Computes a scalar, gradient magnitude image from a multiple channel (pixels are vectors) input. More...

#include <itkVectorGradientMagnitudeImageFilter.h>

Inheritance diagram for itk::VectorGradientMagnitudeImageFilter< TInputImage, TRealType, TOutputImage >:

Collaboration diagram for itk::VectorGradientMagnitudeImageFilter< TInputImage, TRealType, TOutputImage >:

Public Types
typedef ConstNeighborhoodIterator < RealVectorImageType >	ConstNeighborhoodIteratorType
typedef SmartPointer< const Self >	ConstPointer
typedef InputImageType::Pointer	InputImagePointer
typedef TInputImage	InputImageType
typedef TInputImage::PixelType	InputPixelType
typedef OutputImageType::Pointer	OutputImagePointer
typedef Superclass::OutputImageRegionType	OutputImageRegionType
typedef TOutputImage	OutputImageType
typedef TOutputImage::PixelType	OutputPixelType
typedef SmartPointer< Self >	Pointer
typedef ConstNeighborhoodIteratorType::RadiusType	RadiusType
typedef TRealType	RealType
typedef Image< RealVectorType,::itk::GetImageDimension < TInputImage > ::ImageDimension >	RealVectorImageType
typedef Vector< TRealType,::itk::GetVectorDimension < InputPixelType > ::VectorDimension >	RealVectorType
typedef VectorGradientMagnitudeImageFilter	Self
typedef ImageToImageFilter < TInputImage, TOutputImage >	Superclass
typedef FixedArray< TRealType, VectorDimension >	WeightsType
Public Member Functions
virtual ::itk::LightObject::Pointer	CreateAnother (void) const
virtual void	GenerateInputRequestedRegion () throw ( InvalidRequestedRegionError )
virtual const char *	GetNameOfClass () const
virtual bool	GetUseImageSpacing () const
void	SetUseImageSpacing (bool)
void	SetUseImageSpacingOff ()
void	SetUseImageSpacingOn ()
void	SetUsePrincipleComponentsOff ()
	typedef (Concept::HasNumericTraits< RealType >) RealTypeHasNumericTraitsCheck
	typedef (Concept::HasNumericTraits< typename InputPixelType::ValueType >) InputHasNumericTraitsCheck

virtual void	SetDerivativeWeights (WeightsType _arg)
virtual const WeightsType &	GetDerivativeWeights ()

virtual void	SetComponentWeights (WeightsType _arg)
virtual const WeightsType &	GetComponentWeights ()

virtual const RadiusType &	GetNeighborhoodRadius ()
virtual void	SetNeighborhoodRadius (RadiusType _arg)

virtual void	SetUsePrincipleComponents (bool _arg)
virtual bool	GetUsePrincipleComponents () const
void	SetUsePrincipleComponentsOn ()
Static Public Member Functions
static int	CubicSolver (double , double )
static Pointer	New ()
Static Public Attributes
static const unsigned int	ImageDimension = TOutputImage::ImageDimension
static const unsigned int	VectorDimension = InputPixelType::Dimension
Protected Types
typedef InputImageType::Superclass	ImageBaseType
Protected Member Functions
void	BeforeThreadedGenerateData ()
TRealType	EvaluateAtNeighborhood (const ConstNeighborhoodIteratorType &it) const
TRealType	EvaluateAtNeighborhood3D (const ConstNeighborhoodIteratorType &it) const
virtual const ImageBaseType *	GetRealValuedInputImage ()
TRealType	NonPCEvaluateAtNeighborhood (const ConstNeighborhoodIteratorType &it) const
void	PrintSelf (std::ostream &os, Indent indent) const
void	ThreadedGenerateData (const OutputImageRegionType &outputRegionForThread, ThreadIdType threadId)
	VectorGradientMagnitudeImageFilter ()
virtual	~VectorGradientMagnitudeImageFilter ()
Protected Attributes
WeightsType	m_ComponentWeights
WeightsType	m_DerivativeWeights
WeightsType	m_SqrtComponentWeights
Private Member Functions
void	operator= (const Self &)
	VectorGradientMagnitudeImageFilter (const Self &)
Private Attributes
RadiusType	m_NeighborhoodRadius
ImageBaseType::ConstPointer	m_RealValuedInputImage
ThreadIdType	m_RequestedNumberOfThreads
bool	m_UseImageSpacing
bool	m_UsePrincipleComponents

Detailed Description

template<typename TInputImage, typename TRealType = float, typename TOutputImage = Image< TRealType, ::itk::GetImageDimension< TInputImage >::ImageDimension >>
class itk::VectorGradientMagnitudeImageFilter< TInputImage, TRealType, TOutputImage >

Computes a scalar, gradient magnitude image from a multiple channel (pixels are vectors) input.

