image pixel types
do not match the numerical type, those inputs will be cast to an image of appropriate type when the
filter is executed.
Most filters require two images as input, an initial model ψ(X, t = 0), and a feature image, which is
either the image you wish to segment or some preprocessed version. You must specify the isovalue
that represents the surface Γ in your initial model. The single image output of each filter is the
function ψ at the final time step. It is important to note that the contour representing the surface Γ
is the zero level-set of the output image, and not the isovalue you specified for the initial model. To
represent Γ using the original isovalue, simply add that value back to the output.
The solution Γ is calculated to subpixel precision. The best discrete approximation of the surface is
therefore the set of grid positions closest to the zero-crossings in the image, as shown in Figure 4.14.
The itk::ZeroCrossingImageFilter operates by finding exactly those grid positions and can be
used to extract the surface.
There are two important considerations when analyzing the processing time for any particular levelset segmentation task: the surface area of the evolving interface and the total distance that the surface
must travel. Because the level-set equations are usually solved only at pixels near the surface (fast
marching methods are an exception), the time taken at each iteration depends on the number of
points on the surface. This means that as the surface grows, the solver will slow down proportionally.
Because the surface must evolve slowly to prevent numerical instabilities in the solution, the distance
the surface must travel in the image dictates the total number of iterations required.
Some level-set techniques are relatively insensitive to initial conditions and are
therefore suitable for region-growing segmentation.
Other techniques, such as the
itk::LaplacianSegmentationLevelSetImageFilter, can easily become “stuck” on image
386
Chapter 4. Segmentation
Input
itk::Image
Iterations
Sigma
Alpha,Beta
Seeds
Anisotropic
Diffusion
Gradient
Magnitude
Sigmoid
Filter
Fast
Marching
Threshold
Time−Crossing
Map
Binary
Threshold
Binary
Image
Figure 4.15: Collaboration diagram of the FastMarchingImageFilter applied to a segmentation task.
features close to their initialization and should be used only when a reasonable prior segmentation
is available as the initialization. For best efficiency, your initial model of the surface should be the
best guess possible for the solution. When extending the example applications given here to higher
dimensional images, for example, you can improve results and dramatically decrease processing
time by using a multi-scale approach. Start with a downsampled volume and work back to the full
resolution using the results at each intermediate scale as the initialization for the next scale.
4.3.1 Fast Marching Segmentation
The source code for this section can be found in the file
FastMarchingImageFilter.cxx.
When the differential equation governing the level set evolution has a very simple form, a fast
evolution algorithm called fast marching can be used.
The following example illustrates the use of the itk::FastMarchingImageFilter. This filter
implements a fast marching solution to a simple level set evolution problem. In this example, the
speed term used in the differential equation is expected to be provided by the user in the form of an
image. This image is typically computed as a function of the gradient magnitude. Several mappings
are popular in the literature, for example, the negative exponential exp(−x) and the reciprocal 1/(1+
x). In the current example we decided to use a Sigmoid function since it offers a good number of
control parameters that can be customized to shape a nice speed image.
The mapping should be done in such a way that the propagation speed of the front will be very low
close to high image gradients while it will move rather fast in low gradient areas. This arrangement
will make the contour propagate until it reaches the edges of anatomical structures in the image
and then slow down in front of those edges. The output of the FastMarchingImageFilter is a timecrossing map that indicates, for each pixel, how much time it would take for the front to arrive at the
pixel location.
The application of a threshold in the output image is then equivalent to taking a snapshot of the
contour at a particular time during its evolution. It is expected that the contour will take a longer
time to cross over the edges of a particular anatomical structure. This should result in large changes
on the time-crossing map values close to the structure edges. Segmentation is performed with this
filter by locating a time range in which the contour was contained for a long time in a region of the
image space.
Figure 4.15 shows the major components involved in the application of the FastMarchingIm-
4.3. Level Set Segmentation
387
ageFilter to a segmentation task.
It involves an initial stage of smoothing using the
itk::CurvatureAnisotropicDiffusionImageFilter. The smoothed image is passed as
the input to the itk::GradientMagnitudeRecursiveGaussianImageFilter and then to the
itk::SigmoidImageFilter. Finally, the output of the FastMarchingImageFilter is passed to a
itk::BinaryThresholdImageFilter in order to produce a binary mask representing the segmented object.
The code in the following example illustrates the typical setup of a pipeline for performing segmentation with fast marching. First, the input image is smoothed using an edge-preserving filter. Then
the magnitude of its gradient is computed and passed to a sigmoid filter. The result of the sigmoid
filter is the image potential that will be used to affect the speed term of the differential equation.
Let’s start by including the following headers. First we include the header of the CurvatureAnisotropicDiffusionImageFilter that will be used for removing noise from the input image.
#include "itkCurvatureAnisotropicDiffusionImageFilter.h"
The headers of the GradientMagnitudeRecursiveGaussianImageFilter and SigmoidImageFilter are
included below. Together, these two filters will produce the image potential for regulating the speed
term in the differential equation describing the evolution of the level set.
#include "itkGradientMagnitudeRecursiveGaussianImageFilter.h"
#include "itkSigmoidImageFilter.h"
Of course, we will need the itk::Image class and the FastMarchingImageFilter class. Hence we
include their headers.
#include "itkFastMarchingImageFilter.h"
The time-crossing map resulting from the FastMarchingImageFilter will be thresholded using the
BinaryThresholdImageFilter. We include its header here.
#include "itkBinaryThresholdImageFilter.h"
Reading and writing images will be done with the
itk::ImageFileWriter.
itk::ImageFileReader and
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
We now define the image type using a pixel type and a particular dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
388
Chapter 4. Segmentation
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The output image, on the other hand, is declared to be binary.
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
The type of the BinaryThresholdImageFilter filter is instantiated below using the internal image type
and the output image type.
using ThresholdingFilterType =
itk::BinaryThresholdImageFilter< InternalImageType, OutputImageType >;
ThresholdingFilterType::Pointer thresholder = ThresholdingFilterType::New();
The upper threshold passed to the BinaryThresholdImageFilter will define the time snapshot that we
are taking from the time-crossing map. In an ideal application the user should be able to select this
threshold interactively using visual feedback. Here, since it is a minimal example, the value is taken
from the command line arguments.
thresholder->SetLowerThreshold(
0.0 );
thresholder->SetUpperThreshold( timeThreshold );
thresholder->SetOutsideValue( 0 );
thresholder->SetInsideValue( 255 );
We instantiate reader and writer types in the following lines.
using ReaderType = itk::ImageFileReader< InternalImageType >;
using WriterType = itk::ImageFileWriter< OutputImageType >;
The CurvatureAnisotropicDiffusionImageFilter type is instantiated using the internal image type.
using SmoothingFilterType = itk::CurvatureAnisotropicDiffusionImageFilter<
InternalImageType,
InternalImageType >;
Then, the filter is created by invoking the New() method and assigning the result to a
itk::SmartPointer.
4.3. Level Set Segmentation
389
SmoothingFilterType::Pointer smoothing = SmoothingFilterType::New();
The types of the GradientMagnitudeRecursiveGaussianImageFilter and SigmoidImageFilter are instantiated using the internal image type.
using GradientFilterType =
itk::GradientMagnitudeRecursiveGaussianImageFilter<
InternalImageType,
InternalImageType >;
using SigmoidFilterType = itk::SigmoidImageFilter<
InternalImageType,
InternalImageType >;
The corresponding filter objects are instantiated with the New() method.
GradientFilterType::Pointer gradientMagnitude = GradientFilterType::New();
SigmoidFilterType::Pointer sigmoid = SigmoidFilterType::New();
The minimum and maximum values of the SigmoidImageFilter output are defined with the methods
SetOutputMinimum() and SetOutputMaximum(). In our case, we want these two values to be
0.0 and 1.0 respectively in order to get a nice speed image to feed to the FastMarchingImageFilter.
Additional details on the use of the SigmoidImageFilter are presented in Section 2.3.2.
sigmoid->SetOutputMinimum( 0.0
sigmoid->SetOutputMaximum( 1.0
);
);
We now declare the type of the FastMarchingImageFilter.
using FastMarchingFilterType =
itk::FastMarchingImageFilter< InternalImageType,
InternalImageType >;
Then, we construct one filter of this class using the New() method.
FastMarchingFilterType::Pointer fastMarching
= FastMarchingFilterType::New();
The filters are now connected in a pipeline shown in Figure 4.15 using the following lines.
smoothing->SetInput( reader->GetOutput() );
gradientMagnitude->SetInput( smoothing->GetOutput() );
sigmoid->SetInput( gradientMagnitude->GetOutput() );
fastMarching->SetInput( sigmoid->GetOutput() );
thresholder->SetInput( fastMarching->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
390
Chapter 4. Segmentation
The CurvatureAnisotropicDiffusionImageFilter class requires a couple of parameters to be defined.
The following are typical values for 2D images. However they may have to be adjusted depending
on the amount of noise present in the input image. This filter has been discussed in Section 2.7.3.
smoothing->SetTimeStep( 0.125 );
smoothing->SetNumberOfIterations( 5 );
smoothing->SetConductanceParameter( 9.0 );
The GradientMagnitudeRecursiveGaussianImageFilter performs the equivalent of a convolution
with a Gaussian kernel followed by a derivative operator. The sigma of this Gaussian can be used to
control the range of influence of the image edges. This filter has been discussed in Section 2.4.2.
gradientMagnitude->SetSigma( sigma
);
The SigmoidImageFilter class requires two parameters to define the linear transformation to be applied to the sigmoid argument. These parameters are passed using the SetAlpha() and SetBeta()
methods. In the context of this example, the parameters are used to intensify the differences between
regions of low and high values in the speed image. In an ideal case, the speed value should be 1.0 in
the homogeneous regions of anatomical structures and the value should decay rapidly to 0.0 around
the edges of structures. The heuristic for finding the values is the following: From the gradient
magnitude image, let’s call K1 the minimum value along the contour of the anatomical structure to
be segmented. Then, let’s call K2 an average value of the gradient magnitude in the middle of the
structure. These two values indicate the dynamic range that we want to map to the interval [0 : 1] in
the speed image. We want the sigmoid to map K1 to 0.0 and K2 to 1.0. Given that K1 is expected to
be higher than K2 and we want to map those values to 0.0 and 1.0 respectively, we want to select a
negative value for alpha so that the sigmoid function will also do an inverse intensity mapping. This
mapping will produce a speed image such that the level set will march rapidly on the homogeneous
region and will definitely stop on the contour. The suggested value for beta is (K1 + K2)/2 while the
suggested value for alpha is (K2 − K1)/6, which must be a negative number. In our simple example
the values are provided by the user from the command line arguments. The user can estimate these
values by observing the gradient magnitude image.
sigmoid->SetAlpha( alpha );
sigmoid->SetBeta( beta );
The FastMarchingImageFilter requires the user to provide a seed point from which the contour will
expand. The user can actually pass not only one seed point but a set of them. A good set of seed
points increases the chances of segmenting a complex object without missing parts. The use of
multiple seeds also helps to reduce the amount of time needed by the front to visit a whole object
and hence reduces the risk of leaks on the edges of regions visited earlier. For example, when
segmenting an elongated object, it is undesirable to place a single seed at one extreme of the object
since the front will need a long time to propagate to the other end of the object. Placing several
seeds along the axis of the object will probably be the best strategy to ensure that the entire object
4.3. Level Set Segmentation
391
is captured early in the expansion of the front. One of the important properties of level sets is their
natural ability to fuse several fronts implicitly without any extra bookkeeping. The use of multiple
seeds takes good advantage of this property.
The seeds are passed stored in a container. The type of this container is defined as NodeContainer
among the FastMarchingImageFilter traits.
using NodeContainer = FastMarchingFilterType::NodeContainer;
using NodeType = FastMarchingFilterType::NodeType;
NodeContainer::Pointer seeds = NodeContainer::New();
Nodes are created as stack variables and initialized with a value and an itk::Index position.
NodeType node;
constexpr double seedValue = 0.0;
node.SetValue( seedValue );
node.SetIndex( seedPosition );
The list of nodes is initialized and then every node is inserted using the InsertElement().
seeds->Initialize();
seeds->InsertElement( 0, node );
The set of seed nodes is now passed to the FastMarchingImageFilter with the method
SetTrialPoints().
fastMarching->SetTrialPoints( seeds
);
The FastMarchingImageFilter requires the user to specify the size of the image to be produced as
output. This is done using the SetOutputSize() method. Note that the size is obtained here from
the output image of the smoothing filter. The size of this image is valid only after the Update()
method of this filter has been called directly or indirectly.
fastMarching->SetOutputSize(
reader->GetOutput()->GetBufferedRegion().GetSize() );
Since the front representing the contour will propagate continuously over time, it is desirable to stop
the process once a certain time has been reached. This allows us to save computation time under
the assumption that the region of interest has already been computed. The value for stopping the
process is defined with the method SetStoppingValue(). In principle, the stopping value should
be a little bit higher than the threshold value.
392
Chapter 4. Segmentation
Structure
Left Ventricle
Right Ventricle
White matter
Gray matter
Seed Index
(81, 114)
(99, 114)
(56, 92)
(40, 90)
σ
1.0
1.0
1.0
0.5
α
-0.5
-0.5
-0.3
-0.3
β
3.0
3.0
2.0
2.0
Threshold
100
100
200
200
Output Image from left
First
Second
Third
Fourth
Table 4.3: Parameters used for segmenting some brain structures shown in Figure 4.17 using the filter FastMarchingImageFilter. All of them used a stopping value of 100.
fastMarching->SetStoppingValue( stoppingTime
);
The invocation of the Update() method on the writer triggers the execution of the pipeline. As
usual, the call is placed in a try/catch block should any errors occur or exceptions be thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Now let’s run this example using the input image BrainProtonDensitySlice.png provided in
the directory Examples/Data. We can easily segment the major anatomical structures by providing seeds in the appropriate locations. The following table presents the parameters used for some
structures.
Figure 4.16 presents the intermediate outputs of the pipeline illustrated in Figure 4.15. They are from
left to right: the output of the anisotropic diffusion filter, the gradient magnitude of the smoothed
image and the sigmoid of the gradient magnitude which is finally used as the speed image for the
FastMarchingImageFilter.
Notice that the gray matter is not being completely segmented. This illustrates the vulnerability
of the level set methods when the anatomical structures to be segmented do not occupy extended
regions of the image. This is especially true when the width of the structure is comparable to the size
of the attenuation bands generated by the gradient filter. A possible workaround for this limitation is
to use multiple seeds distributed along the elongated object. However, note that white matter versus
gray matter segmentation is not a trivial task, and may require a more elaborate approach than the
one used in this basic example.
4.3. Level Set Segmentation
393
Figure 4.16: Images generated by the segmentation process based on the FastMarchingImageFilter. From left
to right and top to bottom: input image to be segmented, image smoothed with an edge-preserving smoothing
filter, gradient magnitude of the smoothed image, sigmoid of the gradient magnitude. This last image, the
sigmoid, is used to compute the speed term for the front propagation.
394
Chapter 4. Segmentation
Figure 4.17: Images generated by the segmentation process based on the FastMarchingImageFilter. From left
to right: segmentation of the left ventricle, segmentation of the right ventricle, segmentation of the white matter,
attempt of segmentation of the gray matter.
4.3.2 Shape Detection Segmentation
The source code for this section can be found in the file
ShapeDetectionLevelSetFilter.cxx.
The use of the itk::ShapeDetectionLevelSetImageFilter is illustrated in the following example. The implementation of this filter in ITK is based on the paper by Malladi et al [38]. In this implementation, the governing differential equation has an additional curvature-based term. This term
acts as a smoothing term where areas of high curvature, assumed to be due to noise, are smoothed
out. Scaling parameters are used to control the tradeoff between the expansion term and the smoothing term. One consequence of this additional curvature term is that the fast marching algorithm is
no longer applicable, because the contour is no longer guaranteed to always be expanding. Instead,
the level set function is updated iteratively.
The ShapeDetectionLevelSetImageFilter expects two inputs, the first being an initial Level Set in
the form of an itk::Image, and the second being a feature image. For this algorithm, the feature
image is an edge potential image that basically follows the same rules applicable to the speed image
used for the FastMarchingImageFilter discussed in Section 4.3.1.
In this example we use an FastMarchingImageFilter to produce the initial level set as the distance
function to a set of user-provided seeds. The FastMarchingImageFilter is run with a constant speed
value which enables us to employ this filter as a distance map calculator.
Figure 4.18 shows the major components involved in the application of the ShapeDetectionLevelSetImageFilter to a segmentation task. The first stage involves smoothing using the
itk::CurvatureAnisotropicDiffusionImageFilter. The smoothed image is passed as the
input for the itk::GradientMagnitudeRecursiveGaussianImageFilter and then to the
itk::SigmoidImageFilter in order to produce the edge potential image. A set of user-provided
seeds is passed to an FastMarchingImageFilter in order to compute the distance map. A constant
value is subtracted from this map in order to obtain a level set in which the zero set represents the
4.3. Level Set Segmentation
Input
itk::Image
395
Iterations
Sigma
Alpha,Beta
Anisotropic
Diffusion
Gradient
Magnitude
Sigmoid
Filter
Edge
Image
Seeds
Fast
Marching
Distance
Input
LevelSet
Threshold
Edge
Potential
Shape
Detection
Output
LevelSet
Binary
Threshold
Binary
Image
Input
LevelSet
Figure 4.18: Collaboration diagram for the ShapeDetectionLevelSetImageFilter applied to a segmentation task.
396
Chapter 4. Segmentation
initial contour. This level set is also passed as input to the ShapeDetectionLevelSetImageFilter.
Finally, the level set at the output of the ShapeDetectionLevelSetImageFilter is passed to an BinaryThresholdImageFilter in order to produce a binary mask representing the segmented object.
Let’s start by including the headers of the main filters involved in the preprocessing.
#include "itkCurvatureAnisotropicDiffusionImageFilter.h"
#include "itkGradientMagnitudeRecursiveGaussianImageFilter.h"
#include "itkSigmoidImageFilter.h"
The edge potential map is generated using these filters as in the previous example.
We will need the Image class, the FastMarchingImageFilter class and the ShapeDetectionLevelSetImageFilter class. Hence we include their headers here.
#include "itkFastMarchingImageFilter.h"
#include "itkShapeDetectionLevelSetImageFilter.h"
The level set resulting from the ShapeDetectionLevelSetImageFilter will be thresholded at the zero
level in order to get a binary image representing the segmented object. The BinaryThresholdImageFilter is used for this purpose.
#include "itkBinaryThresholdImageFilter.h"
We now define the image type using a particular pixel type and a dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The output image, on the other hand, is declared to be binary.
