Calculate normalized cross correlation using FFTs. More...

#include <itkFFTNormalizedCorrelationImageFilter.h>

Inheritance diagram for itk::FFTNormalizedCorrelationImageFilter< TInputImage, TOutputImage >:

Collaboration diagram for itk::FFTNormalizedCorrelationImageFilter< TInputImage, TOutputImage >:

Public Types
typedef SmartPointer< const Self >	ConstPointer
typedef InputImageType::ConstPointer	InputImageConstPointer
typedef InputImageType::Pointer	InputImagePointer
typedef TInputImage	InputImageType
typedef InputImageType::RegionType	InputRegionType
typedef InputImageType::SizeType	InputSizeType
typedef OutputImageType::Pointer	OutputImagePointer
typedef TOutputImage	OutputImageType
typedef OutputImageType::PixelType	OutputPixelType
typedef SmartPointer< Self >	Pointer
typedef FFTNormalizedCorrelationImageFilter	Self
typedef MaskedFFTNormalizedCorrelationImageFilter < TInputImage, TOutputImage >	Superclass
Public Member Functions
virtual ::itk::LightObject::Pointer	CreateAnother (void) const
virtual const char *	GetNameOfClass () const
Static Public Member Functions
static Pointer	New ()
Static Public Attributes
static const unsigned int	ImageDimension = TOutputImage::ImageDimension
Protected Member Functions
	FFTNormalizedCorrelationImageFilter ()
void	GenerateData ()
void	PrintSelf (std::ostream &os, Indent indent) const
virtual	~FFTNormalizedCorrelationImageFilter ()
Private Member Functions
	FFTNormalizedCorrelationImageFilter (const Self &)
void	operator= (const Self &)

Detailed Description

template<class TInputImage, class TOutputImage>
class itk::FFTNormalizedCorrelationImageFilter< TInputImage, TOutputImage >

Calculate normalized cross correlation using FFTs.

This filter calculates the normalized cross correlation (NCC) of two images using FFTs instead of spatial correlation. It is much faster than spatial correlation for reasonably large structuring elements. This filter is a subclass of the more general MaskedFFTNormalizedCorrelationImageFilter and operates by essentially setting the masks in that algorithm to images of ones. As described in detail in the references below, there is no computational overhead to utilizing the more general masked algorithm because the FFTs of the images of ones are still necessary for the computations.

Inputs: Two images are required as inputs, fixedImage and movingImage. In the context of correlation, inputs are often defined as: "image" and "template". In this filter, the fixedImage plays the role of the image, and the movingImage plays the role of the template. However, this filter is capable of correlating any two images and is not restricted to small movingImages (templates).

Optional parameters: The RequiredNumberOfOverlappingVoxels enables the user to specify how many voxels of the two images must overlap; any location in the correlation map that results from fewer than this number of voxels will be set to zero. Larger values zero-out pixels on a larger border around the correlation image. Thus, larger values remove less stable computations but also limit the capture range. If RequiredNumberOfOverlappingVoxels is set to 0, the default, no zeroing will take place.

Image size: fixedImage and movingImage need not be the same size. Furthermore, whereas some algorithms require that the "template" be smaller than the "image" because of errors in the regions where the two are not fully overlapping, this filter has no such restriction.

Image spacing: Since the computations are done in the pixel domain, all input images must have the same spacing.

Outputs; The output is an image of RealPixelType that is the NCC of the two images and its values range from -1.0 to 1.0. The size of this NCC image is, by definition, size(fixedImage) + size(movingImage) - 1.

Example filter usage:

 typedef itk::FFTNormalizedCorrelationImageFilter< ShortImageType, DoubleImageType > FilterType;
 FilterType::Pointer filter = FilterType::New();
 filter->SetFixedImage( fixedImage );
 filter->SetMovingImage( movingImage );
 filter->SetRequiredNumberOfOverlappingVoxels(20);
 filter->Update();

Warning:: The pixel type of the output image must be of real type (float or double). ConceptChecking is used to enforce the output pixel type. You will get a compilation error if the pixel type of the output image is not float or double.

References: 1) D. Padfield. "Masked object registration in the Fourier domain." Transactions on Image Processing. 2) D. Padfield. "Masked FFT registration". In Proc. Computer Vision and Pattern Recognition, 2010.

Author:: : Dirk Padfield, GE Global Research, padfield@research.ge.com

Definition at line 99 of file itkFFTNormalizedCorrelationImageFilter.h.