ITK/Examples/WishList/Segmentation/MeanShiftClustering

This example uses the "old" statistics framework. MeanShift is not yet implemented (11/30/2010) in the new framework, so it is currently unavailable.

This example is 140 lines, we should probably get it down to < 50.

SampleMeanShiftClustering.cxx

<source lang="cpp">

1. include <itkImageFileReader.h>
2. include <itkImageRegionIterator.h>
3. include <itkScalarImageToListAdaptor.h>
4. include <itkKdTree.h>
5. include <itkKdTreeGenerator.h>
6. include <itkMeanShiftModeCacheMethod.h>
7. include <itkHypersphereKernelMeanShiftModeSeeker.h>
8. include <itkSampleMeanShiftBlurringFilter.h>
9. include <itkSampleMeanShiftClusteringFilter.h>

1. include <itkImageFileWriter.h>

int main(int argc, char* argv[] ) {

 if (argc < 2)
   {
     std::cout << "ERROR: data file name argument missing." 
               << std::endl ;
     return EXIT_FAILURE;
   }

 std::string inputFilename = argv[1];
 
 typedef unsigned char PixelType ;
 typedef itk::Image< PixelType, 2 > ImageType ;
 typedef itk::ImageFileReader< ImageType > ImageReaderType ;
 ImageReaderType::Pointer imageReader = ImageReaderType::New() ;

 imageReader->SetFileName(inputFilename.c_str());
 imageReader->Update() ;
 ImageType::Pointer image = imageReader->GetOutput() ;
 
 typedef itk::Statistics::ScalarImageToListAdaptor< ImageType >
   ListSampleType ;
 
 ListSampleType::Pointer listSample = 
   ListSampleType::New() ;
 listSample->SetImage( image ) ;

 typedef itk::Statistics::KdTreeGenerator< ListSampleType > 
   TreeGeneratorType ;
 TreeGeneratorType::Pointer treeGenerator = TreeGeneratorType::New() ;
 treeGenerator->SetSample( listSample ) ;
 treeGenerator->SetBucketSize( 200 ) ;
 treeGenerator->Update() ;

 typedef TreeGeneratorType::KdTreeType TreeType ;
 TreeType::Pointer tree = treeGenerator->GetOutput() ;

 typedef itk::Statistics::HypersphereKernelMeanShiftModeSeeker< 
   TreeType > ModeSeekerType ;
 ModeSeekerType::Pointer modeSeeker = ModeSeekerType::New() ;
 modeSeeker->SetInputSample( tree ) ;

// modeSeeker->SetInputSample( listSample ) ;

 modeSeeker->SetSearchRadius( 4.0 ) ;

 typedef itk::Statistics::MeanShiftModeCacheMethod< TreeType::MeasurementVectorType > CacheMethodType ;
 CacheMethodType::Pointer cacheMethod = CacheMethodType::New() ;
 cacheMethod->SetMaximumEntries(255) ;
 cacheMethod->SetMaximumConsecutiveFailures(100) ;
 cacheMethod->SetHitRatioThreshold( 0.5 ) ;
 modeSeeker->SetCacheMethod( cacheMethod.GetPointer() ) ;

 typedef itk::Statistics::SampleMeanShiftBlurringFilter< TreeType >
   FilterType ;
 FilterType::Pointer filter = FilterType::New() ;
 filter->SetInputSample( tree ) ;
 filter->SetMeanShiftModeSeeker( modeSeeker ) ;
 filter->Update() ;

 std::cout << "Cache statistics: " << std::endl ;
 cacheMethod->Print(std::cout) ;

 typedef ImageType OutputImageType ;
 OutputImageType::Pointer outputImage = OutputImageType::New() ;
 outputImage->SetRegions( image->GetLargestPossibleRegion() ) ;
 outputImage->Allocate() ;

 typedef itk::ImageRegionIterator< OutputImageType > ImageIteratorType ;
 ImageIteratorType io_iter( outputImage,
                            outputImage->GetLargestPossibleRegion() ) ;
 io_iter.GoToBegin() ;

 FilterType::OutputType::Pointer output = filter->GetOutput() ;
 FilterType::OutputType::Iterator fo_iter = output->Begin() ;
 FilterType::OutputType::Iterator fo_end = output->End() ;

 while ( fo_iter != fo_end )
   {
   io_iter.Set( (PixelType) fo_iter.GetMeasurementVector()[0]) ;
   ++fo_iter ;
   ++io_iter ;
   }

 ListSampleType::Pointer listSample2 = ListSampleType::New() ;
 listSample2->SetImage( outputImage ) ;

 TreeGeneratorType::Pointer treeGenerator2 = TreeGeneratorType::New() ;
 treeGenerator2->SetSample( listSample2 ) ;
 treeGenerator2->SetBucketSize( 200 ) ;
 treeGenerator2->Update() ;

 typedef itk::Statistics::SampleMeanShiftClusteringFilter< TreeType >
   ClusteringMethodType ;

 ClusteringMethodType::Pointer clusteringMethod =
   ClusteringMethodType::New() ;
 clusteringMethod->SetInputSample( treeGenerator2->GetOutput() ) ;
 clusteringMethod->SetThreshold( 0.5 ) ;
 clusteringMethod->SetMinimumClusterSize( 16 ) ;
 clusteringMethod->DebugOn() ;
 clusteringMethod->Update() ;

 // save clustered image
 OutputImageType::Pointer clusterMap = OutputImageType::New() ;
 clusterMap->SetRegions( image->GetLargestPossibleRegion() ) ;
 clusterMap->Allocate() ;
 
 ImageIteratorType m_iter( clusterMap, 
                           clusterMap->GetLargestPossibleRegion() ) ;
 m_iter.GoToBegin() ;
 
 ClusteringMethodType::ClusterLabelsType clusterLabels = 
   clusteringMethod->GetOutput() ;
 
 ClusteringMethodType::ClusterLabelsType::iterator co_iter = 
   clusterLabels.begin() ;
 
 while ( co_iter != clusterLabels.end() )
   {
   m_iter.Set( (PixelType) *co_iter ) ;
   ++co_iter ;
   ++m_iter ;
   }
 
 typedef itk::ImageFileWriter< OutputImageType > ImageWriterType ;
 ImageWriterType::Pointer map_writer = ImageWriterType::New() ;
 map_writer->SetFileName("clustered_sf4.png") ;
 map_writer->SetInput( clusterMap ) ;
 map_writer->Update() ;

 return EXIT_SUCCESS;

}

</source>

CMakeLists.txt

<source lang="cmake"> cmake_minimum_required(VERSION 2.6)

PROJECT(SampleMeanShiftClustering)

FIND_PACKAGE(ITK REQUIRED) INCLUDE(${ITK_USE_FILE})

ADD_EXECUTABLE(SampleMeanShiftClustering SampleMeanShiftClustering.cxx) TARGET_LINK_LIBRARIES(SampleMeanShiftClustering ITKCommon ITKNumerics ITKStatistics ITKIO)

</source>