Overview: This filter has two calculation modes. The first (default) mode calculates gradient magnitude as the difference between the largest two eigenvalues in a principle component analysis of the partial derivatives [1]. The gradient is then based on the direction of maximal change, and is a characterization of how "elongated" the point-spread of the analysis is found to be.

The second, more heuristic, calculation mode finds gradient magnitude as the square-root of the sum of the * individual weighted vector component derivative sums squared. That is, * $\mathbf{magnitude} = \left( \sum_{i=0}^n \sum_{j=0}^m \frac{\delta * \phi_j}{\delta \mathbf{x}_{i}}^2 \right)^{\frac{1}{2}}$ , where $\phi_j$ * is the $j^{\mathbf{th}}$ channel of vector image $\phi$ of dimension $n$ . * Weighting terms are applied to each vector component.

The second mode is computationally much faster than the first and has the advantage that it is automatically multi-threaded (some vnl functions used in the first mode are not thread-safe). The first mode, however, tends to give intuitively better results with less (or no) parameter tuning.

Template Parameters (Input and Output): This filter has one required template parameter which defines the input image type. The pixel type of the input image is assumed to be a vector (e.g., itk::Vector, itk::RGBPixel, itk::FixedArray). The scalar type of the vector components must be castable to floating point. Instantiating with an image of RGBPixel<unsigned short>, for example, is allowed, but the filter will convert it to an image of Vector<float,3> for processing.

The second template parameter, TRealType, can be optionally specified to define the scalar numerical type used in calculations. This is the component type of the output image, which will be of itk::Vector<TRealType, N>, where N is the number of channels in the multiple component input image. The default type of TRealType is float. For extra precision, you may safely change this parameter to double.

The third template parameter is the output image type. The third parameter will be automatically constructed from the first and second parameters, so it is not necessary (or advisable) to set this parameter explicitly. Given an M-channel input image with dimensionality N, and a numerical type specified as TRealType, the output image will be of type itk::Image<itk::Vector<TRealType, M>, N>.

Filter Parameters: The methods Set/GetUsePrincipleComponents and SetUsePrincipleComponentsOn/Off determine controls the calculation mode that is used.

The method SetUseImageSpacingOn will cause derivatives in the image to be scaled (inversely) with the pixel size of the input image, effectively taking derivatives in world coordinates (versus isotropic image space). SetUseImageSpacingOff turns this functionality off. Default is UseImageSpacingOn. The parameter UseImageSpacing can be set directly with the method SetUseImageSpacing(bool).

Weights can be applied to the derivatives directly using the SetDerivativeWeights method. Note that if UseImageSpacing is set to TRUE (ON), then these weights will be overridden by weights derived from the image spacing when the filter is updated. The argument to this method is a C array of TRealValue type.

Weights can be applied to each vector component of the image when the component derivative values are summed during computation. Specify these weights using the SetComponentWeights method. The argument to this method is a C array of TRealValue type.

Constraints: The filter requires an image with at least two dimensions and a vector length of at least 2. The theory supports extension to scalar images, but the implementation of the itk vector classes do not

The template parameter TRealType must be floating point (float or double) or a user-defined "real" numerical type with arithmetic operations defined sufficient to compute derivatives.

Performance: This filter will automatically multithread if run with SetUsePrincipleComponents=Off or on 3D data in UsePrincipleComponents=On mode. Unfortunately the ND eigen solver used is not thread safe (a special 3D solver is), so it cannot multithread for data other than 3D in UsePrincipleComponents=On mode.

References

[1] G. Sapiro and D. Ringach, "Anisotropic Diffusion of Multivalued Images with Application to Color Filtering," IEEE Transactions on Image Processing, Vol 5, No. 11 pp. 1582-1586, 1996

See also:: Image; Neighborhood; NeighborhoodOperator; NeighborhoodIterator

Definition at line 139 of file itkVectorGradientMagnitudeImageFilter.h.