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
The type of the BinaryThresholdImageFilter filter is instantiated below using the internal image type
and the output image type.
using ThresholdingFilterType =
itk::BinaryThresholdImageFilter< InternalImageType, OutputImageType >;
ThresholdingFilterType::Pointer thresholder = ThresholdingFilterType::New();
4.3. Level Set Segmentation
397
The upper threshold of the BinaryThresholdImageFilter is set to 0.0 in order to display the zero set
of the resulting level set. The lower threshold is set to a large negative number in order to ensure
that the interior of the segmented object will appear inside the binary region.
thresholder->SetLowerThreshold( -1000.0 );
thresholder->SetUpperThreshold(
0.0 );
thresholder->SetOutsideValue( 0 );
thresholder->SetInsideValue( 255 );
The CurvatureAnisotropicDiffusionImageFilter type is instantiated using the internal image type.
using SmoothingFilterType = itk::CurvatureAnisotropicDiffusionImageFilter<
InternalImageType,
InternalImageType >;
The filter is instantiated by invoking the New() method and assigning the result to a
itk::SmartPointer.
SmoothingFilterType::Pointer smoothing = SmoothingFilterType::New();
The types of the GradientMagnitudeRecursiveGaussianImageFilter and SigmoidImageFilter are instantiated using the internal image type.
using GradientFilterType =
itk::GradientMagnitudeRecursiveGaussianImageFilter<
InternalImageType,
InternalImageType >;
using SigmoidFilterType = itk::SigmoidImageFilter<
InternalImageType,
InternalImageType >;
The corresponding filter objects are created with the method New().
GradientFilterType::Pointer gradientMagnitude = GradientFilterType::New();
SigmoidFilterType::Pointer sigmoid = SigmoidFilterType::New();
The minimum and maximum values of the SigmoidImageFilter output are defined with the methods
SetOutputMinimum() and SetOutputMaximum(). In our case, we want these two values to be
0.0 and 1.0 respectively in order to get a nice speed image to feed to the FastMarchingImageFilter.
Additional details on the use of the SigmoidImageFilter are presented in Section 2.3.2.
398
sigmoid->SetOutputMinimum( 0.0
sigmoid->SetOutputMaximum( 1.0
Chapter 4. Segmentation
);
);
We now declare the type of the FastMarchingImageFilter that will be used to generate the initial
level set in the form of a distance map.
using FastMarchingFilterType =
itk::FastMarchingImageFilter< InternalImageType, InternalImageType >;
Next we construct one filter of this class using the New() method.
FastMarchingFilterType::Pointer fastMarching
= FastMarchingFilterType::New();
In the following lines we instantiate the type of the ShapeDetectionLevelSetImageFilter and create
an object of this type using the New() method.
using ShapeDetectionFilterType =
itk::ShapeDetectionLevelSetImageFilter< InternalImageType,
InternalImageType >;
ShapeDetectionFilterType::Pointer
shapeDetection = ShapeDetectionFilterType::New();
The filters are now connected in a pipeline indicated in Figure 4.18 with the following code.
smoothing->SetInput( reader->GetOutput() );
gradientMagnitude->SetInput( smoothing->GetOutput() );
sigmoid->SetInput( gradientMagnitude->GetOutput() );
shapeDetection->SetInput( fastMarching->GetOutput() );
shapeDetection->SetFeatureImage( sigmoid->GetOutput() );
thresholder->SetInput( shapeDetection->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
The CurvatureAnisotropicDiffusionImageFilter requires a couple of parameters to be defined. The
following are typical values for 2D images. However they may have to be adjusted depending on
the amount of noise present in the input image. This filter has been discussed in Section 2.7.3.
smoothing->SetTimeStep( 0.125 );
smoothing->SetNumberOfIterations( 5 );
smoothing->SetConductanceParameter( 9.0 );
4.3. Level Set Segmentation
399
The GradientMagnitudeRecursiveGaussianImageFilter performs the equivalent of a convolution
with a Gaussian kernel followed by a derivative operator. The sigma of this Gaussian can be used to
control the range of influence of the image edges. This filter has been discussed in Section 2.4.2.
gradientMagnitude->SetSigma( sigma
);
The SigmoidImageFilter requires two parameters that define the linear transformation to be applied
to the sigmoid argument. These parameters have been discussed in Sections 2.3.2 and 4.3.1.
sigmoid->SetAlpha( alpha );
sigmoid->SetBeta( beta );
The FastMarchingImageFilter requires the user to provide a seed point from which the level set
will be generated. The user can actually pass not only one seed point but a set of them. Note the
FastMarchingImageFilter is used here only as a helper in the determination of an initial level set.
We could have used the itk::DanielssonDistanceMapImageFilter in the same way.
The seeds are stored in a container. The type of this container is defined as NodeContainer among
the FastMarchingImageFilter traits.
using NodeContainer = FastMarchingFilterType::NodeContainer;
using NodeType = FastMarchingFilterType::NodeType;
NodeContainer::Pointer seeds = NodeContainer::New();
Nodes are created as stack variables and initialized with a value and an itk::Index position. Note
that we assign the negative of the value of the user-provided distance to the unique node of the
seeds passed to the FastMarchingImageFilter. In this way, the value will increment as the front is
propagated, until it reaches the zero value corresponding to the contour. After this, the front will
continue propagating until it fills up the entire image. The initial distance is taken from the command
line arguments. The rule of thumb for the user is to select this value as the distance from the seed
points at which the initial contour should be.
NodeType node;
const double seedValue = - initialDistance;
node.SetValue( seedValue );
node.SetIndex( seedPosition );
The list of nodes is initialized and then every node is inserted using InsertElement().
seeds->Initialize();
seeds->InsertElement( 0, node );
The set of seed nodes is now passed to the FastMarchingImageFilter with the method
SetTrialPoints().
400
fastMarching->SetTrialPoints( seeds
Chapter 4. Segmentation
);
Since the FastMarchingImageFilter is used here only as a distance map generator, it does not require
a speed image as input. Instead, the constant value 1.0 is passed using the SetSpeedConstant()
method.
fastMarching->SetSpeedConstant( 1.0 );
The FastMarchingImageFilter requires the user to specify the size of the image to be produced as
output. This is done using the SetOutputSize(). Note that the size is obtained here from the output
image of the smoothing filter. The size of this image is valid only after the Update() methods of
this filter have been called directly or indirectly.
fastMarching->SetOutputSize(
reader->GetOutput()->GetBufferedRegion().GetSize() );
ShapeDetectionLevelSetImageFilter provides two parameters to control the competition between the propagation or expansion term and the curvature smoothing term. The methods
SetPropagationScaling() and SetCurvatureScaling() defines the relative weighting between
the two terms. In this example, we will set the propagation scaling to one and let the curvature scaling be an input argument. The larger the the curvature scaling parameter the smoother the resulting
segmentation. However, the curvature scaling parameter should not be set too large, as it will draw
the contour away from the shape boundaries.
shapeDetection->SetPropagationScaling( propagationScaling );
shapeDetection->SetCurvatureScaling( curvatureScaling );
Once activated, the level set evolution will stop if the convergence criteria or the maximum number
of iterations is reached. The convergence criteria are defined in terms of the root mean squared
(RMS) change in the level set function. The evolution is said to have converged if the RMS change
is below a user-specified threshold. In a real application, it is desirable to couple the evolution of
the zero set to a visualization module, allowing the user to follow the evolution of the zero set. With
this feedback, the user may decide when to stop the algorithm before the zero set leaks through the
regions of low gradient in the contour of the anatomical structure to be segmented.
shapeDetection->SetMaximumRMSError( 0.02 );
shapeDetection->SetNumberOfIterations( 800 );
The invocation of the Update() method on the writer triggers the execution of the pipeline. As
usual, the call is placed in a try/catch block should any errors occur or exceptions be thrown.
4.3. Level Set Segmentation
Structure
Left Ventricle
Right Ventricle
White matter
Gray matter
Seed Index
(81, 114)
(99, 114)
(56, 92)
(40, 90)
401
Distance
5.0
5.0
5.0
5.0
σ
1.0
1.0
1.0
0.5
α
-0.5
-0.5
-0.3
-0.3
β
3.0
3.0
2.0
2.0
Output Image
First in Figure 4.20
Second in Figure 4.20
Third in Figure 4.20
Fourth in Figure 4.20
Table 4.4: Parameters used for segmenting some brain structures shown in Figure 4.19 using the filter ShapeDetectionLevelSetFilter. All of them used a propagation scaling of 1.0 and curvature scaling of 0.05.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Let’s now run this example using as input the image BrainProtonDensitySlice.png provided in
the directory Examples/Data. We can easily segment the major anatomical structures by providing
seeds in the appropriate locations. Table 4.4 presents the parameters used for some structures. For
all of the examples illustrated in this table, the propagation scaling was set to 1.0, and the curvature
scaling set to 0.05.
Figure 4.19 presents the intermediate outputs of the pipeline illustrated in Figure 4.18. They are from
left to right: the output of the anisotropic diffusion filter, the gradient magnitude of the smoothed
image and the sigmoid of the gradient magnitude which is finally used as the edge potential for the
ShapeDetectionLevelSetImageFilter.
Notice that in Figure 4.20 the segmented shapes are rounder than in Figure 4.17 due to the effects of
the curvature term in the driving equation. As with the previous example, segmentation of the gray
matter is still problematic.
A larger number of iterations is reguired for segmenting large structures since it takes longer for the
front to propagate and cover the structure. This drawback can be easily mitigated by setting many
seed points in the initialization of the FastMarchingImageFilter. This will generate an initial level
set much closer in shape to the object to be segmented and hence require fewer iterations to fill and
reach the edges of the anatomical structure.
402
Chapter 4. Segmentation
Figure 4.19: Images generated by the segmentation process based on the ShapeDetectionLevelSetImageFilter. From left to right and top to bottom: input image to be segmented, image smoothed with an edgepreserving smoothing filter, gradient magnitude of the smoothed image, sigmoid of the gradient magnitude. This
last image, the sigmoid, is used to compute the speed term for the front propagation.
4.3. Level Set Segmentation
403
Figure 4.20: Images generated by the segmentation process based on the ShapeDetectionLevelSetImageFilter. From left to right: segmentation of the left ventricle, segmentation of the right ventricle, segmentation of
the white matter, attempt of segmentation of the gray matter.
4.3.3 Geodesic Active Contours Segmentation
The source code for this section can be found in the file
GeodesicActiveContourImageFilter.cxx.
The use of the itk::GeodesicActiveContourLevelSetImageFilter is illustrated in the following example. The implementation of this filter in ITK is based on the paper by Caselles [11]. This
implementation extends the functionality of the itk::ShapeDetectionLevelSetImageFilter by
the addition of a third advection term which attracts the level set to the object boundaries.
GeodesicActiveContourLevelSetImageFilter expects two inputs. The first is an initial level set in
the form of an itk::Image. The second input is a feature image. For this algorithm, the feature
image is an edge potential image that basically follows the same rules used for the ShapeDetectionLevelSetImageFilter discussed in Section 4.3.2. The configuration of this example is quite similar to
the example on the use of the ShapeDetectionLevelSetImageFilter. We omit most of the redundant
description. A look at the code will reveal the great degree of similarity between both examples.
Figure 4.21 shows the major components involved in the application of the GeodesicActiveContourLevelSetImageFilter to a segmentation task. This pipeline is quite similar to the one used by the
ShapeDetectionLevelSetImageFilter in section 4.3.2.
The pipeline involves a first stage of smoothing using the
itk::CurvatureAnisotropicDiffusionImageFilter. The smoothed image is passed as
the input to the itk::GradientMagnitudeRecursiveGaussianImageFilter and then to the
itk::SigmoidImageFilter in order to produce the edge potential image. A set of user-provided
seeds is passed to a itk::FastMarchingImageFilter in order to compute the distance map. A
constant value is subtracted from this map in order to obtain a level set in which the zero set represents the initial contour. This level set is also passed as input to the GeodesicActiveContourLevelSetImageFilter.
Finally, the level set generated by the GeodesicActiveContourLevelSetImageFilter is passed to a
404
Chapter 4. Segmentation
Input
itk::Image
Iterations
Sigma
Alpha,Beta
Anisotropic
Diffusion
Gradient
Magnitude
Sigmoid
Filter
Edge
Image
Seeds
Fast
Marching
Distance
Inflation
Strength
Length
Penalty
Geodesic
Active
Contours
Input
LevelSet
Output
LevelSet
Binary
Threshold
Binary
Image
Figure 4.21: Collaboration diagram for the GeodesicActiveContourLevelSetImageFilter applied to a segmentation task.
4.3. Level Set Segmentation
405
itk::BinaryThresholdImageFilter in order to produce a binary mask representing the segmented object.
Let’s start by including the headers of the main filters involved in the preprocessing.
#include "itkGeodesicActiveContourLevelSetImageFilter.h"
We now define the image type using a particular pixel type and dimension. In this case the float
type is used for the pixels due to the requirements of the smoothing filter.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
In the following lines we instantiate the type of the GeodesicActiveContourLevelSetImageFilter and
create an object of this type using the New() method.
using GeodesicActiveContourFilterType =
itk::GeodesicActiveContourLevelSetImageFilter< InternalImageType,
InternalImageType >;
GeodesicActiveContourFilterType::Pointer geodesicActiveContour =
GeodesicActiveContourFilterType::New();
For the GeodesicActiveContourLevelSetImageFilter, scaling parameters are used to trade off between the propagation (inflation), the curvature (smoothing) and the advection terms. These
parameters are set using methods SetPropagationScaling(), SetCurvatureScaling() and
SetAdvectionScaling(). In this example, we will set the curvature and advection scales to one
and let the propagation scale be a command-line argument.
geodesicActiveContour->SetPropagationScaling( propagationScaling );
geodesicActiveContour->SetCurvatureScaling( 1.0 );
geodesicActiveContour->SetAdvectionScaling( 1.0 );
The filters are now connected in a pipeline indicated in Figure 4.21 using the following lines:
smoothing->SetInput( reader->GetOutput() );
gradientMagnitude->SetInput( smoothing->GetOutput() );
sigmoid->SetInput( gradientMagnitude->GetOutput() );
geodesicActiveContour->SetInput( fastMarching->GetOutput() );
geodesicActiveContour->SetFeatureImage( sigmoid->GetOutput() );
thresholder->SetInput( geodesicActiveContour->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
406
Chapter 4. Segmentation
Structure
Left Ventricle
Right Ventricle
White matter
Gray matter
Seed Index
(81, 114)
(99, 114)
(56, 92)
(40, 90)
Distance
5.0
5.0
5.0
5.0
σ
1.0
1.0
1.0
0.5
α
-0.5
-0.5
-0.3
-0.3
β
3.0
3.0
2.0
2.0
Propag.
2.0
2.0
10.0
10.0
Output Image
First
Second
Third
Fourth
Table 4.5: Parameters used for segmenting some brain structures shown in Figure 4.23 using the filter GeodesicActiveContourLevelSetImageFilter.
The invocation of the Update() method on the writer triggers the execution of the pipeline. As
usual, the call is placed in a try/catch block should any errors occur or exceptions be thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Let’s now run this example using as input the image BrainProtonDensitySlice.png provided in
the directory Examples/Data. We can easily segment the major anatomical structures by providing
seeds in the appropriate locations. Table 4.5 presents the parameters used for some structures.
Figure 4.22 presents the intermediate outputs of the pipeline illustrated in Figure 4.21. They are from
left to right: the output of the anisotropic diffusion filter, the gradient magnitude of the smoothed
image and the sigmoid of the gradient magnitude which is finally used as the edge potential for the
GeodesicActiveContourLevelSetImageFilter.
Segmentations of the main brain structures are presented in Figure 4.23. The results are quite similar
to those obtained with the ShapeDetectionLevelSetImageFilter in Section 4.3.2.
Note that a relatively larger propagation scaling value was required to segment the white matter. This
is due to two factors: the lower contrast at the border of the white matter and the complex shape of
the structure. Unfortunately the optimal value of these scaling parameters can only be determined
by experimentation. In a real application we could imagine an interactive mechanism by which a
user supervises the contour evolution and adjusts these parameters accordingly.
4.3. Level Set Segmentation
407
Figure 4.22: Images generated by the segmentation process based on the GeodesicActiveContourLevelSetImageFilter. From left to right and top to bottom: input image to be segmented, image smoothed with an edgepreserving smoothing filter, gradient magnitude of the smoothed image, sigmoid of the gradient magnitude. This
last image, the sigmoid, is used to compute the speed term for the front propagation.
408
Chapter 4. Segmentation
Figure 4.23: Images generated by the segmentation process based on the GeodesicActiveContourImageFilter.
From left to right: segmentation of the left ventricle, segmentation of the right ventricle, segmentation of the white
matter, attempt of segmentation of the gray matter.
4.3.4 Threshold Level Set Segmentation
The source code for this section can be found in the file
ThresholdSegmentationLevelSetImageFilter.cxx.
The itk::ThresholdSegmentationLevelSetImageFilter is an extension of the threshold
connected-component segmentation to the level set framework. The goal is to define a range of intensity values that classify the tissue type of interest and then base the propagation term on the level
set equation for that intensity range. Using the level set approach, the smoothness of the evolving
surface can be constrained to prevent some of the “leaking” that is common in connected-component
schemes.
The propagation term P from Equation 4.3 is calculated from the FeatureImage input g with
UpperThreshold U and LowerThreshold L according to the following formula.
P(x) =
g(x) − L if g(x) < (U − L)/2 + L
U − g(x) otherwise
(4.4)
Figure 4.25 illustrates the propagation term function. Intensity values in g between L and H yield
positive values in P, while outside intensities yield negative values in P.
The threshold segmentation filter expects two inputs. The first is an initial level set in the form of
an itk::Image. The second input is the feature image g. For many applications, this filter requires
little or no preprocessing of its input. Smoothing the input image is not usually required to produce
reasonable solutions, though it may still be warranted in some cases.
Figure 4.24 shows how the image processing pipeline is constructed. The initial surface is
generated using the fast marching filter. The output of the segmentation filter is passed to a
itk::BinaryThresholdImageFilter to create a binary representation of the segmented object.
Let’s start by including the appropriate header file.
4.3. Level Set Segmentation
409
Seeds
Fast
Marching
Input
itk::Image
Input
LevelSet
Distance
Curvature
Weight
Threshold
Level−set
Segmentation
Feature
Weight
Binary
Threshold
Output
LevelSet
Binary
Image
Figure 4.24: Collaboration diagram for the ThresholdSegmentationLevelSetImageFilter applied to a segmentation task.
Model
Contracts
Model
Expands
Model
Contracts
P
g(x)
P=0
L
U
Figure 4.25: Propagation term for threshold-based level set segmentation. From Equation 4.4.
410
Chapter 4. Segmentation
#include "itkThresholdSegmentationLevelSetImageFilter.h"
We define the image type using a particular pixel type and dimension. In this case we will use 2D
float images.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The following lines instantiate a ThresholdSegmentationLevelSetImageFilter using the New()
method.
using ThresholdSegmentationLevelSetImageFilterType =
itk::ThresholdSegmentationLevelSetImageFilter< InternalImageType,
InternalImageType >;
ThresholdSegmentationLevelSetImageFilterType::Pointer thresholdSegmentation =
ThresholdSegmentationLevelSetImageFilterType::New();
For the ThresholdSegmentationLevelSetImageFilter, scaling parameters are used to balance the influence of the propagation (inflation) and the curvature (surface smoothing) terms from Equation 4.3.
