最近刚刚配置好ITK环境，兴奋之余赶紧跑了图像分割的三个示例，分别是BinaryThresholdImageFilter，WatershedSegmentation和ThresholdSegmentationLevelSetImageFilter，也即是双阈值法，分水岭以及水平集方法，这些cxx源文件在ITK的example下面均可以找到。
一 双阈值分割法
#include "itkBinaryThresholdImageFilter.h"
#include "itkImage.h"
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"int main( int argc, char * argv[] )
{
if( argc < 7 )
{
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImageFile outputImageFile ";
std::cerr << " lowerThreshold upperThreshold ";
std::cerr << " outsideValue insideValue " << std::endl;
return EXIT_FAILURE;
}
typedef unsigned char InputPixelType;
typedef unsigned char OutputPixelType;
typedef itk::Image< InputPixelType, 2 > InputImageType;
typedef itk::Image< OutputPixelType, 2 > OutputImageType;
typedef itk::BinaryThresholdImageFilter<
InputImageType, OutputImageType > FilterType;
typedef itk::ImageFileReader< InputImageType > ReaderType;
typedef itk::ImageFileWriter< OutputImageType > WriterType;
ReaderType::Pointer reader = ReaderType::New();
FilterType::Pointer filter = FilterType::New();
WriterType::Pointer writer = WriterType::New();</span></p><p><span style="white-space: pre;">
writer->SetInput( filter->GetOutput() );
reader->SetFileName( argv[1] );
filter->SetInput( reader->GetOutput() );
const OutputPixelType outsideValue = atoi( argv[5] );
const OutputPixelType insideValue = atoi( argv[6] );
filter->SetOutsideValue( outsideValue );
filter->SetInsideValue( insideValue );
const InputPixelType lowerThreshold = atoi( argv[3] );
const InputPixelType upperThreshold = atoi( argv[4] );
filter->SetLowerThreshold( lowerThreshold );
filter->SetUpperThreshold( upperThreshold );
filter->Update();
writer->SetFileName( argv[2] );
writer->Update();
return EXIT_SUCCESS;
}
下面来一行行分析代码：包含文件#include "itkImage.h"
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
这里面有ITK图像的模板，ITK的图像类型定义与OpenCV略有不同，它采用所谓的数据处理管道结构，如下：<span style="font-family: Arial, Helvetica, sans-serif;">ITK中的数据处理对象分为3种类型</span>1 Source类型对象的输出为数据对象，包括各种图像文件读取类
2 Filter类型对象以一个或多个数据对象作为输入、然后输出一个或多个数据对象，包括数据处理算法类
3 Mapper类型对象是管道处理的最后一步，它将最终处理结果存到硬盘<span style="font-family: Arial, Helvetica, sans-serif;">首先要定义输入输出图像的类型，</span>
typedef unsigned char InputPixelType;
typedef unsigned char OutputPixelType;
typedef itk::Image< InputPixelType, 2 > InputImageType;
typedef itk::Image< OutputPixelType, 2 > OutputImageType;</p><p>ImageType可以是任何类型（uchar,float......）任意维度的图像</p><p>typedef itk::BinaryThresholdImageFilter<InputImageType, OutputImageType > FilterType;</p><p>定义了数据处理算法类。</p><p>所有的类型都要通过new()操作符进行显示初始化。</p><p><span style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);">ReaderType::Pointer reader = ReaderType::New();</span><br style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);" /><span style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);">FilterType::Pointer filter = FilterType::New();</span><br style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);" /><span style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);">WriterType::Pointer writer = WriterType::New();</span>
</p><span style="font-family:Arial, Helvetica, sans-serif;"></span><pre name="code" class="cpp"><p>writer->SetInput( filter->GetOutput() );
