ITK图像配准介绍

完整WORD文档请查看http://download.csdn.net/source/2285738

 

 

1.什么叫配准

 

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.

 

2.配准的框架

 

3．Transform

In the Insight Toolkit, itk::Transform objects encapsulate the mapping of points and vectors from an input space to an output space.

 

ITK provides a variety of transforms from simple translation, rotation and scaling to general affine and kernel transforms.

 

下面详细介绍一些主要的变换方法：

 

3.1 Translation Transform

 

3.2 Scale Transform

 

3.3 Euler2DTransform

 

3.4 CenteredRigid2DTransform

 

 

3.5 Similarity2DTransform

       can be seen as a rigid transform combined with an isotropic scaling factor.

 

3.6 AffineTransform

 

3.7 变换的分类

 

4. Interpolators

 

4.1 为什么要插值?

 

 

4.2 最常用的两种插值方法

 

4.2.1 Nearest Neighbor Interpolation

 

 

4.2.2 Linear Interpolation

 

 

5. Metrics

 

In ITK, itk::ImageToImageMetric 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.

 

下面介绍一些常用的Metrics

 

5.1 Mean Squares Metric

 

The itk::MeanSquaresImageToImageMetric computes the mean squared pixel-wise difference in intensity between image A and B over a user defined region

5.2 Normalized Correlation Metric

 

6. Optimizers

 

The basic input to an optimizer is a cost function object

 

常用的两种方法

 

6.1 Gradient Descent

 

 Advances parameters in the direction of the gradient where the step size is governed by a learning rate ( itk::GradientDescentOptimizer).

 

6.2 Regular Step Gradient Descent

Advances parameters in the direction of the gradient where a bipartition scheme is used to compute the step size (itk::RegularStepGradientDescentOptimizer).

 

7.一个简单的例子

 

 