The advection term is not used in this filter. Set the terms with methods SetPropagationScaling()
and SetCurvatureScaling(). Both terms are set to 1.0 in this example.
thresholdSegmentation->SetPropagationScaling( 1.0 );
if ( argc > 8 )
{
thresholdSegmentation->SetCurvatureScaling( std::stod(argv[8]) );
}
else
{
thresholdSegmentation->SetCurvatureScaling( 1.0 );
}
The convergence criteria MaximumRMSError and MaximumIterations are set as in previous examples. We now set the upper and lower threshold values U and L, and the isosurface value to use in
the initial model.
thresholdSegmentation->SetUpperThreshold( ::std::stod(argv[7]) );
thresholdSegmentation->SetLowerThreshold( ::std::stod(argv[6]) );
thresholdSegmentation->SetIsoSurfaceValue(0.0);
The filters are now connected in a pipeline indicated in Figure 4.24. Remember that before calling
Update() on the file writer object, the fast marching filter must be initialized with the seed points
and the output from the reader object. See previous examples and the source code for this section
for details.
4.3. Level Set Segmentation
Structure
White matter
Ventricle
Gray matter
411
Seed Index
(60, 116)
(81, 112)
(107, 69)
Lower
150
210
180
Upper
180
250
210
Output Image
Second from left
Third from left
Fourth from left
Table 4.6: Segmentation results using the ThresholdSegmentationLevelSetImageFilter for various seed points.
The resulting images are shown in Figure 4.26 .
thresholdSegmentation->SetInput( fastMarching->GetOutput() );
thresholdSegmentation->SetFeatureImage( reader->GetOutput() );
thresholder->SetInput( thresholdSegmentation->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
Invoking the Update() method on the writer triggers the execution of the pipeline. As usual, the
call is placed in a try/catch block should any errors occur or exceptions be thrown.
try
{
reader->Update();
const InternalImageType * inputImage = reader->GetOutput();
fastMarching->SetOutputRegion( inputImage->GetBufferedRegion() );
fastMarching->SetOutputSpacing( inputImage->GetSpacing() );
fastMarching->SetOutputOrigin( inputImage->GetOrigin() );
fastMarching->SetOutputDirection( inputImage->GetDirection() );
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Let’s run this application with the same data and parameters as the example given for
itk::ConnectedThresholdImageFilter in Section 4.1.1. We will use a value of 5 as the initial
distance of the surface from the seed points. The algorithm is relatively insensitive to this initialization. Compare the results in Figure 4.26 with those in Figure 4.1. Notice how the smoothness
constraint on the surface prevents leakage of the segmentation into both ventricles, but also localizes
the segmentation to a smaller portion of the gray matter.
4.3.5 Canny-Edge Level Set Segmentation
The source code for this section can be found in the file
CannySegmentationLevelSetImageFilter.cxx.
412
Chapter 4. Segmentation
Figure 4.26: Images generated by the segmentation process based on the ThresholdSegmentationLevelSetImageFilter. From left to right: segmentation of the left ventricle, segmentation of the right ventricle, segmentation
of the white matter, attempt of segmentation of the gray matter. The parameters used in this segmentations are
presented in Table 4.6.
The itk::CannySegmentationLevelSetImageFilter defines a speed term that minimizes distance to the Canny edges in an image. The initial level set model moves through a gradient advection
field until it locks onto those edges. This filter is more suitable for refining existing segmentations
than as a region-growing algorithm.
The two terms defined for the CannySegmentationLevelSetImageFilter are the advection term and
the propagation term from Equation 4.3. The advection term is constructed by minimizing the
squared distance transform from the Canny edges.
min
Z
D2 ⇒ D∇D
(4.5)
where the distance transform D is calculated using a itk::DanielssonDistanceMapImageFilter
applied to the output of the itk::CannyEdgeDetectionImageFilter.
For cases in which some surface expansion is to be allowed, a non-zero value may be set for the
propagation term. The propagation term is simply D. As with all ITK level set segmentation filters,
the curvature term controls the smoothness of the surface.
CannySegmentationLevelSetImageFilter expects two inputs. The first is an initial level set in the
form of an itk::Image. The second input is the feature image g from which propagation and
advection terms are calculated. It is generally a good idea to do some preprocessing of the feature
image to remove noise.
Figure 4.27 shows how the image processing pipeline is constructed. We read two images: the image
to segment and the image that contains the initial implicit surface. The goal is to refine the initial
model from the second input and not to grow a new segmentation from seed points. The feature
image is preprocessed with a few iterations of an anisotropic diffusion filter.
Let’s start by including the appropriate header file.
4.3. Level Set Segmentation
Input
itk::Image
413
Initial
Model
itk::Image
Gradient
Anisotropic
Diffusion
Canny
Threshold
Canny
Variance
Advection
Weight
Threshold
Level−set
Segmentation
Output
LevelSet
Binary
Threshold
Binary
Image
Maximum
Iterations
Figure 4.27: Collaboration diagram for the CannySegmentationLevelSetImageFilter applied to a segmentation
task.
#include "itkCannySegmentationLevelSetImageFilter.h"
#include "itkGradientAnisotropicDiffusionImageFilter.h"
We define the image type using a particular pixel type and dimension. In this case we will use 2D
float images.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The input image will be processed with a few iterations of feature-preserving diffusion. We create a
filter and set the appropriate parameters.
using DiffusionFilterType =
itk::GradientAnisotropicDiffusionImageFilter< InternalImageType,
InternalImageType>;
DiffusionFilterType::Pointer diffusion = DiffusionFilterType::New();
diffusion->SetNumberOfIterations(5);
diffusion->SetTimeStep(0.125);
diffusion->SetConductanceParameter(1.0);
The following lines define and instantiate a CannySegmentationLevelSetImageFilter.
414
Chapter 4. Segmentation
using CannySegmentationLevelSetImageFilterType =
itk::CannySegmentationLevelSetImageFilter< InternalImageType,
InternalImageType >;
CannySegmentationLevelSetImageFilterType::Pointer cannySegmentation =
CannySegmentationLevelSetImageFilterType::New();
As with the other ITK level set segmentation filters, the terms of the CannySegmentationLevelSetImageFilter level set equation can be weighted by scalars. For this application we will modify the
relative weight of the advection term. The propagation and curvature term weights are set to their
defaults of 0 and 1, respectively.
cannySegmentation->SetAdvectionScaling( ::std::stod(argv[6]) );
cannySegmentation->SetCurvatureScaling( 1.0 );
cannySegmentation->SetPropagationScaling( 0.0 );
The maximum number of iterations is specified from the command line. It may not be desirable in
some applications to run the filter to convergence. Only a few iterations may be required.
cannySegmentation->SetMaximumRMSError( 0.01 );
cannySegmentation->SetNumberOfIterations( ::std::stoi(argv[8]) );
There are two important parameters in the CannySegmentationLevelSetImageFilter to control the
behavior of the Canny edge detection. The variance parameter controls the amount of Gaussian
smoothing on the input image. The threshold parameter indicates the lowest allowed value in the
output image. Thresholding is used to suppress Canny edges whose gradient magnitudes fall below
a certain value.
cannySegmentation->SetThreshold( ::std::stod(argv[4]) );
cannySegmentation->SetVariance( ::std::stod(argv[5]) );
Finally, it is very important to specify the isovalue of the surface in the initial model input image. In a
binary image, for example, the isosurface is found midway between the foreground and background
values.
cannySegmentation->SetIsoSurfaceValue( ::std::stod(argv[7]) );
The filters are now connected in a pipeline indicated in Figure 4.27.
diffusion->SetInput( reader1->GetOutput() );
cannySegmentation->SetInput( reader2->GetOutput() );
cannySegmentation->SetFeatureImage( diffusion->GetOutput() );
thresholder->SetInput( cannySegmentation->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
4.3. Level Set Segmentation
415
Figure 4.28: Results of applying the CannySegmentationLevelSetImageFilter to a prior ventricle segmentation.
Shown from left to right are the original image, the prior segmentation of the ventricle from Figure 4.26, 15
iterations of the CannySegmentationLevelSetImageFilter, and the CannySegmentationLevelSetImageFilter run
to convergence.
Invoking the Update() method on the writer triggers the execution of the pipeline. As usual, the
call is placed in a try/catch block to handle any exceptions that may be thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
We can use this filter to make some subtle refinements to the ventricle segmentation from
the previous example that used the itk::ThresholdSegmentationLevelSetImageFilter.
The application was run using Examples/Data/BrainProtonDensitySlice.png and
Examples/Data/VentricleModel.png as inputs, a threshold of 7.0, variance of 0.1,
advection weight of 10.0, and an initial isosurface value of 127.5. One case was run for 15
iterations and the second was run to convergence. Compare the results in the two rightmost images
of Figure 4.28 with the ventricle segmentation from Figure 4.26 shown in the middle. Jagged edges
are straightened and the small spur at the upper right-hand side of the mask has been removed.
The free parameters of this filter can be adjusted to achieve a wide range of shape variations from the
original model. Finding the right parameters for your particular application is usually a process of
trial and error. As with most ITK level set segmentation filters, examining the propagation (speed)
and advection images can help the process of tuning parameters. These images are available using
Set/Get methods from the filter after it has been updated.
In some cases it is interesting to take a direct look at the speed image used internally by this filter.
416
Chapter 4. Segmentation
This may help for setting the correct parameters for driving the segmentation. In order to obtain such
speed image, the method GenerateSpeedImage() should be invoked first. Then we can recover the
speed image with the GetSpeedImage() method as illustrated in the following lines.
cannySegmentation->GenerateSpeedImage();
using SpeedImageType =
CannySegmentationLevelSetImageFilterType::SpeedImageType;
using SpeedWriterType = itk::ImageFileWriter;
SpeedWriterType::Pointer speedWriter = SpeedWriterType::New();
speedWriter->SetInput( cannySegmentation->GetSpeedImage() );
4.3.6 Laplacian Level Set Segmentation
The source code for this section can be found in the file
LaplacianSegmentationLevelSetImageFilter.cxx.
The itk::LaplacianSegmentationLevelSetImageFilter defines a speed term based on second
derivative features in the image. The speed term is calculated as the Laplacian of the image values.
The goal is to attract the evolving level set surface to local zero-crossings in the Laplacian image.
Like itk::CannySegmentationLevelSetImageFilter, this filter is more suitable for refining
existing segmentations than as a stand-alone, region growing algorithm. It is possible to perform
region growing segmentation, but be aware that the growing surface may tend to become “stuck” at
local edges.
The propagation (speed) term for the LaplacianSegmentationLevelSetImageFilter is constructed by
applying the itk::LaplacianImageFilter to the input feature image. One nice property of using
the Laplacian is that there are no free parameters in the calculation.
LaplacianSegmentationLevelSetImageFilter expects two inputs. The first is an initial level set in the
form of an itk::Image. The second input is the feature image g from which the propagation term
is calculated (see Equation 4.3). Because the filter performs a second derivative calculation, it is
generally a good idea to do some preprocessing of the feature image to remove noise.
Figure 4.29 shows how the image processing pipeline is constructed. We read two images: the image
to segment and the image that contains the initial implicit surface. The goal is to refine the initial
model from the second input to better match the structure represented by the initial implicit surface
(a prior segmentation). The feature image is preprocessed using an anisotropic diffusion filter.
Let’s start by including the appropriate header files.
#include "itkLaplacianSegmentationLevelSetImageFilter.h"
#include "itkGradientAnisotropicDiffusionImageFilter.h"
4.3. Level Set Segmentation
Input
itk::Image
Propagation
Weight
Maximum
Iterations
Laplacian
Level−set
Segmentation
417
Initial
Model
itk::Image
Gradient
Anisotropic
Diffusion
Output
LevelSet
Binary
Threshold
Binary
Image
Figure 4.29: An image processing pipeline using LaplacianSegmentationLevelSetImageFilter for segmentation.
We define the image type using a particular pixel type and dimension. In this case we will use 2D
float images.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The input image will be processed with a few iterations of feature-preserving diffusion. We create
a filter and set the parameters. The number of iterations and the conductance parameter are taken
from the command line.
using DiffusionFilterType =
itk::GradientAnisotropicDiffusionImageFilter< InternalImageType,
InternalImageType >;
DiffusionFilterType::Pointer diffusion = DiffusionFilterType::New();
diffusion->SetNumberOfIterations( std::stoi(argv[4]) );
diffusion->SetTimeStep(0.125);
diffusion->SetConductanceParameter( std::stod(argv[5]) );
The following lines define and instantiate a LaplacianSegmentationLevelSetImageFilter.
using LaplacianSegmentationLevelSetImageFilterType =
itk::LaplacianSegmentationLevelSetImageFilter< InternalImageType,
InternalImageType >;
LaplacianSegmentationLevelSetImageFilterType::Pointer laplacianSegmentation
= LaplacianSegmentationLevelSetImageFilterType::New();
As with the other ITK level set segmentation filters, the terms of the LaplacianSegmentationLevelSetImageFilter level set equation can be weighted by scalars. For this application we will modify the
418
Chapter 4. Segmentation
relative weight of the propagation term. The curvature term weight is set to its default of 1. The
advection term is not used in this filter.
laplacianSegmentation->SetCurvatureScaling( 1.0 );
laplacianSegmentation->SetPropagationScaling( ::std::stod(argv[6]) );
The maximum number of iterations is set from the command line. It may not be desirable in some
applications to run the filter to convergence. Only a few iterations may be required.
laplacianSegmentation->SetMaximumRMSError( 0.002 );
laplacianSegmentation->SetNumberOfIterations( ::std::stoi(argv[8]) );
Finally, it is very important to specify the isovalue of the surface in the initial model input image. In a
binary image, for example, the isosurface is found midway between the foreground and background
values.
laplacianSegmentation->SetIsoSurfaceValue( ::std::stod(argv[7]) );
The filters are now connected in a pipeline indicated in Figure 4.29.
diffusion->SetInput( reader1->GetOutput() );
laplacianSegmentation->SetInput( reader2->GetOutput() );
laplacianSegmentation->SetFeatureImage( diffusion->GetOutput() );
thresholder->SetInput( laplacianSegmentation->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
Invoking the Update() method on the writer triggers the execution of the pipeline. As usual, the
call is placed in a try/catch block to handle any exceptions that may be thrown.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
We can use this filter to make some subtle refinements to the ventricle segmentation
from the example using the filter itk::ThresholdSegmentationLevelSetImageFilter.
This application was run using Examples/Data/BrainProtonDensitySlice.png and
Examples/Data/VentricleModel.png as inputs. We used 10 iterations of the diffusion filter with a conductance of 2.0. The propagation scaling was set to 1.0 and the filter was run
4.3. Level Set Segmentation
419
Figure 4.30: Results of applying LaplacianSegmentationLevelSetImageFilter to a prior ventricle segmentation.
Shown from left to right are the original image, the prior segmentation of the ventricle from Figure 4.26, and the
refinement of the prior using LaplacianSegmentationLevelSetImageFilter.
until convergence. Compare the results in the rightmost images of Figure 4.30 with the ventricle
segmentation from Figure 4.26 shown in the middle. Jagged edges are straightened and the small
spur at the upper right-hand side of the mask has been removed.
4.3.7 Geodesic Active Contours Segmentation With Shape Guidance
The source code for this section can be found in the file
GeodesicActiveContourShapePriorLevelSetImageFilter.cxx.
In medical imaging applications, the general shape, location and orientation of an anatomical structure of interest is typically known a priori. This information can be used to aid the segmentation
process especially when image contrast is low or when the object boundary is not distinct.
extended the geodesic active contours method
In [33], Leventon et al.
with
an
additional
shape-influenced
term
in
the
driving
PDE.
The
itk::GeodesicActiveContourShapePriorLevelSetFilter is a generalization of Leventon’s approach and its use is illustrated in the following example.
To support shape-guidance, the generic level set equation (Eqn( 4.3)) is extended to incorporate a
shape guidance term:
ξ (ψ∗ (x) − ψ(x))
(4.6)
where ψ∗ is the signed distance function of the “best-fit” shape with respect to a shape model. The
new term has the effect of driving the contour towards the best-fit shape. The scalar ξ weights the
420
Chapter 4. Segmentation
influence of the shape term in the overall evolution. In general, the best-fit shape is not known ahead
of time and has to be iteratively estimated in conjunction with the contour evolution.
As with the itk::GeodesicActiveContourLevelSetImageFilter, the GeodesicActiveContourShapePriorLevelSetImageFilter expects two input images: the first is an initial level set and the second a feature image that represents the image edge potential. The configuration of this example is
quite similar to the example in Section 4.3.3 and hence the description will focus on the new objects
involved in the segmentation process as shown in Figure 4.31.
The process pipeline begins with centering the input image using the the
itk::ChangeInformationImageFilter to simplify the estimation of the pose of the shape,
to be explained later. The centered image is then smoothed using non-linear diffusion to remove
noise and the gradient magnitude is computed from the smoothed image. For simplicity, this
example uses the itk::BoundedReciprocalImageFilter to produce the edge potential image.
The itk::FastMarchingImageFilter creates an initial level set using three user specified
seed positions and a initial contour radius. Three seeds are used in this example to facilitate
the segmentation of long narrow objects in a smaller number of iterations. The output of the
FastMarchingImageFilter is passed as the input to the GeodesicActiveContourShapePriorLevelSetImageFilter. At then end of the segmentation process, the output level set is passed to the
itk::BinaryThresholdImageFilter to produce a binary mask representing the segmented object.
The remaining objects in Figure 4.31 are used for shape modeling and estimation. The
itk::PCAShapeSignedDistanceFunction represents a statistical shape model defined by a mean
signed distance and the first K principal components modes; while the itk::Euler2DTransform is
used to represent the pose of the shape. In this implementation, the best-fit shape estimation problem
is reformulated as a minimization problem where the itk::ShapePriorMAPCostFunction is the
cost function to be optimized using the itk::OnePlusOneEvolutionaryOptimizer.
It should be noted that, although particular shape model, transform cost function, and optimizer
are used in this example, the implementation is generic, allowing different instances of these components to be plugged in. This flexibility allows a user to tailor the behavior of the segmentation
process to suit the circumstances of the targeted application.
Let’s start the example by including the headers of the new filters involved in the segmentation.
#include "itkGeodesicActiveContourShapePriorLevelSetImageFilter.h"
#include "itkChangeInformationImageFilter.h"
#include "itkBoundedReciprocalImageFilter.h"
Next, we include the headers of the objects involved in shape modeling and estimation.