reader->SetFileName( argv[1] );
</p><p>filter->SetInput( reader->GetOutput() );</p><span style="font-family:Arial, Helvetica, sans-serif;"><span style="font-family: monospace; white-space: pre; background-color: rgb(240, 240, 240);">最后使用Update</span><span style="white-space: pre; background-color: rgb(240, 240, 240);">方法写入硬盘。
</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);">最终获得的效果如下：</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);"><img src="https://img-blog.csdn.net/20141108194941533" alt="" /><img src="https://img-blog.csdn.net/20141108195042484" alt="" />
</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);">总结：这是简单的阈值法，所以对于有噪声，以及由于偏移场造成空间异构的MRI医学图像，无法获得较好的分割结果。</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);">
</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);">二<span style="white-space:pre"> </span>分水岭分割法
</span></span><span style="font-family:Arial, Helvetica, sans-serif;"><span style="white-space: pre; background-color: rgb(240, 240, 240);"></span></span><pre name="code" class="cpp">#include "itkVectorGradientAnisotropicDiffusionImageFilter.h"
#include "itkVectorGradientMagnitudeImageFilter.h"
#include "itkWatershedImageFilter.h"
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkVectorCastImageFilter.h"
#include "itkScalarToRGBPixelFunctor.h"
int main( int argc, char *argv[] )
{
if (argc < 8 )
{
std::cerr << "Missing Parameters " << std::endl;
std::cerr << "Usage: " << argv[0];
std::cerr << " inputImage outputImage conductanceTerm diffusionIterations lowerThreshold outputScaleLevel gradientMode " << std::endl;
return 1;
}
typedef itk::RGBPixel< unsigned char > RGBPixelType;
typedef itk::Image< RGBPixelType, 2 > RGBImageType;
typedef itk::Vector< float, 3 > VectorPixelType;
typedef itk::Image< VectorPixelType, 2 > VectorImageType;
typedef itk::Image< itk::IdentifierType, 2 > LabeledImageType;
typedef itk::Image< float, 2 > ScalarImageType;
typedef itk::ImageFileReader< RGBImageType > FileReaderType;
typedef itk::VectorCastImageFilter< RGBImageType, VectorImageType >
CastFilterType;
typedef itk::VectorGradientAnisotropicDiffusionImageFilter<
VectorImageType, VectorImageType >
DiffusionFilterType;
typedef itk::VectorGradientMagnitudeImageFilter< VectorImageType >
GradientMagnitudeFilterType;
typedef itk::WatershedImageFilter< ScalarImageType >
WatershedFilterType;
typedef itk::ImageFileWriter<RGBImageType> FileWriterType;
FileReaderType::Pointer reader = FileReaderType::New();
reader->SetFileName(argv[1]);
CastFilterType::Pointer caster = CastFilterType::New();
DiffusionFilterType::Pointer diffusion = DiffusionFilterType::New();
diffusion->SetNumberOfIterations( atoi(argv[4]) );
diffusion->SetConductanceParameter( atof(argv[3]) );
diffusion->SetTimeStep(0.125);
GradientMagnitudeFilterType::Pointer
gradient = GradientMagnitudeFilterType::New();
gradient->SetUsePrincipleComponents(atoi(argv[7]));
WatershedFilterType::Pointer watershed = WatershedFilterType::New();
watershed->SetLevel( atof(argv[6]) );
watershed->SetThreshold( atof(argv[5]) );
typedef itk::Functor::ScalarToRGBPixelFunctor<unsigned long>
ColorMapFunctorType;
typedef itk::UnaryFunctorImageFilter<LabeledImageType,