实现平移配准

#include "itkImageRegistrationMethod.h" #include "itkTranslationTransform.h" #include "itkLinearInterpolateImageFunction.h" #include "itkMeanSquaresImageToImageMetric.h" #include "itkRegularStepGradientDescentOptimizer.h" #include "itkImage.h"   #include "itkImageFileReader.h" #include "itkImageFileWriter.h"   #include "itkResampleImageFilter.h"   #include "itkCommand.h"   class CommandIterationUpdate : public itk::Command { public:        typedef CommandIterationUpdate Self;        typedef itk::Command           Superclass;          typedef itk::SmartPointer<Self> Pointer;          itkNewMacro( Self );   protected:        CommandIterationUpdate()  {};   public:        typedef itk::RegularStepGradientDescentOptimizer OptimizerType;        typedef const OptimizerType      *      OptimizerPointer;          void Execute(itk::Object *caller, const itk::EventObject & event)        {               Execute((const itk::Object *)caller,event);        }          void Execute(const itk::Object * object, const itk::EventObject & event)        {               OptimizerPointer optimizer = dynamic_cast< OptimizerPointer > ( object );               if( ! itk::IterationEvent().CheckEvent( &event))               {                      return;               }               std::cout << optimizer->GetCurrentIteration() << " " << std::endl;               std::cout <<  optimizer->GetValue() << " " << std::endl;               std::cout <<  optimizer->GetCurrentPosition() << " " << std::endl;          } };   int main() {        const unsigned int Dimension = 2;        typedef unsigned char PixelType;          typedef itk::Image<PixelType,Dimension> FixImageType;        typedef itk::Image<PixelType,Dimension> MoveImageType;          typedef itk::TranslationTransform<double,Dimension> TransformType;        typedef itk::LinearInterpolateImageFunction< MoveImageType , double > InterpolateType;        typedef itk::MeanSquaresImageToImageMetric< FixImageType , MoveImageType > MetricType;        typedef itk::RegularStepGradientDescentOptimizer OptimizerType;        typedef itk::ImageRegistrationMethod< FixImageType , MoveImageType > RegistrationType;          TransformType::Pointer transform = TransformType::New(); InterpolateType::Pointer interpolator = InterpolateType::New();        MetricType::Pointer metric = MetricType::New();        OptimizerType::Pointer optimizer = OptimizerType::New();        RegistrationType::Pointer registration = RegistrationType::New();          registration->SetTransform( transform );        registration->SetInterpolator( interpolator );        registration->SetMetric( metric );        registration->SetOptimizer( optimizer );          typedef itk::ImageFileReader< FixImageType > FixedImageReaderType;        typedef itk::ImageFileReader< MoveImageType > MoveImageReaderType;          FixedImageReaderType::Pointer fixedImageReader = FixedImageReaderType::New();        MoveImageReaderType::Pointer  moveImageReader  = MoveImageReaderType::New();          const char* FixedImageRoot = "D://pic//BFixed.png";        const char* MoveImageRoot = "D://pic//BMove2.png";               fixedImageReader->SetFileName( FixedImageRoot );        moveImageReader->SetFileName( MoveImageRoot );          registration->SetFixedImage( fixedImageReader->GetOutput());        registration->SetMovingImage( moveImageReader->GetOutput());          fixedImageReader->Update();        registration->SetFixedImageRegion( fixedImageReader->GetOutput()->GetBufferedRegion());          typedef RegistrationType::ParametersType ParametersType;        ParametersType initialParameters( transform->GetNumberOfParameters() );          initialParameters[0] = 0.0;        initialParameters[1] = 0.0;              registration->SetInitialTransformParameters ( initialParameters );          optimizer->SetMaximumStepLength ( 4.00 );        optimizer->SetMinimumStepLength ( 0.05 );          optimizer->SetNumberOfIterations( 200 );          CommandIterationUpdate::Pointer observer = CommandIterationUpdate::New();          optimizer->AddObserver( itk::IterationEvent(), observer );          try        {               registration->StartRegistration();               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;               system("pause");               return -1;        }          ParametersType finalParameters = registration->GetLastTransformParameters();          const double TranslationAlongX = finalParameters[0];        const double TranslationAlongY = finalParameters[1];          const unsigned int numberOfIterations = optimizer->GetCurrentIteration();        const double bestValue = optimizer->GetValue();          std::cout << " TranslationAlongX = " << TranslationAlongX << std::endl;        std::cout << " TranslationAlongY = " << TranslationAlongY << std::endl;          std::cout << " numberOfIterations = " << numberOfIterations << std::endl;          system("pause");          typedef itk::ResampleImageFilter< MoveImageType , FixImageType > ResampleFilterType;               ResampleFilterType::Pointer resampler = ResampleFilterType::New();        resampler->SetInput( moveImageReader->GetOutput() );               resampler->SetTransform( registration->GetOutput()->Get() );               FixImageType::Pointer fixedImage = fixedImageReader->GetOutput();        resampler->SetSize( fixedImage->GetLargestPossibleRegion().GetSize() );        resampler->SetOutputOrigin(  fixedImage->GetOrigin() );        resampler->SetOutputSpacing( fixedImage->GetSpacing() );        resampler->SetOutputDirection( fixedImage->GetDirection() );        resampler->SetDefaultPixelValue( 100 );                      //     typedef itk::ImageFileWriter< OutputImageType >  WriterType;        typedef itk::Image<PixelType,Dimension> OutputImageType;        typedef itk::ImageFileWriter< OutputImageType >  WriterType;        WriterType::Pointer      writer =  WriterType::New();        //CastFilterType::Pointer  caster =  CastFilterType::New();               const char* writeFile = "D://pic//write.png";        writer->SetFileName( writeFile );               //caster->SetInput( resampler->GetOutput() );        //writer->SetInput( caster->GetOutput() );        writer->SetInput( resampler->GetOutput());               try        {               writer->Update();               }        catch( itk::ExceptionObject &err )        {               std::cerr << "ExceptionObject caught ! " << std::endl;               std::cerr << err << std::endl;               system("pause");               return -1;        }                      return 0 ;   }

 

 

8. Multi-Resolution Registration

 

 

Multi-Resolution配准方法，可以增强配准的准确性，速度以及鲁棒性。如图所示，它为多次配准。采用金字塔模型，从第一次的粗糙配准到后面越来越细致的配准，以此来达到程序的精确匹配。

 

代码片段

void Execute(itk::Object * object, const itk::EventObject & event) {      if(!(itk::IterationEvent().CheckEvent( &event )))      {          return;      }      RegistrationPointer registration =      dynamic_cast<RegistrationPointer>( object );      OptimizerPointer optimizer = dynamic_cast< OptimizerPointer >(      registration->GetOptimizer() );      std::cout << "-------------------------------------" << std::endl;      std::cout << "MultiResolution Level : " << registration->GetCurrentLevel() << std::endl;      std::cout << std::endl;        if( registration->GetCurrentLevel() == 0 )      {          optimizer->SetMaximumStepLength( 16.00 );          optimizer->SetMinimumStepLength( 0.01 );      }      else      {          optimizer->SetMaximumStepLength( optimizer->GetMaximumStepLength() / 4.0 );          optimizer->SetMinimumStepLength( optimizer->GetMinimumStepLength() / 10.0 );      }   }

 

9.Resample Image Filter

 

       适用于配准后的图像输出.

 

       Inputs: an image , transform , interpolator .

 

       它针对的是空间坐标系，而不是像素，单位是毫米。它是把像素点换算成坐标点，然后再赋值的。

 

它的坐标系起点为左下角。