#include
#include
#include
#include
#include
"itkPCAShapeSignedDistanceFunction.h"
"itkEuler2DTransform.h"
"itkOnePlusOneEvolutionaryOptimizer.h"
"itkNormalVariateGenerator.h"
"itkNumericSeriesFileNames.h"
4.3. Level Set Segmentation
421
Sigma
Input
Image
Change
Information
(Center Image)
Anisotropic
Diffusion
Gradient
Magnitude
Bounded
Reciprocal
Seeds/
Distance
Fast
Marching
Initial
LevelSet
Prop./
Shape
Scaling
Geodesic
ActiveContour
ShapePrior
Binary
Threshold
Edge
Image
Output
Binary
Image
ShapePriorMAP
CostFunction
NormalVariate
Generator
OnePlusOne
Evolutionary
Optimizer
Euler2DTransform
PCAShape
SignedDistance
Mean
Shape
Image
Shape
Mode
Images
Figure 4.31: Collaboration diagram for the GeodesicActiveContourShapePriorLevelSetImageFilter applied to
a segmentation task.
422
Chapter 4. Segmentation
Given the numerous parameters involved in tuning this segmentation method it is not uncommon
for a segmentation process to run for several minutes and still produce an unsatisfactory result. For
debugging purposes it is quite helpful to track the evolution of the segmentation as it progresses.
The following defines a custom itk::Command class for monitoring the RMS change and shape
parameters at each iteration.
#include "itkCommand.h"
template
class CommandIterationUpdate : public itk::Command
{
public:
using Self = CommandIterationUpdate;
using Superclass = itk::Command;
using Pointer = itk::SmartPointer;
itkNewMacro( Self );
protected:
CommandIterationUpdate() = default;
public:
void Execute(itk::Object *caller,
const itk::EventObject & event) override
{
Execute( (const itk::Object *) caller, event);
}
void Execute(const itk::Object * object,
const itk::EventObject & event) override
{
const auto * filter = static_cast< const TFilter * >( object );
if( typeid( event ) != typeid( itk::IterationEvent ) )
{ return; }
std::cout << filter->GetElapsedIterations() << ": ";
std::cout << filter->GetRMSChange() << " ";
std::cout << filter->GetCurrentParameters() << std::endl;
}
};
We define the image type using a particular pixel type and dimension. In this case we will use 2D
float images.
using InternalPixelType = float;
constexpr unsigned int Dimension = 2;
using InternalImageType = itk::Image< InternalPixelType, Dimension >;
The following line instantiate a itk::GeodesicActiveContourShapePriorLevelSetImageFilter
using the New() method.
4.3. Level Set Segmentation
423
using GeodesicActiveContourFilterType =
itk::GeodesicActiveContourShapePriorLevelSetImageFilter<
InternalImageType, InternalImageType >;
GeodesicActiveContourFilterType::Pointer geodesicActiveContour =
GeodesicActiveContourFilterType::New();
The itk::ChangeInformationImageFilter is the first filter in the preprocessing stage and is
used to force the image origin to the center of the image.
using CenterFilterType =
itk::ChangeInformationImageFilter;
CenterFilterType::Pointer center = CenterFilterType::New();
center->CenterImageOn();
In this example, we will use the bounded reciprocal 1/(1 + x) of the image gradient magnitude as
the edge potential feature image.
using ReciprocalFilterType = itk::BoundedReciprocalImageFilter<
InternalImageType,
InternalImageType >;
ReciprocalFilterType::Pointer reciprocal = ReciprocalFilterType::New();
In the GeodesicActiveContourShapePriorLevelSetImageFilter, scaling parameters are used to
trade off between the propagation (inflation), the curvature (smoothing), the advection, and the
shape influence terms. These parameters are set using methods SetPropagationScaling(),
SetCurvatureScaling(), SetAdvectionScaling() and SetShapePriorScaling(). In this example, we will set the curvature and advection scales to one and let the propagation and shape prior
scale be command-line arguments.
geodesicActiveContour->SetPropagationScaling( propagationScaling );
geodesicActiveContour->SetShapePriorScaling( shapePriorScaling );
geodesicActiveContour->SetCurvatureScaling( 1.0 );
geodesicActiveContour->SetAdvectionScaling( 1.0 );
Each iteration, the current “best-fit” shape is estimated from the edge potential image and the current
contour. To increase speed, only information within the sparse field layers of the current contour is
used in the estimation. The default number of sparse field layers is the same as the ImageDimension which does not contain enough information to get a reliable best-fit shape estimate. Thus, we
override the default and set the number of layers to 4.
geodesicActiveContour->SetNumberOfLayers( 4 );
The filters are then connected in a pipeline as illustrated in Figure 4.31.
424
Chapter 4. Segmentation
center->SetInput( reader->GetOutput() );
smoothing->SetInput( center->GetOutput() );
gradientMagnitude->SetInput( smoothing->GetOutput() );
reciprocal->SetInput( gradientMagnitude->GetOutput() );
geodesicActiveContour->SetInput( fastMarching->GetOutput() );
geodesicActiveContour->SetFeatureImage( reciprocal->GetOutput() );
thresholder->SetInput( geodesicActiveContour->GetOutput() );
writer->SetInput( thresholder->GetOutput() );
Next, we define the shape model. In this example, we use an implicit shape model based on the
principal components such that:
ψ∗ (x) = µ(x) + ∑ αk uk (x)
(4.7)
k
where µ(x) is the mean signed distance computed from training set of segmented objects and uk (x)
are the first K principal components of the offset (signed distance - mean). The coefficients {αk }
form the set of shape parameters.
Given a set of training data, the itk::ImagePCAShapeModelEstimator can be used to obtain the
mean and principal mode shape images required by PCAShapeSignedDistanceFunction.
using ShapeFunctionType = itk::PCAShapeSignedDistanceFunction<
double,
Dimension,
InternalImageType >;
ShapeFunctionType::Pointer shape = ShapeFunctionType::New();
shape->SetNumberOfPrincipalComponents( numberOfPCAModes );
In this example, we will read the mean shape and principal mode images from file. We will assume
that the filenames of the mode images form a numeric series starting from index 0.
ReaderType::Pointer meanShapeReader = ReaderType::New();
meanShapeReader->SetFileName( argv[13] );
meanShapeReader->Update();
std::vector shapeModeImages( numberOfPCAModes );
itk::NumericSeriesFileNames::Pointer fileNamesCreator =
itk::NumericSeriesFileNames::New();
fileNamesCreator->SetStartIndex( 0 );
fileNamesCreator->SetEndIndex( numberOfPCAModes - 1 );
fileNamesCreator->SetSeriesFormat( argv[15] );
4.3. Level Set Segmentation
425
const std::vector & shapeModeFileNames =
fileNamesCreator->GetFileNames();
for (unsigned int k = 0; k < numberOfPCAModes; ++k )
{
ReaderType::Pointer shapeModeReader = ReaderType::New();
shapeModeReader->SetFileName( shapeModeFileNames[k].c_str() );
shapeModeReader->Update();
shapeModeImages[k] = shapeModeReader->GetOutput();
}
shape->SetMeanImage( meanShapeReader->GetOutput() );
shape->SetPrincipalComponentImages( shapeModeImages );
Further we assume that the shape modes have been normalized by multiplying with the corresponding singular value. Hence, we can set the principal component standard deviations to all ones.
ShapeFunctionType::ParametersType pcaStandardDeviations( numberOfPCAModes );
pcaStandardDeviations.Fill( 1.0 );
shape->SetPrincipalComponentStandardDeviations( pcaStandardDeviations );
Next, we instantiate a itk::Euler2DTransform and connect it to the PCASignedDistanceFunction. The transform represent the pose of the shape. The parameters of the transform forms the set
of pose parameters.
using TransformType = itk::Euler2DTransform;
TransformType::Pointer transform = TransformType::New();
shape->SetTransform( transform );
Before updating the level set at each iteration, the parameters of the current best-fit shape is estimated
by minimizing the itk::ShapePriorMAPCostFunction. The cost function is composed of four
terms: contour fit, image fit, shape prior and pose prior. The user can specify the weights applied to
each term.
using CostFunctionType = itk::ShapePriorMAPCostFunction<
InternalImageType,
InternalPixelType >;
CostFunctionType::Pointer costFunction = CostFunctionType::New();
CostFunctionType::WeightsType weights;
weights[0] = 1.0; // weight for contour fit term
weights[1] = 20.0; // weight for image fit term
weights[2] = 1.0; // weight for shape prior term
weights[3] = 1.0; // weight for pose prior term
426
Chapter 4. Segmentation
costFunction->SetWeights( weights );
Contour fit measures the likelihood of seeing the current evolving contour for a given set of shape/pose parameters. This is computed by counting the number of pixels inside the current contour but
outside the current shape.
Image fit measures the likelihood of seeing certain image features for a given set of shape/pose
parameters. This is computed by assuming that ( 1 - edge potential ) approximates a zero-mean,
unit variance Gaussian along the normal of the evolving contour. Image fit is then computed by
computing the Laplacian goodness of fit of the Gaussian:
∑ (G(ψ(x)) − |1 − g(x)|)2
(4.8)
where G is a zero-mean, unit variance Gaussian and g is the edge potential feature image.
The pose parameters are assumed to have a uniform distribution and hence do not contribute to the
cost function. The shape parameters are assumed to have a Gaussian distribution. The parameters
of the distribution are user-specified. Since we assumed the principal modes have already been
normalized, we set the distribution to zero mean and unit variance.
CostFunctionType::ArrayType mean(
shape->GetNumberOfShapeParameters() );
CostFunctionType::ArrayType stddev( shape->GetNumberOfShapeParameters() );
mean.Fill( 0.0 );
stddev.Fill( 1.0 );
costFunction->SetShapeParameterMeans( mean );
costFunction->SetShapeParameterStandardDeviations( stddev );
In this example, we will use the itk::OnePlusOneEvolutionaryOptimizer to optimize the cost
function.
using OptimizerType = itk::OnePlusOneEvolutionaryOptimizer;
OptimizerType::Pointer optimizer = OptimizerType::New();
The evolutionary optimization algorithm is based on testing random permutations of the parameters.
As such, we need to provide the optimizer with a random number generator. In the following lines,
we create a itk::NormalVariateGenerator, seed it, and connect it to the optimizer.
using GeneratorType = itk::Statistics::NormalVariateGenerator;
GeneratorType::Pointer generator = GeneratorType::New();
generator->Initialize( 20020702 );
optimizer->SetNormalVariateGenerator( generator );
4.3. Level Set Segmentation
427
The cost function has K + 3 parameters. The first K parameters are the principal component multipliers, followed by the 2D rotation parameter (in radians) and the x- and y- translation parameters (in
mm). We need to carefully scale the different types of parameters to compensate for the differences
in the dynamic ranges of the parameters.
OptimizerType::ScalesType scales( shape->GetNumberOfParameters() );
scales.Fill( 1.0 );
for( unsigned int k = 0; k < numberOfPCAModes; k++ )
{
scales[k] = 20.0; // scales for the pca mode multiplier
}
scales[numberOfPCAModes] = 350.0; // scale for 2D rotation
optimizer->SetScales( scales );
Next, we specify the initial radius, the shrink and grow mutation factors and termination criteria of
the optimizer. Since the best-fit shape is re-estimated each iteration of the curve evolution, we do
not need to spend too much time finding the true minimizing solution each time; we only need to
head towards it. As such, we only require a small number of optimizer iterations.
double initRadius = 1.05;
double grow = 1.1;
double shrink = pow(grow, -0.25);
optimizer->Initialize(initRadius, grow, shrink);
optimizer->SetEpsilon(1.0e-6); // minimal search radius
optimizer->SetMaximumIteration(15);
Before starting the segmentation process we need to also supply the initial best-fit shape estimate. In
this example, we start with the unrotated mean shape with the initial x- and y- translation specified
through command-line arguments.
ShapeFunctionType::ParametersType parameters(
shape->GetNumberOfParameters() );
parameters.Fill( 0.0 );
parameters[numberOfPCAModes + 1] = std::stod( argv[16] ); // startX
parameters[numberOfPCAModes + 2] = std::stod( argv[17] ); // startY
Finally, we connect all the components to the filter and add our observer.
geodesicActiveContour->SetShapeFunction( shape );
geodesicActiveContour->SetCostFunction( costFunction );
geodesicActiveContour->SetOptimizer( optimizer );
geodesicActiveContour->SetInitialParameters( parameters );
using CommandType = CommandIterationUpdate;
CommandType::Pointer observer = CommandType::New();
geodesicActiveContour->AddObserver( itk::IterationEvent(), observer );
428
Chapter 4. Segmentation
Figure 4.32: The input image to the GeodesicActiveContourShapePriorLevelSetImageFilter is a synthesized
MR-T1 mid-sagittal slice (217 × 180 pixels, 1 × 1 mm spacing) of the brain (left) and the initial best-fit shape
(right) chosen to roughly overlap the corpus callosum in the image to be segmented.
The invocation of the Update() method on the writer triggers the execution of the pipeline. As
usual, the call is placed in a try/catch block to handle exceptions should errors occur.
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Deviating from previous examples,
we will demonstrate this example using
BrainMidSagittalSlice.png (Figure 4.32, left) from the Examples/Data directory.
The aim here is to segment the corpus callosum from the image using a shape model
defined by CorpusCallosumMeanShape.mha and the first three principal components
CorpusCallosumMode0.mha, CorpusCallosumMode1.mha and CorpusCallosumMode12.mha.
As shown in Figure 4.33, the first mode captures scaling, the second mode captures the shifting of
mass between the rostrum and the splenium and the third mode captures the degree of curvature.
Segmentation results with and without shape guidance are shown in Figure 4.34.
A sigma value of 1.0 was used to compute the image gradient and the propagation and shape prior
scaling are respectively set to 0.5 and 0.02. An initial level set was created by placing one seed
point in the rostrum (60, 102), one in the splenium (120, 85) and one centrally in the body (88, 83)
of the corpus callosum with an initial radius of 6 pixels at each seed position. The best-fit shape was
initially placed with a translation of (10, 0)mm so that it roughly overlapped the corpus callosum in
the image as shown in Figure 4.32 (right).
From Figure 4.34 it can be observed that without shape guidance (left), segmentation using geodesic
active contour leaks in the regions where the corpus callosum blends into the surrounding brain
4.4. Feature Extraction
429
−3σ
mean
+3σ
mode 0:
mode 1:
mode 2:
Figure 4.33: First three PCA modes of a low-resolution (58 × 31 pixels, 2 × 2 mm spacing) corpus callosum
model used in the shape guided geodesic active contours example.
Figure 4.34: Corpus callosum segmentation using geodesic active contours without (left) and with (center)
shape guidance. The image on the right represents the best-fit shape at the end of the segmentation process.
tissues. With shape guidance (center), the segmentation is constrained by the global shape model to
prevent leaking.
The final best-fit shape parameters after the segmentation process is:
Parameters: [-0.384988, -0.578738, 0.557793, 0.275202, 16.9992, 4.73473]
and is shown in Figure 4.34 (right). Note that a 0.28 radian (15.8 degree) rotation has been introduced to match the model to the corpus callosum in the image. Additionally, a negative weight for
the first mode shrinks the size relative to the mean shape. A negative weight for the second mode
shifts the mass to splenium, and a positive weight for the third mode increases the curvature. It can
also be observed that the final segmentation is a combination of the best-fit shape with additional local deformation. The combination of both global and local shape allows the segmentation to capture
fine details not represented in the shape model.
4.4 Feature Extraction
Extracting salient features from images is an important task on image processing. It is typically used
for guiding segmentation methods, preparing data for registration methods, or as a mechanism for
430
Chapter 4. Segmentation
recognizing anatomical structures in images. The following section introduce some of the feature
extraction methods available in ITK.
4.4.1 Hough Transform
The Hough transform is a widely used technique for detection of geometrical features in images.
It is based on mapping the image into a parametric space in which it may be easier to identify if
particular geometrical features are present in the image. The transformation is specific for each
desired geometrical shape.
Line Extraction
The source code for this section can be found in the file
HoughTransform2DLinesImageFilter.cxx.
This example illustrates the use of the itk::HoughTransform2DLinesImageFilter to find
straight lines in a 2-dimensional image.
First, we include the header files of the filter.
#include "itkHoughTransform2DLinesImageFilter.h"
Next, we declare the pixel type and image dimension and specify the image type to be used as input.
We also specify the image type of the accumulator used in the Hough transform filter.
using PixelType = unsigned char;
using AccumulatorPixelType = float;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
using AccumulatorImageType = itk::Image< AccumulatorPixelType, Dimension >;
We setup a reader to load the input image.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
try
{
reader->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
4.4. Feature Extraction
431
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
ImageType::Pointer localImage = reader->GetOutput();
Once the image is loaded, we apply a itk::GradientMagnitudeImageFilter to segment edges.
This casts the input image using a itk::CastImageFilter.
using CastingFilterType =
itk::CastImageFilter< ImageType, AccumulatorImageType >;
CastingFilterType::Pointer caster = CastingFilterType::New();
std::cout << "Applying gradient magnitude filter" << std::endl;
using GradientFilterType =
itk::GradientMagnitudeImageFilter< AccumulatorImageType,
AccumulatorImageType >;
GradientFilterType::Pointer gradFilter = GradientFilterType::New();
caster->SetInput(localImage);
gradFilter->SetInput(caster->GetOutput());
gradFilter->Update();
The next step is to apply a threshold filter on the gradient magnitude image to keep only bright
values. Only pixels with a high value will be used by the Hough transform filter.
std::cout << "Thresholding" << std::endl;
using ThresholdFilterType = itk::ThresholdImageFilter;
ThresholdFilterType::Pointer threshFilter = ThresholdFilterType::New();
threshFilter->SetInput( gradFilter->GetOutput());
threshFilter->SetOutsideValue(0);
unsigned char threshBelow = 0;
unsigned char threshAbove = 255;
threshFilter->ThresholdOutside(threshBelow,threshAbove);
threshFilter->Update();
We create the HoughTransform2DLinesImageFilter based on the pixel type of the input image (the
resulting image from the ThresholdImageFilter).
std::cout << "Computing Hough Map" << std::endl;
using HoughTransformFilterType =
itk::HoughTransform2DLinesImageFilter< AccumulatorPixelType,
AccumulatorPixelType>;
HoughTransformFilterType::Pointer houghFilter
= HoughTransformFilterType::New();
432
Chapter 4. Segmentation
We set the input to the filter to be the output of the ThresholdImageFilter. We set also the number
of lines we are looking for. Basically, the filter computes the Hough map, blurs it using a certain
variance and finds maxima in the Hough map. After a maximum is found, the local neighborhood,
a circle, is removed from the Hough map. SetDiscRadius() defines the radius of this disc.
The output of the filter is the accumulator.
houghFilter->SetInput(threshFilter->GetOutput());
houghFilter->SetNumberOfLines(std::stoi(argv[3]));
if(argc > 4 )
{
houghFilter->SetVariance(std::stod(argv[4]));
}
if(argc > 5 )
{
houghFilter->SetDiscRadius(std::stod(argv[5]));
}
houghFilter->Update();
AccumulatorImageType::Pointer localAccumulator = houghFilter->GetOutput();
We can also get the lines as itk::LineSpatialObject. The GetLines() function return a list of
those.
HoughTransformFilterType::LinesListType lines;
lines = houghFilter->GetLines();
std::cout << "Found " << lines.size() << " line(s)." << std::endl;
We can then allocate an image to draw the resulting lines as binary objects.
