Convolutions.H File Reference

---

Detailed Description

basic 1-D and 2-D filtering operations

Definition in file Convolutions.H.

#include "Image/Image.H"
#include "Util/Promotions.H"

Include dependency graph for Convolutions.H:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Enumerations
enum	ConvolutionBoundaryStrategy { CONV_BOUNDARY_ZERO, CONV_BOUNDARY_CLEAN, CONV_BOUNDARY_REPLICATE }
Functions
template<class T_or_RGB>
Image< typename promote_trait< T_or_RGB, float >::TP >	convolve (const Image< T_or_RGB > &src, const float *filter, const int Nx, const int Ny, ConvolutionBoundaryStrategy boundary)
	Brute force, super inefficient 2D convolution, with C filter of Nx*Ny.
template<class T_or_RGB>
Image< typename promote_trait< T_or_RGB, float >::TP >	convolve (const Image< T_or_RGB > &src, const Image< float > &filter, ConvolutionBoundaryStrategy boundary)
	Brute force, super inefficient 2D convolution.
template<class T>
Image< typename promote_trait< T, float >::TP >	optConvolve (const Image< T > &src, const Image< float > &filter)
	This is a somewhat-optimized 2-D convolution.
template<class T>
Image< typename promote_trait< T, float >::TP >	convolveHmax (const Image< T > &src, const Image< float > &filter)
	Brute force, super inefficient 2D convolution (truncated filter boundary).
template<class T_or_RGB>
Image< typename promote_trait< T_or_RGB, float >::TP >	sepFilter (const Image< T_or_RGB > &src, const Image< float > &hFilter, const Image< float > &vFilter, ConvolutionBoundaryStrategy boundary)
	Separable filter, any size (size=0 for no filter).
template<class T_or_RGB>
Image< typename promote_trait< T_or_RGB, float >::TP >	sepFilter (const Image< T_or_RGB > &src, const float *hFilter, const float *vFilter, const int hfsize, const int vfsize, ConvolutionBoundaryStrategy boundary)

---

Enumeration Type Documentation

enum ConvolutionBoundaryStrategy

Enumerator: CONV_BOUNDARY_ZERO  zero-pad, a.k.a. truncated CONV_BOUNDARY_CLEAN  normalize by the sum of the used filter coefficients CONV_BOUNDARY_REPLICATE  replicate the nearest image pixel value Definition at line 47 of file Convolutions.H.

---

Function Documentation

template<class T_or_RGB> Image<typename promote_trait<T_or_RGB, float>::TP> convolve (  const Image< T_or_RGB > &  src, const Image< float > &  filter, ConvolutionBoundaryStrategy  boundary )  [inline]

Brute force, super inefficient 2D convolution. Definition at line 64 of file Convolutions.H. References convolve(), Image< T >::getArrayPtr(), Image< T >::getHeight(), and Image< T >::getWidth().

template<class T_or_RGB> Image<typename promote_trait<T_or_RGB, float>::TP> convolve (  const Image< T_or_RGB > &  src, const float *  filter, const int  Nx, const int  Ny, ConvolutionBoundaryStrategy  boundary )

Brute force, super inefficient 2D convolution, with C filter of Nx*Ny.

template<class T> Image<typename promote_trait<T, float>::TP> convolveHmax (  const Image< T > &  src, const Image< float > &  filter )

Brute force, super inefficient 2D convolution (truncated filter boundary). With normalization by the local image energy, a la HMAX

template<class T> Image<typename promote_trait<T, float>::TP> optConvolve (  const Image< T > &  src, const Image< float > &  filter )

This is a somewhat-optimized 2-D convolution. It is optimized in the sense that there are special-case inner loops for (1) cases where the filter doesn't completely overlap with the local image patch because we're at the image boundary (in this case we need a specialized inner loop to handle these boundary conditions), and (2) cases where the filter and local image patch overlap completely (in which case we can use a more efficient inner loop).

template<class T_or_RGB> Image<typename promote_trait<T_or_RGB, float>::TP> sepFilter (  const Image< T_or_RGB > &  src, const Image< float > &  hFilter, const Image< float > &  vFilter, ConvolutionBoundaryStrategy  boundary )

Separable filter, any size (size=0 for no filter). If you want to do horizontal-only or vertical-only filtering, then simply pass an empty filter (i.e., Image<float>()) for the other filter. Parameters: boundary  controls the boundary-handling strategy (i.e., zero-fill, clean, replicate) HISTORICAL NOTE: This function takes over for what used to be many functions, including sepFilter(), sepFiltClean(), xFilter(), yFilter(), xFilterClean(), and yFilterClean(). First, the x+y variants are now collapsed into a single function; if you want x-only or y-only filtering, simply pass an empty image for the other filter. Second, the functions are collapsed across boundary handling strategy, where you can now select the boundary strategy at run-time with the ConvolutionBoundaryStrategy parameter.

---