Deformably register two images using the fast curvature algorithm. More...

#include <itkCurvatureRegistrationFilter.h>

Inheritance diagram for itk::CurvatureRegistrationFilter< TFixedImage, TMovingImage, TDisplacementField, TImageForceFunction >:

Collaboration diagram for itk::CurvatureRegistrationFilter< TFixedImage, TMovingImage, TDisplacementField, TImageForceFunction >:

List of all members.

Public Types
typedef SmartPointer< const Self >	ConstPointer
typedef DisplacementFieldComponentImageType::Pointer	DisplacementFieldComponentImagePointer
typedef Image< RealTypeDFT, TDisplacementField::ImageDimension >	DisplacementFieldComponentImageType
typedef DisplacementFieldPixelType::ValueType	DisplacementFieldComponentType
typedef TDisplacementField::PixelType	DisplacementFieldPixelType
typedef Superclass::DisplacementFieldPointer	DisplacementFieldPointer
typedef Superclass::DisplacementFieldType	DisplacementFieldType
typedef Superclass::FiniteDifferenceFunctionType	FiniteDifferenceFunctionType
typedef Superclass::FixedImagePointer	FixedImagePointer
typedef Superclass::FixedImageType	FixedImageType
typedef Superclass::MovingImagePointer	MovingImagePointer
typedef Superclass::MovingImageType	MovingImageType
typedef SmartPointer< Self >	Pointer
typedef double	RealTypeDFT
typedef TImageForceFunction	RegistrationFunctionType
typedef CurvatureRegistrationFilter	Self
typedef PDEDeformableRegistrationFilter < TFixedImage, TMovingImage, TDisplacementField >	Superclass
typedef Superclass::TimeStepType	TimeStepType
Public Member Functions
virtual ::itk::LightObject::Pointer	CreateAnother (void) const
virtual double	GetMetric () const
virtual const char *	GetNameOfClass () const
void	SetConstraintWeight (const float w)
void	SetTimeStep (const TimeStepType ts)
Static Public Member Functions
static Pointer	New ()
Static Public Attributes
static const unsigned int	DeformationVectorDimension = DisplacementFieldPixelType::Dimension
static const unsigned int	ImageDimension = FixedImageType::ImageDimension
Protected Member Functions
virtual void	ApplyUpdate (const TimeStepType &dt)
	CurvatureRegistrationFilter ()
virtual void	Initialize ()
void	PrintSelf (std::ostream &os, Indent indent) const
	~CurvatureRegistrationFilter ()
Private Member Functions
	CurvatureRegistrationFilter (const Self &)
void	operator= (const Self &)
Private Attributes
float	m_ConstraintWeight
RealTypeDFT *	m_DiagonalElements [ImageDimension]
RealTypeDFT *	m_DisplacementFieldComponentImage
RealTypeDFT *	m_DisplacementFieldComponentImageDCT
unsigned int	m_FixedImageDimensions [ImageDimension]
fftw_plan	m_PlanBackwardDCT
fftw_plan	m_PlanForwardDCT
TimeStepType	m_TimeStep

Detailed Description

template<class TFixedImage, class TMovingImage, class TDisplacementField, class TImageForceFunction = MeanSquareRegistrationFunction< TFixedImage, TMovingImage, TDisplacementField >>
class itk::CurvatureRegistrationFilter< TFixedImage, TMovingImage, TDisplacementField, TImageForceFunction >

Deformably register two images using the fast curvature algorithm.

CurvatureRegistrationFilter implements the fast (i.e., O(n log n) ) registration method described in B. Fischer and J. Modersitzki, "A unified approach to fast image registration and a new curvature based registration technique," Linear Algebra and its Applications, vol. 380, pp. 107-124, 2004.

A deformation field is represented as a image whose pixel type is some vector type with at least N elements, where N is the dimension of the fixed image. The vector type must support element access via operator []. It is assumed that the vector elements behave like floating point scalars.

This class is templated over the fixed image type, moving image type, the deformation field type, and the images forces function. Fundamentally, any image force function in the PDE framework should work (e.g., mean squares, demons, correlation, mutual information, etc.).

The input fixed and moving images are set via methods SetFixedImage and SetMovingImage respectively. An initial deformation field maybe set via SetInitialDisplacementField or SetInput. If no initial field is set, a zero field is used as the initial condition.

The output deformation field can be obtained via methods GetOutput or GetDisplacementField.

This class makes use of the finite difference solver hierarchy. Update for each iteration is computed in the function class defined by the template parameter TImageForceFunction.

The algorithm has three parameters: the number of iteration to be performed, the time step, and the weight of the curvature regularization term in the overall cost function.

Warning:: The algorithm parameters (time step and constraint weight) as given by F&M (dt=100; alpha=0.01) do not seem to work with this implementation. In fact, they seem to be off by orders of magnitude. This may be an effect of different normalization of the image forces, or this implementation may be incorrect. Be aware.; This filter assumes that the fixed image type, moving image type and deformation field type all have the same number of dimensions.; There is something sketchy going on with the DCT (see FFTW documentation) regarding numerical stability of the "ordinary" DCT. Also, F&M are slightly ambiguous in their use of the DCT operator. So it is quite possible that the implementation of their algorithm is not quite correct. Ultimately, they are saying that the curvature smoother is equivalent to a special low-pass filter that is applied to the DCT coefficients.

Note:: This class was developed with funding from:

"CNS Deficits: Interaction of Age and Alcoholism" NIAAA AA05965, PI: A. Pfefferbaum

"INIA: Imaging Core" NIAAA AA13521, PI: A. Pfefferbaum

See also:: CurvatureRegistrationFunction

Author:: Torsten Rohlfing, SRI International, Neuroscience Program

Definition at line 100 of file itkCurvatureRegistrationFilter.h.