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
OutputImageType::Pointer localOutputImage = OutputImageType::New();
OutputImageType::RegionType region(localImage->GetLargestPossibleRegion());
localOutputImage->SetRegions(region);
localOutputImage->CopyInformation(localImage);
localOutputImage->Allocate(true); // initialize buffer to zero
We iterate through the list of lines and we draw them.
using LineIterator = HoughTransformFilterType::LinesListType::const_iterator;
LineIterator itLines = lines.begin();
while( itLines != lines.end() )
{
4.4. Feature Extraction
433
We get the list of points which consists of two points to represent a straight line. Then, from these
two points, we compute a fixed point u and a unit vector ~v to parameterize the line.
using PointListType = HoughTransformFilterType::LineType::PointListType;
PointListType
PointListType::const_iterator
pointsList = (*itLines)->GetPoints();
itPoints = pointsList.begin();
double u[2];
u[0] = (*itPoints).GetPosition()[0];
u[1] = (*itPoints).GetPosition()[1];
itPoints++;
double v[2];
v[0] = u[0]-(*itPoints).GetPosition()[0];
v[1] = u[1]-(*itPoints).GetPosition()[1];
double norm = std::sqrt(v[0]*v[0]+v[1]*v[1]);
v[0] /= norm;
v[1] /= norm;
We draw a white pixels in the output image to represent the line.
ImageType::IndexType localIndex;
itk::Size<2> size = localOutputImage->GetLargestPossibleRegion().GetSize();
float diag = std::sqrt((float)( size[0]*size[0] + size[1]*size[1] ));
for(auto i=static_cast(-diag); i(diag); i++)
{
localIndex[0]=(long int)(u[0]+i*v[0]);
localIndex[1]=(long int)(u[1]+i*v[1]);
OutputImageType::RegionType outputRegion =
localOutputImage->GetLargestPossibleRegion();
if( outputRegion.IsInside( localIndex ) )
{
localOutputImage->SetPixel( localIndex, 255 );
}
}
itLines++;
}
We setup a writer to write out the binary image created.
using WriterType = itk::ImageFileWriter< OutputImageType >;
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[2] );
writer->SetInput( localOutputImage );
try
{
434
Chapter 4. Segmentation
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
Circle Extraction
The source code for this section can be found in the file
HoughTransform2DCirclesImageFilter.cxx.
This example illustrates the use of the itk::HoughTransform2DCirclesImageFilter to find
circles in a 2-dimensional image.
First, we include the header files of the filter.
#include "itkHoughTransform2DCirclesImageFilter.h"
Next, we declare the pixel type and image dimension and specify the image type to be used as input.
We also specify the image type of the accumulator used in the Hough transform filter.
using PixelType = unsigned char;
using AccumulatorPixelType = unsigned;
using RadiusPixelType = float;
constexpr unsigned int Dimension = 2;
using ImageType = itk::Image< PixelType, Dimension >;
ImageType::IndexType localIndex;
using AccumulatorImageType = itk::Image< AccumulatorPixelType, Dimension >;
We setup a reader to load the input image.
using ReaderType = itk::ImageFileReader< ImageType >;
ReaderType::Pointer reader = ReaderType::New();
reader->SetFileName( argv[1] );
try
{
reader->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
4.4. Feature Extraction
435
}
ImageType::Pointer localImage = reader->GetOutput();
We create the HoughTransform2DCirclesImageFilter based on the pixel type of the input image (the
resulting image from the ThresholdImageFilter).
std::cout << "Computing Hough Map" << std::endl;
using HoughTransformFilterType =
itk::HoughTransform2DCirclesImageFilter;
HoughTransformFilterType::Pointer houghFilter
= HoughTransformFilterType::New();
We set the input of the filter to be the output of the ImageFileReader. We set also the number of
circles we are looking for. Basically, the filter computes the Hough map, blurs it using a certain
variance and finds maxima in the Hough map. After a maximum is found, the local neighborhood,
a circle, is removed from the Hough map. SetDiscRadiusRatio() defines the radius of this disc proportional to the radius of the disc found. The Hough map is computed by looking at the points
above a certain threshold in the input image. Then, for each point, a Gaussian derivative function
is computed to find the direction of the normal at that point. The standard deviation of the derivative function can be adjusted by SetSigmaGradient(). The accumulator is filled by drawing a line
along the normal and the length of this line is defined by the minimum radius (SetMinimumRadius()) and the maximum radius (SetMaximumRadius()). Moreover, a sweep angle can be defined
by SetSweepAngle() (default 0.0) to increase the accuracy of detection.
The output of the filter is the accumulator.
houghFilter->SetInput( reader->GetOutput() );
houghFilter->SetNumberOfCircles( std::stoi(argv[3]) );
houghFilter->SetMinimumRadius( std::stod(argv[4]) );
houghFilter->SetMaximumRadius( std::stod(argv[5]) );
if( argc > 6 )
{
houghFilter->SetSweepAngle( std::stod(argv[6]) );
}
if( argc > 7 )
{
houghFilter->SetSigmaGradient( std::stoi(argv[7]) );
}
if( argc > 8 )
{
houghFilter->SetVariance( std::stod(argv[8]) );
}
if( argc > 9 )
436
Chapter 4. Segmentation
{
houghFilter->SetDiscRadiusRatio( std::stod(argv[9]) );
}
houghFilter->Update();
AccumulatorImageType::Pointer localAccumulator = houghFilter->GetOutput();
We can also get the circles as itk::EllipseSpatialObject. The GetCircles() function return
a list of those.
HoughTransformFilterType::CirclesListType circles;
circles = houghFilter->GetCircles();
std::cout << "Found " << circles.size() << " circle(s)." << std::endl;
We can then allocate an image to draw the resulting circles as binary objects.
using OutputPixelType = unsigned char;
using OutputImageType = itk::Image< OutputPixelType, Dimension >;
OutputImageType::Pointer localOutputImage = OutputImageType::New();
OutputImageType::RegionType region;
region.SetSize(localImage->GetLargestPossibleRegion().GetSize());
region.SetIndex(localImage->GetLargestPossibleRegion().GetIndex());
localOutputImage->SetRegions( region );
localOutputImage->SetOrigin(localImage->GetOrigin());
localOutputImage->SetSpacing(localImage->GetSpacing());
localOutputImage->Allocate(true); // initializes buffer to zero
We iterate through the list of circles and we draw them.
using CirclesListType = HoughTransformFilterType::CirclesListType;
CirclesListType::const_iterator itCircles = circles.begin();
while( itCircles != circles.end() )
{
std::cout << "Center: ";
std::cout << (*itCircles)->GetCenterPoint()
<< std::endl;
std::cout << "Radius: " << (*itCircles)->GetRadius()[0] << std::endl;
We draw white pixels in the output image to represent each circle.
for( double angle = 0;
angle <= itk::Math::twopi;
angle += itk::Math::pi/60.0 )
{
const HoughTransformFilterType::CircleType::PointType centerPoint =
4.4. Feature Extraction
437
(*itCircles)->GetCenterPoint();
using IndexValueType = ImageType::IndexType::IndexValueType;
localIndex[0] =
itk::Math::Round(centerPoint[0]
+ (*itCircles)->GetRadius()[0]*std::cos(angle));
localIndex[1] =
itk::Math::Round(centerPoint[1]
+ (*itCircles)->GetRadius()[0]*std::sin(angle));
OutputImageType::RegionType outputRegion =
localOutputImage->GetLargestPossibleRegion();
if( outputRegion.IsInside( localIndex ) )
{
localOutputImage->SetPixel( localIndex, 255 );
}
}
itCircles++;
}
We setup a writer to write out the binary image created.
using WriterType = itk::ImageFileWriter< ImageType
WriterType::Pointer writer = WriterType::New();
writer->SetFileName( argv[2] );
writer->SetInput(localOutputImage );
try
{
writer->Update();
}
catch( itk::ExceptionObject & excep )
{
std::cerr << "Exception caught !" << std::endl;
std::cerr << excep << std::endl;
return EXIT_FAILURE;
}
>;
CHAPTER
FIVE
STATISTICS
This chapter introduces the statistics functionalities in Insight. The statistics subsystem’s primary
purpose is to provide general capabilities for statistical pattern classification. However, its use is not
limited for classification. Users might want to use data containers and algorithms in the statistics
subsystem to perform other statistical analysis or to preprocess image data for other tasks.
The statistics subsystem mainly consists of three parts: data container classes, statistical algorithms,
and the classification framework. In this chapter, we will discuss each major part in that order.
5.1 Data Containers
An itk::Statistics::Sample object is a data container of elements that we call measurement
vectors. A measurement vector is an array of values (of the same type) measured on an object (In
images, it can be a vector of the gray intensity value and/or the gradient value of a pixel). Strictly
speaking from the design of the Sample class, a measurement vector can be any class derived from
itk::FixedArray, including FixedArray itself.
Sample
ListSampleBase
ListSample
Histogram
Subsample
ImageToListSampleAdaptor PointSetToListSampleAdaptor
ScalarImageToListSampleAdaptor
JointDomainImageToListSampleAdaptor
Figure 5.1: Sample class inheritance diagram.
MembershipSample
440
Chapter 5. Statistics
5.1.1 Sample Interface
The source code for this section can be found in the file
ListSample.cxx.
This example illustrates the common interface of the Sample class in Insight.
Different subclasses of itk::Statistics::Sample expect different sets of template arguments. In
this example, we use the itk::Statistics::ListSample class that requires the type of measurement vectors. The ListSample uses STL vector to store measurement vectors. This class conforms
to the common interface of Sample. Most methods of the Sample class interface are for retrieving
measurement vectors, the size of a container, and the total frequency. In this example, we will see
those information retrieving methods in addition to methods specific to the ListSample class for data
input.
To use the ListSample class, we include the header file for the class.
We need another header for measurement vectors. We are going to use the itk::Vector class
which is a subclass of the itk::FixedArray class.
#include "itkListSample.h"
#include "itkVector.h"
The following code snippet defines the measurement vector type as a three component float
itk::Vector. The MeasurementVectorType is the measurement vector type in the SampleType.
An object is instantiated at the third line.
using MeasurementVectorType = itk::Vector< float, 3 >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
In the above code snippet, the namespace specifier for ListSample is itk::Statistics:: instead
of the usual namespace specifier for other ITK classes, itk::.
The newly instantiated object does not have any data in it. We have two different ways of storing
data elements. The first method is using the PushBack method.
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack(mv);
The previous code increases the size of the container by one and stores mv as the first data element
in it.
5.1. Data Containers
441
The other way to store data elements is calling the Resize method and then calling the
SetMeasurementVector() method with a measurement vector. The following code snippet increases the size of the container to three and stores two measurement vectors at the second and the
third slot. The measurement vector stored using the PushBack method above is still at the first slot.
sample->Resize(3);
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->SetMeasurementVector(1, mv);
mv[0] = 3.0;
mv[1] = 8.0;
mv[2] = 6.0;
sample->SetMeasurementVector(2, mv);
We have seen how to create an ListSample object and store measurement vectors using the
ListSample-specific interface. The following code shows the common interface of the Sample class.
The Size method returns the number of measurement vectors in the sample. The primary data
stored in Sample subclasses are measurement vectors. However, each measurement vector has its
associated frequency of occurrence within the sample. For the ListSample and the adaptor classes
(see Section 5.1.2), the frequency value is always one. itk::Statistics::Histogram can have a
varying frequency (float type) for each measurement vector. We retrieve measurement vectors using the GetMeasurementVector(unsigned long instance identifier), and frequency using
the GetFrequency(unsigned long instance identifier).
for ( unsigned long i = 0; i < sample->Size(); ++i )
{
std::cout << "id = " << i
<< "\t measurement vector = "
<< sample->GetMeasurementVector(i)
<< "\t frequency = "
<< sample->GetFrequency(i)
<< std::endl;
}
The output should look like the following:
id = 0 measurement vector = 1 2 4 frequency = 1
id = 1 measurement vector = 2 4 5 frequency = 1
id = 2 measurement vector = 3 8 6 frequency = 1
We can get the same result with its iterator.
SampleType::Iterator iter = sample->Begin();
442
Chapter 5. Statistics
while( iter != sample->End() )
{
std::cout << "id = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency()
<< std::endl;
++iter;
}
The last method defined in the Sample class is the GetTotalFrequency() method that returns
the sum of frequency values associated with every measurement vector in a container. In the case of
ListSample and the adaptor classes, the return value should be exactly the same as that of the Size()
method, because the frequency values are always one for each measurement vector. However, for the
itk::Statistics::Histogram, the frequency values can vary. Therefore, if we want to develop
a general algorithm to calculate the sample mean, we must use the GetTotalFrequency() method
instead of the Size() method.
std::cout << "Size = " << sample->Size() << std::endl;
std::cout << "Total frequency = "
<< sample->GetTotalFrequency() << std::endl;
5.1.2 Sample Adaptors
There are two adaptor classes that provide the common itk::Statistics::Sample interfaces for itk::Image and itk::PointSet, two fundamental data container classes
found in ITK. The adaptor classes do not store any real data elements themselves. These
data come from the source data container plugged into them.
First, we will describe how to create an itk::Statistics::ImageToListSampleAdaptor and then an
itk::Statistics::PointSetToListSampleAdaptor object.
ImageToListSampleAdaptor
The source code for this section can be found in the file
ImageToListSampleAdaptor.cxx.
This example shows how to instantiate an itk::Statistics::ImageToListSampleAdaptor object and plug-in an itk::Image object as the data source for the adaptor.
In this example, we use the ImageToListSampleAdaptor class that requires the input type of Image as the template argument. To users of the ImageToListSampleAdaptor, the pixels of the input
5.1. Data Containers
443
image are treated as measurement vectors. The ImageToListSampleAdaptor is one of two adaptor
classes among the subclasses of the itk::Statistics::Sample. That means an ImageToListSampleAdaptor object does not store any real data. The data comes from other ITK data container
classes. In this case, an instance of the Image class is the source of the data.
To use an ImageToListSampleAdaptor object, include the header file for the class. Since we are
using an adaptor, we also should include the header file for the Image class. For illustration, we use
the itk::RandomImageSource that generates an image with random pixel values. So, we need to
include the header file for this class. Another convenient filter is the itk::ComposeImageFilter
which creates an image with pixels of array type from one or more input images composed of pixels
of scalar type. Since an element of a Sample object is a measurement vector, you cannot plug in an
image of scalar pixels. However, if we want to use an image of scalar pixels without the help from the
ComposeImageFilter, we can use the itk::Statistics::ScalarImageToListSampleAdaptor
class that is derived from the itk::Statistics::ImageToListSampleAdaptor. The usage of
the ScalarImageToListSampleAdaptor is identical to that of the ImageToListSampleAdaptor.
#include
#include
#include
#include
"itkImageToListSampleAdaptor.h"
"itkImage.h"
"itkRandomImageSource.h"
"itkComposeImageFilter.h"
We assume you already know how to create an image. The following code snippet will create a 2D
image of float pixels filled with random values.
using FloatImage2DType = itk::Image;
itk::RandomImageSource::Pointer random;
random = itk::RandomImageSource::New();
random->SetMin(
0.0 );
random->SetMax( 1000.0 );
using SpacingValueType = FloatImage2DType::SpacingValueType;
using SizeValueType = FloatImage2DType::SizeValueType;
using PointValueType = FloatImage2DType::PointValueType;
SizeValueType size[2] = {20, 20};
random->SetSize( size );
SpacingValueType spacing[2] = {0.7, 2.1};
random->SetSpacing( spacing );
PointValueType origin[2] = {15, 400};
random->SetOrigin( origin );
We now have an instance of Image and need to cast it to an Image object with an array pixel
type (anything derived from the itk::FixedArray class such as itk::Vector, itk::Point,
itk::RGBPixel, or itk::CovariantVector).
444
Chapter 5. Statistics
Since the image pixel type is float in this example, we will use a single element float FixedArray
as our measurement vector type. And that will also be our pixel type for the cast filter.
using MeasurementVectorType = itk::FixedArray< float, 1 >;
using ArrayImageType = itk::Image< MeasurementVectorType, 2 >;
using CasterType =
itk::ComposeImageFilter< FloatImage2DType, ArrayImageType >;
CasterType::Pointer caster = CasterType::New();
caster->SetInput( random->GetOutput() );
caster->Update();
Up to now, we have spent most of our time creating an image suitable for the adaptor. Actually, the
hard part of this example is done. Now, we just define an adaptor with the image type and instantiate
an object.
using SampleType = itk::Statistics::ImageToListSampleAdaptor;
SampleType::Pointer sample = SampleType::New();
The final task is to plug in the image object to the adaptor. After that, we can use the common
methods and iterator interfaces shown in Section 5.1.1.
sample->SetImage( caster->GetOutput() );
If we are interested only in pixel values, the ScalarImageToListSampleAdaptor (scalar pixels) or the ImageToListSampleAdaptor (vector pixels) would be sufficient.
However, if
we want to perform some statistical analysis on spatial information (image index or pixel’s
physical location) and pixel values altogether, we want to have a measurement vector
that consists of a pixel’s value and physical position.
In that case, we can use the
itk::Statistics::JointDomainImageToListSampleAdaptor class. With this class, when
we call the GetMeasurementVector() method, the returned measurement vector is composed
of the physical coordinates and pixel values. The usage is almost the same as with ImageToListSampleAdaptor. One important difference between JointDomainImageToListSampleAdaptor and the other two image adaptors is that the JointDomainImageToListSampleAdaptor has the
SetNormalizationFactors() method. Each component of a measurement vector from the JointDomainImageToListSampleAdaptor is divided by the corresponding component value from the supplied normalization factors.
PointSetToListSampleAdaptor
The source code for this section can be found in the file
PointSetToListSampleAdaptor.cxx.
5.1. Data Containers
445
We will describe how to use itk::PointSet as a itk::Statistics::Sample using an adaptor
in this example.
The itk::Statistics::PointSetToListSampleAdaptor class requires a PointSet as input. The
PointSet class is an associative data container. Each point in a PointSet object can have an associated
optional data value. For the statistics subsystem, the current implementation of PointSetToListSampleAdaptor takes only the point part into consideration. In other words, the measurement vectors
from a PointSetToListSampleAdaptor object are points from the PointSet object that is plugged into
the adaptor object.
To use an PointSetToListSampleAdaptor class, we include the header file for the class.
#include "itkPointSetToListSampleAdaptor.h"
Since we are using an adaptor, we also include the header file for the PointSet class.
#include "itkPointSet.h"
#include "itkVector.h"
Next we create a PointSet object. The following code snippet will create a PointSet object that stores
points (its coordinate value type is float) in 3D space.
using PointSetType = itk::PointSet< short >;
PointSetType::Pointer pointSet = PointSetType::New();
Note that the short type used in the declaration of PointSetType pertains to the pixel
type associated with every point, not to the type used to represent point coordinates. If we
want to change the type of the point in terms of the coordinate value and/or dimension, we
have to modify the TMeshTraits (one of the optional template arguments for the PointSet
class). The easiest way of creating a custom mesh traits instance is to specialize the existing itk::DefaultStaticMeshTraits. By specifying the TCoordRep template argument,
we can change the coordinate value type of a point. By specifying the VPointDimension
template argument, we can change the dimension of the point. As mentioned earlier, a
PointSetToListSampleAdaptor object cares only about the points, and the type of measurement
vectors is the type of points.