RGBImageType, ColorMapFunctorType> ColorMapFilterType;
ColorMapFilterType::Pointer colormapper = ColorMapFilterType::New();
FileWriterType::Pointer writer = FileWriterType::New();
writer->SetFileName(argv[2]);
caster->SetInput(reader->GetOutput());
diffusion->SetInput(caster->GetOutput());
gradient->SetInput(diffusion->GetOutput());
watershed->SetInput(gradient->GetOutput());
colormapper->SetInput(watershed->GetOutput());
writer->SetInput(colormapper->GetOutput());
try
{
writer->Update(); }
catch (itk::ExceptionObject &e)
{
std::cerr << e << std::endl;
}
return 0;
}包含文件#include "itkVectorGradientAnisotropicDiffusionImageFilter.h"
#include "itkVectorGradientMagnitudeImageFilter.h"
#include "itkWatershedImageFilter.h"
#include "itkImageFileReader.h"
#include "itkImageFileWriter.h"
#include "itkVectorCastImageFilter.h"
#include "itkScalarToRGBPixelFunctor.h"VectorCastImageFilter包括了从RGB像素类型至数值像素类型的转换，ITK中存在<span style="font-family: Arial, Helvetica, sans-serif;">RGBPixel这样的结构，而在OpenCV中RGB像素类型与灰度像素类型的区别只是通过通道数的不同来区分。</span><pre name="code" class="cpp">CastFilterType这样的一个算法类似于与OpenCV中的convert函数，整个分水岭算法的流程是这样的；<pre name="code" class="cpp">caster->SetInput(reader->GetOutput()); //输入图像转换成浮点数实值类型，输入为uchar型图像
diffusion->SetInput(caster->GetOutput()); //扩散滤波，输入为实值类型图像，输出为经过扩散滤波的图像
gradient->SetInput(diffusion->GetOutput());//梯度场滤波，输入为经过扩散滤波的图像，输出为梯度图像
watershed->SetInput(gradient->GetOutput());//输出梯度图像，其实好的分水岭算法，输出不是简单的梯度图像，往往是带标记的梯度极值图像，输出则是标签图像//标签图像是什么呢，就是经过分水岭浸没之后，图像会形成一个个的区域，不同的区域采用不同的整数值（1,2,3...）来标记，而区域与区域之间的边界只有一个像素宽，我//初学图像处理时，自己动手写过带标记的分水岭的源程序，这个一个像素宽的边界就没有写出来
colormapper->SetInput(watershed->GetOutput());//将标签图像转换成彩色图像，以便显示
writer->SetInput(colormapper->GetOutput());使用默认的参数2 10 0.001 0.15 0，所得到的结果如下，<span style="color:#ff0000;">效果简直没法忍啊！</span>不过今天的目的只是观看下原始的ITK算法，以后将会来仔细研究分水岭算法用于彩色图像分割。<img src="https://img-blog.csdn.net/20141109134204563" alt="" /><img src="https://img-blog.csdn.net/20141109134218042" alt="" />
<pre name="code" class="cpp"><span style="font-family:Arial, Helvetica, sans-serif;">三<span> </span>水平集分割法</span>水平集方法相对以上的方法来说就要复杂很多，此处只是演示，具体的方法以后再说。ITK里面的水平集方法的步骤是；参数如下：60 116 5 150 180分别对应seedX seedY InitialDistance LowerThreshold UpperThreshold"，它们都是什么呢，首先我们要解释一个函数，快速步进法<span style="font-family: Arial, Helvetica, sans-serif;">FastMarchingImageFilter。</span><span style="font-family: Arial, Helvetica, sans-serif;">快速步进法是由Sethian和Tsitsiklis独立提出的基于模板变形的演变方法。它通过将问题放到高一维的情景中进行求精，比如要跟踪曲线的变形，转化为跟踪曲面的等高线的方法，因为在图像中，曲面的演化非常容易（就是灰度值的变化），而曲线的演化，很难用数学的公司来表述，站在更高维度空间进行思考，这就是水平集思想的精髓。那么它的演化就通过几个变量进行控制：（1）初始演化曲面（对于二类分割，就是正负区域，此处通过seedX和seedY来进行双阈值化来提供初始演化曲面）（2）曲面演化方向，也就是朝着梯度方向进行演化（3）迭代次数，步长以及等等一些控制参数。水平集方法的流程就是：初始演化曲面（种子区域）-> 按照一定的速度演化，迭代直至收敛-> 阈值化得到零水平集，即分割线。得到的结果如下：</span> <img src="https://img-blog.csdn.net/20141109143012534" alt="" /><img src="https://img-blog.csdn.net/20141109143024842" alt="" /><span style="color:#ff0000;">换一个图片，就极有可能要换一种参数，这就是水平集最大的软肋。具体的细节，咱们留待以后慢慢讨论。</span>可能看完这篇博客你什么都没学习，依旧稀里糊涂；我也是，因为毕竟自己也是刚学，而且主要目标是分享很具体的实例。但是万里长城第一步嘛，以后会一步接一步。想进一步了解ITK文件的读写：请进http://blog.163.com/jacky_ling0/blog/static/13739257120108335625694</pre><pre><span style="font-family:Arial, Helvetica, sans-serif;">
</span><span style="font-family:Arial, Helvetica, sans-serif;">
</span>