Computes the Kubler Lieblach(KL) metric between the histogram of the two images to be registered and a training histogram. More...

#include <itkKullbackLeiblerCompareHistogramImageToImageMetric.h>

Inheritance diagram for itk::KullbackLeiblerCompareHistogramImageToImageMetric< TFixedImage, TMovingImage >:

Collaboration diagram for itk::KullbackLeiblerCompareHistogramImageToImageMetric< TFixedImage, TMovingImage >:

List of all members.

Public Types
typedef SmartPointer< const Self >	ConstPointer
typedef Superclass::DerivativeType	DerivativeType
typedef Superclass::FixedImageConstPointer	FixedImageConstPointer
typedef Superclass::FixedImageType	FixedImageType
typedef Superclass::GradientPixelType	GradientPixelType
typedef Superclass::HistogramFrequencyType	HistogramFrequencyType
typedef Superclass::HistogramIteratorType	HistogramIteratorType
typedef Superclass::MeasurementVectorType	HistogramMeasurementVectorType
typedef Superclass::HistogramPointerType	HistogramPointerType
typedef Superclass::HistogramSizeType	HistogramSizeType
typedef Superclass::HistogramType	HistogramType
typedef Superclass::InterpolatorPointer	InterpolatorPointer
typedef Superclass::InterpolatorType	InterpolatorType
typedef Superclass::MeasureType	MeasureType
typedef Superclass::MovingImageConstPointer	MovingImageConstPointer
typedef Superclass::MovingImageType	MovingImageType
typedef SmartPointer< Self >	Pointer
typedef Superclass::RealType	RealType
typedef KullbackLeiblerCompareHistogramImageToImageMetric	Self
typedef CompareHistogramImageToImageMetric < TFixedImage, TMovingImage >	Superclass
typedef Superclass::ConstPointer	TransformConstPointer
typedef Superclass::TransformJacobianType	TransformJacobianType
typedef Superclass::TransformParametersType	TransformParametersType
typedef Superclass::TransformPointer	TransformPointer
typedef Superclass::TransformType	TransformType
Public Member Functions
virtual ::itk::LightObject::Pointer	CreateAnother (void) const
virtual const double &	GetEpsilon ()
virtual const char *	GetNameOfClass () const
unsigned int	GetNumberOfParameters (void) const
void	Initialize () throw ( ExceptionObject )
virtual void	SetEpsilon (double _arg)
Static Public Member Functions
static Pointer	New ()
Protected Member Functions
virtual MeasureType	EvaluateMeasure (HistogramType &histogram) const
void	FormTrainingHistogram () throw ( ExceptionObject )

	KullbackLeiblerCompareHistogramImageToImageMetric ()
virtual	~KullbackLeiblerCompareHistogramImageToImageMetric ()
void	PrintSelf (std::ostream &os, Indent indent) const
Protected Attributes
double	m_Epsilon
Private Member Functions
	KullbackLeiblerCompareHistogramImageToImageMetric (Self const &)
void	operator= (Self const &)

Detailed Description

template<class TFixedImage, class TMovingImage>
class itk::KullbackLeiblerCompareHistogramImageToImageMetric< TFixedImage, TMovingImage >

Computes the Kubler Lieblach(KL) metric between the histogram of the two images to be registered and a training histogram.

This class is templated over the type of the fixed and moving images to be compared.

This class computers the KL-metric by comparing the histograms of the testing histogram formed by the overlap of intensities in the images, to a training histogram. It is based on the following paper:

Albert C.S. Chung, William M. Wells III, Alexander Norbash, and W. Eric L. Grimson, Multi-modal Image Registration by Minimising Kullback-Leibler Distance, In Medical Image Computing and Computer-Assisted Intervention - MICCAI 2002, LNCS 2489, pp. 525 - 532.

The metric is given by KL(P_test||P_train) = Sum_{i1,i2} P_test(i1,i2) vcl_log(P_test(i1,i2)/P_train(i1,i2)) where P_test and P_train are probabilities given my normalized histograms, and i1 and i2 are the intensity bins in the histogram.

PARAMETERS: Epsilon is added to every bin in both histograms. This prevents division by zero problems. Epsilon should generally be set to a number smaller than one divided by the total number bins in the histogram. So, for a 256 by 256 histogram, Epsilon should be much less than 1e-5. Tests have shown that choices of epsilon are not very important as long as it is small enough. The default is 1e-12. I doubt you will need to change it.

Author:: Samson Timoner

SUPPORT: This work was supported by the Functional Imaging Research in Schizophrenia Testbed (FIRST) Biomedical Informatics Research Network (BIRN, http://www.birncommunity.org/), which is funded by the National Center for Research Resources at the National Institutes of Health (NIH). This work is also funded by the Neuroimage Analysis Center (P41 RR13218).

Definition at line 72 of file itkKullbackLeiblerCompareHistogramImageToImageMetric.h.