To make the example a little bit realistic, we add two points into the pointSet.
PointSetType::PointType point;
point[0] = 1.0;
point[1] = 2.0;
point[2] = 3.0;
pointSet->SetPoint( 0UL, point);
point[0] = 2.0;
446
Chapter 5. Statistics
point[1] = 4.0;
point[2] = 6.0;
pointSet->SetPoint( 1UL, point );
Now we have a PointSet object with two points in it. The PointSet is ready to be plugged into the
adaptor. First, we create an instance of the PointSetToListSampleAdaptor class with the type of the
input PointSet object.
using SampleType =
itk::Statistics::PointSetToListSampleAdaptor;
SampleType::Pointer sample = SampleType::New();
Second, all we have to do is plug in the PointSet object to the adaptor. After that, we can use the
common methods and iterator interfaces shown in Section 5.1.1.
sample->SetPointSet( pointSet );
SampleType::Iterator iter = sample->Begin();
while( iter != sample->End() )
{
std::cout << "id = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency()
<< std::endl;
++iter;
}
The source code for this section can be found in the file
PointSetToAdaptor.cxx.
We will describe how to use itk::PointSet as a Sample using an adaptor in this example.
itk::Statistics::PointSetToListSampleAdaptor class requires the type of input
itk::PointSet object. The itk::PointSet class is an associative data container. Each
point in a PointSet object can have its associated data value (optional). For the statistics subsystem, current implementation of PointSetToListSampleAdaptor takes only the point part into
consideration. In other words, the measurement vectors from a PointSetToListSampleAdaptor
object are points from the PointSet object that is plugged-into the adaptor object.
To use, an itk::PointSetToListSampleAdaptor object, we include the header file for the class.
5.1. Data Containers
447
#include "itkPointSetToListSampleAdaptor.h"
Since, we are using an adaptor, we also include the header file for the itk::PointSet class.
#include "itkPointSet.h"
We assume you already know how to create an itk::PointSet object. The following code snippet
will create a 2D image of float pixels filled with random values.
using FloatPointSet2DType = itk::PointSet;
itk::RandomPointSetSource::Pointer random;
random = itk::RandomPointSetSource::New();
random->SetMin(0.0);
random->SetMax(1000.0);
unsigned long size[2] = {20, 20};
random->SetSize(size);
float spacing[2] = {0.7, 2.1};
random->SetSpacing( spacing );
float origin[2] = {15, 400};
random->SetOrigin( origin );
We now have an itk::PointSet object and need to cast it to an itk::PointSet object with array
type (anything derived from the itk::FixedArray class) pixels.
Since, the itk::PointSet object’s pixel type is float, We will use single element float
itk::FixedArray as our measurement vector type. And that will also be our pixel type for the
cast filter.
using MeasurementVectorType = itk::FixedArray< float, 1 >;
using ArrayPointSetType = itk::PointSet< MeasurementVectorType, 2 >;
using CasterType = itk::ScalarToArrayCastPointSetFilter< FloatPointSet2DType,
ArrayPointSetType >;
CasterType::Pointer caster = CasterType::New();
caster->SetInput( random->GetOutput() );
caster->Update();
Up to now, we spend most of time to prepare an itk::PointSet object suitable for the adaptor.
Actually, the hard part of this example is done. Now, we must define an adaptor with the image type
and instantiate an object.
using SampleType = itk::Statistics::PointSetToListSampleAdaptor<
ArrayPointSetType >;
SampleType::Pointer sample = SampleType::New();
448
Chapter 5. Statistics
1.1
3.1
5.1
2.6 (0, 0)
0.0
(1, 0)
2.0
(2, 0)
3.0
4.6
(1, 1)
0.5
(2, 1)
1.0
(1, 2)
(2, 2)
0.0
(0, 1)
2.0
6.6
(0, 2)
5.0
2.5
7.1
8.6
Figure 5.2: Conceptual histogram data structure.
The final thing we have to is to plug-in the image object to the adaptor. After that, we can use the
common methods and iterator interfaces shown in 5.1.1.
sample->SetPointSet( caster->GetOutput() );
5.1.3 Histogram
The source code for this section can be found in the file
Histogram.cxx.
This example shows how to create an itk::Statistics::Histogram object and use it.
The Histogram differs from
We call an instance in a Histogram object a bin.
the itk::Statistics::ListSample, itk::Statistics::ImageToListSampleAdaptor, or
itk::Statistics::PointSetToListSampleAdaptor in significant ways. Histograms can have
a variable number of values (float type) for each measurement vector, while the three other classes
have a fixed value (one) for all measurement vectors. Also those array-type containers can have multiple instances (data elements) with identical measurement vector values. However, in a Histogram
object, there is one unique instance for any given measurement vector.
#include "itkHistogram.h"
#include "itkDenseFrequencyContainer2.h"
Here we create a histogram with dense frequency containers. In this example we will not have any
zero-frequency measurements, so the dense frequency container is the appropriate choice. If the
5.1. Data Containers
449
histogram is expected to have many empty (zero) bins, a sparse frequency container would be the
better option. Here we also set the size of the measurement vectors to be 2 components.
using MeasurementType = float;
using FrequencyContainerType = itk::Statistics::DenseFrequencyContainer2;
using FrequencyType = FrequencyContainerType::AbsoluteFrequencyType;
constexpr unsigned int numberOfComponents = 2;
using HistogramType = itk::Statistics::Histogram< MeasurementType,
FrequencyContainerType >;
HistogramType::Pointer histogram = HistogramType::New();
histogram->SetMeasurementVectorSize( numberOfComponents );
We initialize it as a 3 × 3 histogram with equal size intervals.
HistogramType::SizeType size( numberOfComponents );
size.Fill(3);
HistogramType::MeasurementVectorType lowerBound( numberOfComponents );
HistogramType::MeasurementVectorType upperBound( numberOfComponents );
lowerBound[0] = 1.1;
lowerBound[1] = 2.6;
upperBound[0] = 7.1;
upperBound[1] = 8.6;
histogram->Initialize(size, lowerBound, upperBound );
Now the histogram is ready for storing frequency values. We will fill each bin’s frequency according
to the Figure 5.2. There are three ways of accessing data elements in the histogram:
• using instance identifiers—just like any other Sample object;
• using n-dimensional indices—just like an Image object;
• using an iterator—just like any other Sample object.
In this example, the index (0, 0) refers the same bin as the instance identifier (0) refers to. The
instance identifier of the index (0, 1) is (3), (0, 2) is (6), (2, 2) is (8), and so on.
histogram->SetFrequency(0UL,
histogram->SetFrequency(1UL,
histogram->SetFrequency(2UL,
histogram->SetFrequency(3UL,
histogram->SetFrequency(4UL,
histogram->SetFrequency(5UL,
histogram->SetFrequency(6UL,
histogram->SetFrequency(7UL,
histogram->SetFrequency(8UL,
static_cast(0.0));
static_cast(2.0));
static_cast(3.0));
static_cast(2.0f));
static_cast(0.5f));
static_cast(1.0f));
static_cast(5.0f));
static_cast(2.5f));
static_cast(0.0f));
450
Chapter 5. Statistics
Let us examine if the frequency is set correctly by calling the GetFrequency(index) method. We
can use the GetFrequency(instance identifier) method for the same purpose.
HistogramType::IndexType index( numberOfComponents );
index[0] = 0;
index[1] = 2;
std::cout << "Frequency of the bin at index " << index
<< " is " << histogram->GetFrequency(index)
<< ", and the bin's instance identifier is "
<< histogram->GetInstanceIdentifier(index) << std::endl;
For test purposes, we create a measurement vector and an index that belongs to the center bin.
HistogramType::MeasurementVectorType mv( numberOfComponents );
mv[0] = 4.1;
mv[1] = 5.6;
index.Fill(1);
We retrieve the measurement vector at the index value (1, 1), the center bin’s measurement vector.
The output is [4.1, 5.6].
std::cout << "Measurement vector at the center bin is "
<< histogram->GetMeasurementVector(index) << std::endl;
Since all the measurement vectors are unique in the Histogram class, we can determine the index
from a measurement vector.
HistogramType::IndexType resultingIndex;
histogram->GetIndex(mv,resultingIndex);
std::cout << "Index of the measurement vector " << mv
<< " is " << resultingIndex << std::endl;
In a similar way, we can get the instance identifier from the index.
std::cout << "Instance identifier of index " << index
<< " is " << histogram->GetInstanceIdentifier(index)
<< std::endl;
If we want to check if an index is valid, we use the method IsIndexOutOfBounds(index). The
following code snippet fills the index variable with (100, 100). It is obviously not a valid index.
index.Fill(100);
if ( histogram->IsIndexOutOfBounds(index) )
{
std::cout << "Index " << index << " is out of bounds." << std::endl;
}
5.1. Data Containers
451
The following code snippets show how to get the histogram size and frequency dimension.
std::cout << "Number of bins = " << histogram->Size()
<< " Total frequency = " << histogram->GetTotalFrequency()
<< " Dimension sizes = " << histogram->GetSize() << std::endl;
The Histogram class has a quantile calculation method, Quantile(dimension, percent). The
following code returns the 50th percentile along the first dimension. Note that the quantile calculation considers only one dimension.
std::cout << "50th percentile along the first dimension = "
<< histogram->Quantile(0, 0.5) << std::endl;
5.1.4 Subsample
The source code for this section can be found in the file
Subsample.cxx.
The itk::Statistics::Subsample is a derived sample. In other words, it requires another
itk::Statistics::Sample object for storing measurement vectors. The Subsample class stores a
subset of instance identifiers from another Sample object. Any Sample’s subclass can be the source
Sample object. You can create a Subsample object out of another Subsample object. The Subsample
class is useful for storing classification results from a test Sample object or for just extracting some
part of interest in a Sample object. Another good use of Subsample is sorting a Sample object. When
we use an itk::Image object as the data source, we do not want to change the order of data elements in the image. However, we sometimes want to sort or select data elements according to their
order. Statistics algorithms for this purpose accepts only Subsample objects as inputs. Changing the
order in a Subsample object does not change the order of the source sample.
To use a Subsample object, we include the header files for the class itself and a Sample class. We
will use the itk::Statistics::ListSample as the input sample.
#include "itkListSample.h"
#include "itkSubsample.h"
We need another header for measurement vectors. We are going to use the itk::Vector class in
this example.
#include "itkVector.h"
The following code snippet will create a ListSample object with three-component float measurement
vectors and put three measurement vectors into the list.
452
Chapter 5. Statistics
using MeasurementVectorType = itk::Vector< float, 3 >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack(mv);
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->PushBack(mv);
mv[0] = 3.0;
mv[1] = 8.0;
mv[2] = 6.0;
sample->PushBack(mv);
To create a Subsample instance, we define the type of the Subsample with the source sample type,
in this case, the previously defined SampleType. As usual, after that, we call the New() method to
create an instance. We must plug in the source sample, sample, using the SetSample() method.
However, with regard to data elements, the Subsample is empty. We specify which data elements,
among the data elements in the Sample object, are part of the Subsample. There are two ways of
doing that. First, if we want to include every data element (instance) from the sample, we simply
call the InitializeWithAllInstances() method like the following:
subsample->InitializeWithAllInstances();
This method is useful when we want to create a Subsample object for sorting all the data elements
in a Sample object. However, in most cases, we want to include only a subset of a Sample object.
For this purpose, we use the AddInstance(instance identifier) method in this example. In
the following code snippet, we include only the first and last instance in our subsample object from
the three instances of the Sample class.
using SubsampleType = itk::Statistics::Subsample< SampleType >;
SubsampleType::Pointer subsample = SubsampleType::New();
subsample->SetSample( sample );
subsample->AddInstance( 0UL );
subsample->AddInstance( 2UL );
The Subsample is ready for use. The following code snippet shows how to use Iterator interfaces.
5.1. Data Containers
453
SubsampleType::Iterator iter = subsample->Begin();
while ( iter != subsample->End() )
{
std::cout << "instance identifier = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency()
<< std::endl;
++iter;
}
As mentioned earlier, the instances in a Subsample can be sorted without changing the order in the
source Sample. For this purpose, the Subsample provides an additional instance indexing scheme.
The indexing scheme is just like the instance identifiers for the Sample. The index is an integer
value starting at 0, and the last value is one less than the number of all instances in a Subsample.
The Swap(0, 1) method, for example, swaps two instance identifiers of the first data element and
the second element in the Subsample. Internally, the Swap() method changes the instance identifiers
in the first and second position. Using indices, we can print out the effects of the Swap() method.
We use the GetMeasurementVectorByIndex(index) to get the measurement vector at the index
position. However, if we want to use the common methods of Sample that accepts instance identifiers, we call them after we get the instance identifiers using GetInstanceIdentifier(index)
method.
subsample->Swap(0, 1);
for ( int index = 0; index < subsample->Size(); ++index )
{
std::cout << "instance identifier = "
<< subsample->GetInstanceIdentifier(index)
<< "\t measurement vector = "
<< subsample->GetMeasurementVectorByIndex(index)
<< std::endl;
}
Since we are using a ListSample object as the source sample, the following code snippet will return
the same value (2) for the Size() and the GetTotalFrequency() methods. However, if we used a
Histogram object as the source sample, the two return values might be different because a Histogram
allows varying frequency values for each instance.
std::cout << "Size = " << subsample->Size() << std::endl;
std::cout << "Total frequency = "
<< subsample->GetTotalFrequency() << std::endl;
If we want to remove all instances that are associated with the Subsample, we call the Clear()
method. After this invocation, the Size() and the GetTotalFrequency() methods return 0.
454
Chapter 5. Statistics
subsample->Clear();
std::cout << "Size = " << subsample->Size() << std::endl;
std::cout << "Total frequency = "
<< subsample->GetTotalFrequency() << std::endl;
5.1.5 MembershipSample
The source code for this section can be found in the file
MembershipSample.cxx.
The
itk::Statistics::MembershipSample
is
derived
from
the
class
itk::Statistics::Sample that associates a class label with each measurement vector. It
needs another Sample object for storing measurement vectors. A MembershipSample object stores
a subset of instance identifiers from another Sample object. Any subclass of Sample can be the
source Sample object. The MembershipSample class is useful for storing classification results from
a test Sample object. The MembershipSample class can be considered as an associative container
that stores measurement vectors, frequency values, and class labels.
To use a MembershipSample object, we include the header files for the class itself and the Sample
class. We will use the itk::Statistics::ListSample as the input sample. We need another
header for measurement vectors. We are going to use the itk::Vector class which is a subclass of
the itk::FixedArray.
#include "itkListSample.h"
#include "itkMembershipSample.h"
#include "itkVector.h"
The following code snippet will create a ListSample object with three-component float measurement vectors and put three measurement vectors in the ListSample object.
using MeasurementVectorType = itk::Vector< float, 3 >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack(mv);
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->PushBack(mv);
5.1. Data Containers
455
mv[0] = 3.0;
mv[1] = 8.0;
mv[2] = 6.0;
sample->PushBack(mv);
To create a MembershipSample instance, we define the type of the MembershipSample using the
source sample type using the previously defined SampleType. As usual, after that, we call the
New() method to create an instance. We must plug in the source sample, Sample, using the
SetSample() method. We provide class labels for data instances in the Sample object using the
AddInstance() method. As the required initialization step for the membershipSample, we must
call the SetNumberOfClasses() method with the number of classes. We must add all instances in
the source sample with their class labels. In the following code snippet, we set the first instance’
class label to 0, the second to 0, the third (last) to 1. After this, the membershipSample has two
Subsample objects. And the class labels for these two Subsample objects are 0 and 1. The 0 class
Subsample object includes the first and second instances, and the 1 class includes the third instance.
using MembershipSampleType = itk::Statistics::MembershipSample;
MembershipSampleType::Pointer membershipSample =
MembershipSampleType::New();
membershipSample->SetSample(sample);
membershipSample->SetNumberOfClasses(2);
membershipSample->AddInstance(0U, 0UL );
membershipSample->AddInstance(0U, 1UL );
membershipSample->AddInstance(1U, 2UL );
The Size() and GetTotalFrequency() returns the same information that Sample does.
std::cout << "Total frequency = "
<< membershipSample->GetTotalFrequency() << std::endl;
The membershipSample is ready for use. The following code snippet shows how to use the
Iterator interface. The MembershipSample’s Iterator has an additional method that returns
the class label (GetClassLabel()).
MembershipSampleType::ConstIterator iter = membershipSample->Begin();
while ( iter != membershipSample->End() )
{
std::cout << "instance identifier = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency()
<< "\t class label = "
<< iter.GetClassLabel()
456
Chapter 5. Statistics
<< std::endl;
++iter;
}
To see the numbers of instances in each class subsample, we use the Size() method of the
ClassSampleType instance returned by the GetClassSample(index) method.
std::cout <<
<<
std::cout <<
<<
"class label = 0 sample size = "
membershipSample->GetClassSample(0)->Size() << std::endl;
"class label = 1 sample size = "
membershipSample->GetClassSample(1)->Size() << std::endl;
We call the GetClassSample() method to get the class subsample in the membershipSample.
The MembershipSampleType::ClassSampleType is actually a specialization of the
itk::Statistics::Subsample. We print out the instance identifiers, measurement vectors,
and frequency values that are part of the class. The output will be two lines for the two instances
that belong to the class 0.
MembershipSampleType::ClassSampleType::ConstPointer classSample =
membershipSample->GetClassSample( 0 );
MembershipSampleType::ClassSampleType::ConstIterator c_iter =
classSample->Begin();
while ( c_iter != classSample->End() )
{
std::cout << "instance identifier = " << c_iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< c_iter.GetMeasurementVector()
<< "\t frequency = "
<< c_iter.GetFrequency() << std::endl;
++c_iter;
}
5.1.6 MembershipSampleGenerator
The source code for this section can be found in the file
MembershipSampleGenerator.cxx.
To use, an MembershipSample object, we include the header files for the class itself and a Sample
class. We will use the ListSample as the input sample.
#include "itkListSample.h"
#include "itkMembershipSample.h"
5.1. Data Containers
457
We need another header for measurement vectors. We are going to use the itk::Vector class
which is a subclass of the itk::FixedArray in this example.
#include "itkVector.h"
The following code snippet will create a ListSample object with three-component float measurement vectors and put three measurement vectors in the ListSample object.
using MeasurementVectorType = itk::Vector< float, 3 >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack(mv);
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->PushBack(mv);
mv[0] = 3.0;
mv[1] = 8.0;
mv[2] = 6.0;
sample->PushBack(mv);
To create a MembershipSample instance, we define the type of the MembershipSample with
the source sample type, in this case, previously defined SampleType. As usual, after that, we
call New() method to instantiate an instance. We must plug in the source sample, sample object using the SetSample(source sample) method. However, in regard to class labels, the
membershipSample is empty. We provide class labels for data instances in the sample object using
the AddInstance(class label, instance identifier) method. As the required initialization
step for the membershipSample, we must call the SetNumberOfClasses(number of classes)
method with the number of classes. We must add all instances in the source sample with their class
labels. In the following code snippet, we set the first instance class label to 0, the second to 0, the
third (last) to 1. After this, the membershipSample has two Subclass objects. And the class labels
for these two Subclass are 0 and 1. The 0 class Subsample object includes the first and second
instances, and the 1 class includes the third instance.
using MembershipSampleType =
itk::Statistics::MembershipSample< SampleType >;
MembershipSampleType::Pointer membershipSample =
MembershipSampleType::New();
membershipSample->SetSample(sample);
458
Chapter 5. Statistics
membershipSample->SetNumberOfClasses(2);
membershipSample->AddInstance(0U, 0UL );
membershipSample->AddInstance(0U, 1UL );
membershipSample->AddInstance(1U, 2UL );
The Size() and GetTotalFrequency() methods return the same values as the sample.
std::cout << "Size = " << membershipSample->Size() << std::endl;
std::cout << "Total frequency = "
<< membershipSample->GetTotalFrequency() << std::endl;
The membershipSample is ready for use. The following code snippet shows how to use Iterator
interfaces. The MembershipSample Iterator has an additional method that returns the class label
(GetClassLabel()).
MembershipSampleType::Iterator iter = membershipSample->Begin();
while ( iter != membershipSample->End() )
{
std::cout << "instance identifier = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency()
<< "\t class label = "
<< iter.GetClassLabel()
<< std::endl;
++iter;
}
To see the numbers of instances in each class subsample, we use the GetClassSampleSize(class
label) method.
std::cout <<
<<
std::cout <<
<<
"class label = 0 sample size = "
membershipSample->GetClassSampleSize(0) << std::endl;
"class label = 1 sample size = "
membershipSample->GetClassSampleSize(0) << std::endl;
We call the GetClassSample(class label) method to get the class subsample in the
membershipSample. The MembershipSampleType::ClassSampleType is actually an specialization of the itk::Statistics::Subsample. We print out the instance identifiers, measurement
vectors, and frequency values that are part of the class. The output will be two lines for the two
instances that belong to the class 0.
5.1. Data Containers
459
MembershipSampleType::ClassSampleType::Pointer classSample =
membershipSample->GetClassSample(0);
MembershipSampleType::ClassSampleType::Iterator c_iter =
classSample->Begin();
while ( c_iter != classSample->End() )
{
std::cout << "instance identifier = " << c_iter.GetInstanceIdentifier()
<< "\t measurement vector = "
<< c_iter.GetMeasurementVector()
<< "\t frequency = "
<< c_iter.GetFrequency() << std::endl;
++c_iter;
}
5.1.7 K-d Tree
The source code for this section can be found in the file
KdTree.cxx.
The itk::Statistics::KdTree implements a data structure that separates samples in a kdimension space. The std::vector class is used here as the container for the measurement vectors
from a sample.
#include
#include
#include
#include
#include
"itkVector.h"
"itkMath.h"
"itkListSample.h"
"itkWeightedCentroidKdTreeGenerator.h"
"itkEuclideanDistanceMetric.h"
We define the measurement vector type and instantiate a itk::Statistics::ListSample object,
and then put 1000 measurement vectors in the object.
using MeasurementVectorType = itk::Vector< float, 2 >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
sample->SetMeasurementVectorSize( 2 );
MeasurementVectorType mv;
for (unsigned int i = 0; i < 1000; ++i )
{
mv[0] = (float) i;
mv[1] = (float) ((1000 - i) / 2 );
sample->PushBack( mv );
}
460
Chapter 5. Statistics
The following code snippet shows how to create two KdTree objects. The first object
itk::Statistics::KdTreeGenerator has a minimal set of information (partition dimension,
partition value, and pointers to the left and right child nodes). The second tree from the
itk::Statistics::WeightedCentroidKdTreeGenerator has additional information such as the
number of children under each node, and the vector sum of the measurement vectors belonging to
children of a particular node. WeightedCentroidKdTreeGenerator and the resulting k-d tree structure
were implemented based on the description given in the paper by Kanungo et al [28].
The instantiation and input variables are exactly the same for both tree generators. Using the
SetSample() method we plug-in the source sample. The bucket size input specifies the limit on
the maximum number of measurement vectors that can be stored in a terminal (leaf) node. A bigger
bucket size results in a smaller number of nodes in a tree. It also affects the efficiency of search.
With many small leaf nodes, we might experience slower search performance because of excessive
boundary comparisons.
using TreeGeneratorType = itk::Statistics::KdTreeGenerator< SampleType >;
TreeGeneratorType::Pointer treeGenerator = TreeGeneratorType::New();
treeGenerator->SetSample( sample );
treeGenerator->SetBucketSize( 16 );
treeGenerator->Update();
using CentroidTreeGeneratorType =
itk::Statistics::WeightedCentroidKdTreeGenerator;
CentroidTreeGeneratorType::Pointer centroidTreeGenerator =
CentroidTreeGeneratorType::New();
centroidTreeGenerator->SetSample( sample );
centroidTreeGenerator->SetBucketSize( 16 );
centroidTreeGenerator->Update();
After the generation step, we can get the pointer to the kd-tree from the generator by calling the
GetOutput() method. To traverse a kd-tree, we have to use the GetRoot() method. The method
will return the root node of the tree. Every node in a tree can have its left and/or right child node.
To get the child node, we call the Left() or the Right() method of a node (these methods do not
belong to the kd-tree but to the nodes).
We can get other information about a node by calling the methods described below in addition to the
child node pointers.
using TreeType = TreeGeneratorType::KdTreeType;
using NodeType = TreeType::KdTreeNodeType;
TreeType::Pointer tree = treeGenerator->GetOutput();
TreeType::Pointer centroidTree = centroidTreeGenerator->GetOutput();
NodeType* root = tree->GetRoot();
5.1. Data Containers
461
if ( root->IsTerminal() )
{
std::cout << "Root node is a terminal node." << std::endl;
}
else
{
std::cout << "Root node is not a terminal node." << std::endl;
}
unsigned int partitionDimension;
float partitionValue;
root->GetParameters( partitionDimension, partitionValue);
std::cout << "Dimension chosen to split the space = "
<< partitionDimension << std::endl;
std::cout << "Split point on the partition dimension = "
<< partitionValue << std::endl;
std::cout << "Address of the left chile of the root node = "
<< root->Left() << std::endl;
std::cout << "Address of the right chile of the root node = "
<< root->Right() << std::endl;
root = centroidTree->GetRoot();
std::cout << "Number of the measurement vectors under the root node"
<< " in the tree hierarchy = " << root->Size() << std::endl;
NodeType::CentroidType centroid;
root->GetWeightedCentroid( centroid );
std::cout << "Sum of the measurement vectors under the root node = "
<< centroid << std::endl;
std::cout << "Number of the measurement vectors under the left child"
<< " of the root node = " << root->Left()->Size() << std::endl;
In the following code snippet, we query the three nearest neighbors of the queryPoint on the two
tree. The results and procedures are exactly the same for both. First we define the point from which
distances will be measured.
MeasurementVectorType queryPoint;
queryPoint[0] = 10.0;
queryPoint[1] = 7.0;
Then we instantiate the type of a distance metric, create an object of this type and set the origin
of coordinates for measuring distances. The GetMeasurementVectorSize() method returns the
length of each measurement vector stored in the sample.
using DistanceMetricType =
itk::Statistics::EuclideanDistanceMetric;
DistanceMetricType::Pointer distanceMetric = DistanceMetricType::New();
462
Chapter 5. Statistics
DistanceMetricType::OriginType origin( 2 );
for ( unsigned int i = 0; i < sample->GetMeasurementVectorSize(); ++i )
{
origin[i] = queryPoint[i];
}
distanceMetric->SetOrigin( origin );
We can now set the number of neighbors to be located and the point coordinates to be used as a
reference system.
unsigned int numberOfNeighbors = 3;
TreeType::InstanceIdentifierVectorType neighbors;
tree->Search( queryPoint, numberOfNeighbors, neighbors);
std::cout <<
"\n*** kd-tree knn search result using an Euclidean distance metric:"
<< std::endl
<< "query point = [" << queryPoint << "]" << std::endl
<< "k = " << numberOfNeighbors << std::endl;
std::cout << "measurement vector : distance from querry point " << std::endl;
std::vector distances1 (numberOfNeighbors);
for ( unsigned int i = 0; i < numberOfNeighbors; ++i )
{
distances1[i] = distanceMetric->Evaluate(
tree->GetMeasurementVector( neighbors[i] ));
std::cout << "[" << tree->GetMeasurementVector( neighbors[i] )
<< "] : "
<< distances1[i]
<< std::endl;
}
Instead of using an Euclidean distance metric, Tree itself can also return the distance vector. Here
we get the distance values from tree and compare them with previous values.
std::vector distances2;
tree->Search( queryPoint, numberOfNeighbors, neighbors, distances2 );
std::cout << "\n*** kd-tree knn search result directly from tree:"
<< std::endl
<< "query point = [" << queryPoint << "]" << std::endl
<< "k = " << numberOfNeighbors << std::endl;
std::cout << "measurement vector : distance from querry point " << std::endl;
for ( unsigned int i = 0; i < numberOfNeighbors; ++i )
{
std::cout << "[" << tree->GetMeasurementVector( neighbors[i] )
<< "] : "
<< distances2[i]
<< std::endl;
if ( itk::Math::NotAlmostEquals( distances2[i], distances1[i] ) )
{
5.1. Data Containers
463
std::cerr << "Mismatched distance values by tree." << std::endl;
return EXIT_FAILURE;
}
}
As previously indicated, the interface for finding nearest neighbors in the centroid tree is very similar.
std::vector distances3;
centroidTree->Search(
queryPoint, numberOfNeighbors, neighbors, distances3 );
centroidTree->Search( queryPoint, numberOfNeighbors, neighbors );
std::cout << "\n*** Weighted centroid kd-tree knn search result:"
<< std::endl
<< "query point = [" << queryPoint << "]" << std::endl
<< "k = " << numberOfNeighbors << std::endl;
std::cout << "measurement vector : distance_by_distMetric : distance_by_tree"
<< std::endl;
std::vector distances4 (numberOfNeighbors);
for ( unsigned int i = 0; i < numberOfNeighbors; ++i )
{
distances4[i] = distanceMetric->Evaluate(
centroidTree->GetMeasurementVector( neighbors[i]));
std::cout << "[" << centroidTree->GetMeasurementVector( neighbors[i] )
<< "]
:
"
<< distances4[i]
<< "
:
"
<< distances3[i]
<< std::endl;
if ( itk::Math::NotAlmostEquals( distances2[i], distances1[i] ) )
{
std::cerr << "Mismatched distance values by centroid tree." << std::endl;
return EXIT_FAILURE;
}
}
KdTree also supports searching points within a hyper-spherical kernel. We specify the radius and
call the Search() method. In the case of the KdTree, this is done with the following lines of code.
double radius = 437.0;
tree->Search( queryPoint, radius, neighbors );
std::cout << "\nSearching points within a hyper-spherical kernel:"
<< std::endl;
std::cout << "*** kd-tree radius search result:" << std::endl
<< "query point = [" << queryPoint << "]" << std::endl
<< "search radius = " << radius << std::endl;
std::cout << "measurement vector : distance" << std::endl;
for ( auto neighbor : neighbors)
464
Chapter 5. Statistics
{
std::cout << "[" << tree->GetMeasurementVector( neighbor )
<< "] : "
<< distanceMetric->Evaluate(
tree->GetMeasurementVector( neighbor))
<< std::endl;
}
In the case of the centroid KdTree, the Search() method is used as illustrated by the following
code.
centroidTree->Search( queryPoint, radius, neighbors );
std::cout << "\n*** Weighted centroid kd-tree radius search result:"
<< std::endl
<< "query point = [" << queryPoint << "]" << std::endl
<< "search radius = " << radius << std::endl;
std::cout << "measurement vector : distance" << std::endl;
for ( auto neighbor : neighbors)
{
std::cout << "[" << centroidTree->GetMeasurementVector( neighbor )
<< "] : "
<< distanceMetric->Evaluate(
centroidTree->GetMeasurementVector( neighbor))
<< std::endl;
}
5.2 Algorithms and Functions
In the previous section, we described the data containers in the ITK statistics subsystem. We also
need data processing algorithms and statistical functions to conduct statistical analysis or statistical
classification using these containers. Here we define an algorithm to be an operation over a set of
measurement vectors in a sample. A function is an operation over individual measurement vectors.
For example, if we implement a class ( itk::Statistics::EuclideanDistance) to calculate the
Euclidean distance between two measurement vectors, we call it a function, while if we implemented
a class ( itk::Statistics::MeanCalculator) to calculate the mean of a sample, we call it an
algorithm.
5.2.1 Sample Statistics
We will show how to get sample statistics such as means and covariance from the (
itk::Statistics::Sample) classes. Statistics can tells us characteristics of a sample. Such
sample statistics are very important for statistical classification. When we know the form of the
5.2. Algorithms and Functions
465
sample distributions and their parameters (statistics), we can conduct Bayesian classification. In
ITK, sample mean and covariance calculation algorithms are implemented. Each algorithm also
has its weighted version (see Section 5.2.1). The weighted versions are used in the expectationmaximization parameter estimation process.
Mean and Covariance
The source code for this section can be found in the file
SampleStatistics.cxx.
We include the header file for the itk::Vector class that will be our measurement vector template
in this example.
#include "itkVector.h"
We will use the itk::Statistics::ListSample as our sample template. We include the header
for the class too.
#include "itkListSample.h"
The following headers are for sample statistics algorithms.
#include "itkMeanSampleFilter.h"
#include "itkCovarianceSampleFilter.h"
The following code snippet will create a ListSample object with three-component float measurement
vectors and put five measurement vectors in the ListSample object.
constexpr unsigned int MeasurementVectorLength = 3;
using MeasurementVectorType = itk::Vector< float, MeasurementVectorLength >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
sample->SetMeasurementVectorSize( MeasurementVectorLength );
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack( mv );
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->PushBack( mv );
mv[0] = 3.0;
466
Chapter 5. Statistics
mv[1] = 8.0;
mv[2] = 6.0;
sample->PushBack( mv );
mv[0] = 2.0;
mv[1] = 7.0;
mv[2] = 4.0;
sample->PushBack( mv );
mv[0] = 3.0;
mv[1] = 2.0;
mv[2] = 7.0;
sample->PushBack( mv );
To
calculate
the
mean
(vector)
of
a
sample,
we
instantiate
the
itk::Statistics::MeanSampleFilter class that implements the mean algorithm and plug
in the sample using the SetInputSample(sample*) method. By calling the Update() method,
we run the algorithm. We get the mean vector using the GetMean() method. The output from the
GetOutput() method is the pointer to the mean vector.
using MeanAlgorithmType = itk::Statistics::MeanSampleFilter< SampleType >;
MeanAlgorithmType::Pointer meanAlgorithm = MeanAlgorithmType::New();
meanAlgorithm->SetInput( sample );
meanAlgorithm->Update();
std::cout << "Sample mean = " << meanAlgorithm->GetMean() << std::endl;
The covariance calculation algorithm will also calculate the mean while performing the covariance
matrix calculation. The mean can be accessed using the GetMean() method while the covariance
can be accessed using the GetCovarianceMatrix() method.
using CovarianceAlgorithmType =
itk::Statistics::CovarianceSampleFilter;
CovarianceAlgorithmType::Pointer covarianceAlgorithm =
CovarianceAlgorithmType::New();
covarianceAlgorithm->SetInput( sample );
covarianceAlgorithm->Update();
std::cout << "Mean = " << std::endl;
std::cout << covarianceAlgorithm->GetMean() << std::endl;
std::cout << "Covariance = " << std::endl;
std::cout << covarianceAlgorithm->GetCovarianceMatrix() << std::endl;
5.2. Algorithms and Functions
467
Weighted Mean and Covariance
The source code for this section can be found in the file
WeightedSampleStatistics.cxx.
We include the header file for the itk::Vector class that will be our measurement vector template
in this example.
#include "itkVector.h"
We will use the itk::Statistics::ListSample as our sample template. We include the header
for the class too.
#include "itkListSample.h"
The following headers are for the weighted covariance algorithms.
#include "itkWeightedMeanSampleFilter.h"
#include "itkWeightedCovarianceSampleFilter.h"
The following code snippet will create a ListSample instance with three-component float measurement vectors and put five measurement vectors in the ListSample object.
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
sample->SetMeasurementVectorSize( 3 );
MeasurementVectorType mv;
mv[0] = 1.0;
mv[1] = 2.0;
mv[2] = 4.0;
sample->PushBack( mv );
mv[0] = 2.0;
mv[1] = 4.0;
mv[2] = 5.0;
sample->PushBack( mv );
mv[0] = 3.0;
mv[1] = 8.0;
mv[2] = 6.0;
sample->PushBack( mv );
mv[0] = 2.0;
mv[1] = 7.0;
mv[2] = 4.0;
sample->PushBack( mv );
468
Chapter 5. Statistics
mv[0] = 3.0;
mv[1] = 2.0;
mv[2] = 7.0;
sample->PushBack( mv );
Robust versions of covariance algorithms require weight values for measurement vectors. We have
two ways of providing weight values for the weighted mean and weighted covariance algorithms.
The first method is to plug in an array of weight values. The size of the weight value array should be
equal to that of the measurement vectors. In both algorithms, we use the SetWeights(weights).
using WeightedMeanAlgorithmType =
itk::Statistics::WeightedMeanSampleFilter;
WeightedMeanAlgorithmType::WeightArrayType weightArray( sample->Size() );
weightArray.Fill( 0.5 );
weightArray[2] = 0.01;
weightArray[4] = 0.01;
WeightedMeanAlgorithmType::Pointer weightedMeanAlgorithm =
WeightedMeanAlgorithmType::New();
weightedMeanAlgorithm->SetInput( sample );
weightedMeanAlgorithm->SetWeights( weightArray );
weightedMeanAlgorithm->Update();
std::cout << "Sample weighted mean = "
<< weightedMeanAlgorithm->GetMean() << std::endl;
using WeightedCovarianceAlgorithmType =
itk::Statistics::WeightedCovarianceSampleFilter;
WeightedCovarianceAlgorithmType::Pointer weightedCovarianceAlgorithm =
WeightedCovarianceAlgorithmType::New();
weightedCovarianceAlgorithm->SetInput( sample );
weightedCovarianceAlgorithm->SetWeights( weightArray );
weightedCovarianceAlgorithm->Update();
std::cout << "Sample weighted covariance = " << std::endl;
std::cout << weightedCovarianceAlgorithm->GetCovarianceMatrix() << std::endl;
The second method for computing weighted statistics is to plug-in a function that returns a weight
value that is usually a function of each measurement vector. Since the weightedMeanAlgorithm
and weightedCovarianceAlgorithm already have the input sample plugged in, we only need to
call the SetWeightingFunction(weights) method.
ExampleWeightFunction::Pointer weightFunction = ExampleWeightFunction::New();
weightedMeanAlgorithm->SetWeightingFunction( weightFunction );
5.2. Algorithms and Functions
469
weightedMeanAlgorithm->Update();
std::cout << "Sample weighted mean = "
<< weightedMeanAlgorithm->GetMean() << std::endl;
weightedCovarianceAlgorithm->SetWeightingFunction( weightFunction );
weightedCovarianceAlgorithm->Update();
std::cout << "Sample weighted covariance = " << std::endl;
std::cout << weightedCovarianceAlgorithm->GetCovarianceMatrix();
std::cout << "Sample weighted mean (from WeightedCovarainceSampleFilter) = "
<< std::endl << weightedCovarianceAlgorithm->GetMean()
<< std::endl;
5.2.2 Sample Generation
SampleToHistogramFilter
The source code for this section can be found in the file
SampleToHistogramFilter.cxx.
Sometimes we want to work with a histogram instead of a list of measurement vectors (e.g.
itk::Statistics::ListSample,
itk::Statistics::ImageToListSampleAdaptor,
or
itk::Statistics::PointSetToListSample) to use less memory or to
perform a particular type od analysis.
In such cases, we can import data
itk::Statistics::Histogram object using the
from a sample type to a
itk::Statistics::SampleToHistogramFiler.
We use a ListSample object as the input for the filter. We include the header files for the ListSample
and Histogram classes, as well as the filter.
#include "itkListSample.h"
#include "itkHistogram.h"
#include "itkSampleToHistogramFilter.h"
We need another header for the type of the measurement vectors. We are going to use the
itk::Vector class which is a subclass of the itk::FixedArray in this example.
#include "itkVector.h"
The following code snippet creates a ListSample object with two-component int measurement vectors and put the measurement vectors: [1,1] - 1 time, [2,2] - 2 times, [3,3] - 3 times, [4,4] - 4 times,
[5,5] - 5 times into the listSample.
470
Chapter 5. Statistics
using MeasurementType = int;
constexpr unsigned int MeasurementVectorLength = 2;
using MeasurementVectorType =
itk::Vector< MeasurementType , MeasurementVectorLength >;
using ListSampleType = itk::Statistics::ListSample< MeasurementVectorType >;
ListSampleType::Pointer listSample = ListSampleType::New();
listSample->SetMeasurementVectorSize( MeasurementVectorLength );
MeasurementVectorType mv;
for (unsigned int i = 1; i < 6; ++i)
{
for (unsigned int j = 0; j < 2; ++j)
{
mv[j] = ( MeasurementType ) i;
}
for (unsigned int j = 0; j < i; ++j)
{
listSample->PushBack(mv);
}
}
Here, we set up the size and bound of the output histogram.
using HistogramMeasurementType = float;
constexpr unsigned int numberOfComponents = 2;
using HistogramType = itk::Statistics::Histogram;
HistogramType::SizeType size( numberOfComponents );
size.Fill(5);
HistogramType::MeasurementVectorType lowerBound( numberOfComponents );
HistogramType::MeasurementVectorType upperBound( numberOfComponents );
lowerBound[0] = 0.5;
lowerBound[1] = 0.5;
upperBound[0] = 5.5;
upperBound[1] = 5.5;
Now, we set up the SampleToHistogramFilter object by passing listSample as the
input and initializing the histogram size and bounds with the SetHistogramSize(),
SetHistogramBinMinimum(), and SetHistogramBinMaximum() methods. We execute the filter
by calling the Update() method.
using FilterType = itk::Statistics::SampleToHistogramFilter< ListSampleType,
HistogramType >;
FilterType::Pointer filter = FilterType::New();
filter->SetInput( listSample );
5.2. Algorithms and Functions
471
filter->SetHistogramSize( size );
filter->SetHistogramBinMinimum( lowerBound );
filter->SetHistogramBinMaximum( upperBound );
filter->Update();
The Size() and GetTotalFrequency() methods return the same values as the sample does.
const HistogramType* histogram = filter->GetOutput();
HistogramType::ConstIterator iter = histogram->Begin();
while ( iter != histogram->End() )
{
std::cout << "Measurement vectors = " << iter.GetMeasurementVector()
<< " frequency = " << iter.GetFrequency() << std::endl;
++iter;
}
std::cout << "Size = " << histogram->Size() << std::endl;
std::cout << "Total frequency = "
<< histogram->GetTotalFrequency() << std::endl;
NeighborhoodSampler
The source code for this section can be found in the file
NeighborhoodSampler.cxx.
When we want to create an itk::Statistics::Subsample object that includes only
the measurement vectors within a radius from a center in a sample, we can use
the
itk::Statistics::NeighborhoodSampler.
In this example, we will use the
itk::Statistics::ListSample as the input sample.
We include the header files for the ListSample and the NeighborhoodSampler classes.
#include "itkListSample.h"
#include "itkNeighborhoodSampler.h"
We need another header for measurement vectors. We are going to use the itk::Vector class
which is a subclass of the itk::FixedArray.
#include "itkVector.h"
The following code snippet will create a ListSample object with two-component int measurement
vectors and put the measurement vectors: [1,1] - 1 time, [2,2] - 2 times, [3,3] - 3 times, [4,4] - 4
times, [5,5] - 5 times into the listSample.
472
Chapter 5. Statistics
using MeasurementType = int;
constexpr unsigned int MeasurementVectorLength = 2;
using MeasurementVectorType =
itk::Vector< MeasurementType , MeasurementVectorLength >;
using SampleType = itk::Statistics::ListSample< MeasurementVectorType >;
SampleType::Pointer sample = SampleType::New();
sample->SetMeasurementVectorSize( MeasurementVectorLength );
MeasurementVectorType mv;
for (unsigned int i = 1; i < 6; ++i)
{
for (unsigned int j = 0; j < 2; ++j)
{
mv[j] = ( MeasurementType ) i;
}
for (unsigned int j = 0; j < i; ++j)
{
sample->PushBack(mv);
}
}
We plug-in the sample to the NeighborhoodSampler using the SetInputSample(sample*). The
two required inputs for the NeighborhoodSampler are a center and a radius. We set these two inputs using the SetCenter(center vector*) and the SetRadius(double*) methods respectively.
And then we call the Update() method to generate the Subsample object. This sampling procedure
subsamples measurement vectors within a hyper-spherical kernel that has the center and radius specified.
using SamplerType = itk::Statistics::NeighborhoodSampler< SampleType >;
SamplerType::Pointer sampler = SamplerType::New();
sampler->SetInputSample( sample );
SamplerType::CenterType center( MeasurementVectorLength );
center[0] = 3;
center[1] = 3;
double radius = 1.5;
sampler->SetCenter( ¢er );
sampler->SetRadius( &radius );
sampler->Update();
SamplerType::OutputType::Pointer output = sampler->GetOutput();
The SamplerType::OutputType is in fact itk::Statistics::Subsample. The following code
prints out the resampled measurement vectors.
SamplerType::OutputType::Iterator iter = output->Begin();
while ( iter != output->End() )
{
std::cout << "instance identifier = " << iter.GetInstanceIdentifier()
<< "\t measurement vector = "
5.2. Algorithms and Functions
473
<< iter.GetMeasurementVector()
<< "\t frequency = "
<< iter.GetFrequency() << std::endl;
++iter;
}
5.2.3 Sample Sorting
The source code for this section can be found in the file
SampleSorting.cxx.
Sometimes we want to sort the measurement vectors in a sample. The sorted vectors may reveal
some characteristics of the sample. The insert sort, the heap sort, and the introspective sort algorithms [42] for samples are implemented in ITK. To learn pros and cons of each algorithm, please
refer to [18]. ITK also offers the quick select algorithm.
Among the subclasses of the
itk::Statistics::Sample,
only the class
itk::Statistics::Subsample allows users to change the order of the measurement vector.
Therefore, we must create a Subsample to do any sorting or selecting.
We include the header files for the itk::Statistics::ListSample and the Subsample classes.
#include "itkListSample.h"
The sorting and selecting related functions are in the include file itkStatisticsAlgorithm.h.
Note that all functions in this file are in the itk::Statistics::Algorithm namespace.
#include "itkStatisticsAlgorithm.h"
We need another header for measurement vectors. We are going to use the itk::Vector class
which is a subclass of the itk::FixedArray in this example.
We define the types of the measurement vectors, the sample, and the subsample.
#include "itkVector.h"
We define two functions for convenience. The first one clears the content of the subsample and fill
it with the measurement vectors from the sample.
void initializeSubsample(SubsampleType* subsample, SampleType* sample)
{
subsample->Clear();
subsample->SetSample(sample);
474
Chapter 5. Statistics
subsample->InitializeWithAllInstances();
}
The second one prints out the content of the subsample using the Subsample’s iterator interface.
void printSubsample(SubsampleType* subsample, const char* header)
{
std::cout << std::endl;
std::cout << header << std::endl;
SubsampleType::Iterator iter = subsample->Begin();
while ( iter != subsample->End() )
{
std::cout << "instance identifier = " << iter.GetInstanceIdentifier()
<< " \t measurement vector = "
<< iter.GetMeasurementVector()
<< std::endl;
++iter;
}
}
The following code snippet will create a ListSample object with two-component int measurement
vectors and put the measurement vectors: [5,5] - 5 times, [4,4] - 4 times, [3,3] - 3 times, [2,2] - 2
times,[1,1] - 1 time into the sample.
SampleType::Pointer sample = SampleType::New();
MeasurementVectorType mv;
for (unsigned int i = 5; i > 0; --i )
{
for (unsigned int j = 0; j < 2; ++j)
{
mv[j] = ( MeasurementType ) i;
}
for (unsigned int j = 0; j < i; ++j)
{
sample->PushBack(mv);
}
}
We create a Subsample object and plug-in the sample.
SubsampleType::Pointer subsample = SubsampleType::New();
subsample->SetSample(sample);
initializeSubsample(subsample, sample);
printSubsample(subsample, "Unsorted");
The common parameters to all the algorithms are the Subsample object (subsample), the dimension
(activeDimension) that will be considered for the sorting or selecting (only the component belong-
5.2. Algorithms and Functions
475
ing to the dimension of the measurement vectors will be considered), the beginning index, and the
ending index of the measurement vectors in the subsample. The sorting or selecting algorithms are
applied only to the range specified by the beginning index and the ending index. The ending index
should be the actual last index plus one.
The itk::InsertSort function does not require any other optional arguments. The following
function call will sort the all measurement vectors in the subsample. The beginning index is 0, and
the ending index is the number of the measurement vectors in the subsample.
int activeDimension = 0;
itk::Statistics::Algorithm::InsertSort< SubsampleType >( subsample,
activeDimension, 0, subsample->Size() );
printSubsample(subsample, "InsertSort");
We sort the subsample using the heap sort algorithm. The arguments are identical to those of the
insert sort.
initializeSubsample(subsample, sample);
itk::Statistics::Algorithm::HeapSort< SubsampleType >( subsample,
activeDimension, 0, subsample->Size() );
printSubsample(subsample, "HeapSort");
The introspective sort algorithm needs an additional argument that specifies when to stop the introspective sort loop and sort the fragment of the sample using the heap sort algorithm. Since we set the
threshold value as 16, when the sort loop reach the point where the number of measurement vectors
in a sort loop is not greater than 16, it will sort that fragment using the insert sort algorithm.
initializeSubsample(subsample, sample);
itk::Statistics::Algorithm::IntrospectiveSort< SubsampleType >
( subsample, activeDimension, 0, subsample->Size(), 16 );
printSubsample(subsample, "IntrospectiveSort");
We query the median of the measurements along the activeDimension. The last argument tells the
algorithm that we want to get the subsample->Size()/2-th element along the activeDimension.
The quick select algorithm changes the order of the measurement vectors.
initializeSubsample(subsample, sample);
SubsampleType::MeasurementType median =
itk::Statistics::Algorithm::QuickSelect< SubsampleType >( subsample,
activeDimension,
0, subsample->Size(),
subsample->Size()/2 );
std::cout << std::endl;
std::cout << "Quick Select: median = " << median << std::endl;
476
Chapter 5. Statistics
5.2.4 Probability Density Functions
The probability density function (PDF) for a specific distribution returns the probability density for
a measurement vector. To get the probability density from a PDF, we use the Evaluate(input)
method. PDFs for different distributions require different sets of distribution parameters. Before
calling the Evaluate() method, make sure to set the proper values for the distribution parameters.
Gaussian Distribution
The source code for this section can be found in the file
GaussianMembershipFunction.cxx.
The Gaussian probability density function itk::Statistics::GaussianMembershipFunction
requires two distribution parameters—the mean vector and the covariance matrix.
We include the header files for the class and the itk::Vector.
#include "itkVector.h"
#include "itkGaussianMembershipFunction.h"
We define the type of the measurement vector that will be input to the Gaussian membership function.
using MeasurementVectorType = itk::Vector< float, 2 >;
The instantiation of the function is done through the usual New() method and a smart pointer.
using DensityFunctionType =
itk::Statistics::GaussianMembershipFunction;
DensityFunctionType::Pointer densityFunction = DensityFunctionType::New();
The length of the measurement vectors in the membership function, in this case a vector of length
2, is specified using the SetMeasurementVectorSize() method.
densityFunction->SetMeasurementVectorSize( 2 );
We create the two distribution parameters and set them. The mean is [0, 0], and the covariance
matrix is a 2 x 2 matrix:
4 0
0 4
We obtain the probability density for the measurement vector:
Evaluate(measurement vector) method and print it out.
[0,
0] using the
5.2. Algorithms and Functions
477
DensityFunctionType::MeanVectorType mean( 2 );
mean.Fill( 0.0 );
DensityFunctionType::CovarianceMatrixType cov;
cov.SetSize( 2, 2 );
cov.SetIdentity();
cov *= 4;
densityFunction->SetMean( mean );
densityFunction->SetCovariance( cov );
MeasurementVectorType mv;
mv.Fill( 0 );
std::cout << densityFunction->Evaluate( mv ) << std::endl;
5.2.5 Distance Metric
Euclidean Distance
The source code for this section can be found in the file
EuclideanDistanceMetric.cxx.
The Euclidean distance function ( itk::Statistics::EuclideanDistanceMetric requires as
template parameter the type of the measurement vector. We can use this function for any subclass of
the itk::FixedArray. As a subclass of the itk::Statistics::DistanceMetric, it has two basic methods, the SetOrigin(measurement vector) and the Evaluate(measurement vector).
The Evaluate() method returns the distance between its argument (a measurement vector) and the
measurement vector set by the SetOrigin() method.
In addition to the two methods, EuclideanDistanceMetric has two more methods that return the
distance of two measurements — Evaluate(measurement vector, measurement vector) and
the coordinate distance between two measurements (not vectors) — Evaluate(measurement,
measurement). The argument type of the latter method is the type of the component of the measurement vector.
We include the header files for the class and the itk::Vector.
#include "itkVector.h"
#include "itkArray.h"
#include "itkEuclideanDistanceMetric.h"
We define the type of the measurement vector that will be input of the Euclidean distance function.
As a result, the measurement type is float.
478
Chapter 5. Statistics
using MeasurementVectorType = itk::Array< float >;
The instantiation of the function is done through the usual New() method and a smart pointer.
using DistanceMetricType =
itk::Statistics::EuclideanDistanceMetric;
DistanceMetricType::Pointer distanceMetric = DistanceMetricType::New();
We create three measurement vectors, the originPoint, the queryPointA, and the queryPointB.
The type of the originPoint is fixed in the
itk::Statistics::DistanceMetric
base class as itk::Vector< double, length of the measurement vector of the each
distance metric instance>.
The Distance metric does not know about the length of the measurement vectors. We must set it
explicitly using the SetMeasurementVectorSize() method.
DistanceMetricType::OriginType originPoint( 2 );
MeasurementVectorType queryPointA( 2 );
MeasurementVectorType queryPointB( 2 );
originPoint[0] = 0;
originPoint[1] = 0;
queryPointA[0] = 2;
queryPointA[1] = 2;
queryPointB[0] = 3;
queryPointB[1] = 3;
In the following code snippet, we show the uses of the three different Evaluate() methods.
distanceMetric->SetOrigin( originPoint );
std::cout << "Euclidean distance between the origin and the query point A = "
<< distanceMetric->Evaluate( queryPointA )
<< std::endl;
std::cout << "Euclidean distance between the two query points (A and B) = "
<< distanceMetric->Evaluate( queryPointA, queryPointB )
<< std::endl;
std::cout <<
<<
<<
<<
"Coordinate distance between "
"the first components of the two query points = "
distanceMetric->Evaluate( queryPointA[0], queryPointB[0] )
std::endl;
5.2. Algorithms and Functions
479
5.2.6 Decision Rules
A decision rule is a function that returns the index of one data element in a vector of data elements.
The index returned depends on the internal logic of each decision rule. The decision rule is an
essential part of the ITK statistical classification framework. The scores from a set of membership
functions (e.g. probability density functions, distance metrics) are compared by a decision rule and a
class label is assigned based on the output of the decision rule. The common interface is very simple.
Any decision rule class must implement the Evaluate() method. In addition to this method, certain
decision rule class can have additional method that accepts prior knowledge about the decision task.
The itk::MaximumRatioDecisionRule is an example of such a class.
The argument type for the Evaluate() method is std::vector< double >. The decision rule
classes are part of the itk namespace instead of itk::Statistics namespace.
For a project that uses a decision rule, it must link the itkCommon library. Decision rules are not
templated classes.
Maximum Decision Rule
The source code for this section can be found in the file
MaximumDecisionRule.cxx.
The itk::MaximumDecisionRule returns the index of the largest discriminant score among the
discriminant scores in the vector of discriminant scores that is the input argument of the Evaluate()
method.
To begin the example, we include the header files for the class and the MaximumDecisionRule. We
also include the header file for the std::vector class that will be the container for the discriminant
scores.
#include "itkMaximumDecisionRule.h"
#include
The instantiation of the function is done through the usual New() method and a smart pointer.
using DecisionRuleType = itk::Statistics::MaximumDecisionRule;
DecisionRuleType::Pointer decisionRule = DecisionRuleType::New();
We create the discriminant score vector and fill it with three values.
The Evaluate(
discriminantScores ) will return 2 because the third value is the largest value.
DecisionRuleType::MembershipVectorType discriminantScores;
discriminantScores.push_back( 0.1 );
discriminantScores.push_back( 0.3 );
discriminantScores.push_back( 0.6 );
480
Chapter 5. Statistics
std::cout << "MaximumDecisionRule: The index of the chosen = "
<< decisionRule->Evaluate( discriminantScores )
<< std::endl;
Minimum Decision Rule
The source code for this section can be found in the file
MinimumDecisionRule.cxx.
The Evaluate() method of the itk::MinimumDecisionRule returns the index of the smallest
discriminant score among the vector of discriminant scores that it receives as input.
To begin this example, we include the class header file. We also include the header file for the
std::vector class that will be the container for the discriminant scores.
#include "itkMinimumDecisionRule.h"
#include