convolve (Separable)

Separable Convolution. More...

Functions
AFAPI array	convolve (const array &col_filter, const array &row_filter, const array &signal, const convMode mode=AF_CONV_DEFAULT)
	C++ Interface for separable convolution on two dimensional signals. More...
AFAPI af_err	af_convolve2_sep (af_array *out, const af_array col_filter, const af_array row_filter, const af_array signal, const af_conv_mode mode)
	C Interface for separable convolution on two dimensional signals. More...

Detailed Description

Separable Convolution.

Separable Convolution is faster equivalent of the canonical 2D convolution with an additional prerequisite that the filter/kernel can be decomposed into two separate spatial vectors. A classic example of such separable kernels is sobel operator. Given below is decomposition of vertical gradient of sobel operator.

\( \begin{bmatrix} -1 & 0 & +1 \\ -2 & 0 & +2 \\ -1 & 0 & +1 \\ \end{bmatrix} \)

can be decomposed into two vectors shown below.

\( \begin{bmatrix} 1 \\ 2 \\ 1 \\ \end{bmatrix} \)

\( \begin{bmatrix} -1 & 0 & +1 \\ \end{bmatrix} \)

Function Documentation

af_convolve2_sep()

AFAPI af_err af_convolve2_sep	(	af_array *	out,
		const af_array	col_filter,
		const af_array	row_filter,
		const af_array	signal,
		const af_conv_mode	mode
	)

C Interface for separable convolution on two dimensional signals.

Parameters

[out]	out	is convolved array
[in]	col_filter	is filter that has to be applied along the coloumns
[in]	row_filter	is filter that has to be applied along the rows
[in]	signal	is the input array
[in]	mode	indicates if the convolution should be expanded or not(where output size equals input)

Returns

AF_SUCCESS if the convolution is successful, otherwise an appropriate error code is returned.

Note

Separable convolution only supports two(ONE-to-ONE and MANY-to-ONE) batch modes from the ones described in the detailed description section.

convolve()

AFAPI array convolve	(	const array &	col_filter,
		const array &	row_filter,
		const array &	signal,
		const convMode	mode = AF_CONV_DEFAULT
	)

C++ Interface for separable convolution on two dimensional signals.

    // vector<dim4> numDims;
    // vector<vector<float> > in;
    array signal(numDims[0], &(in[0].front()));
    // signal dims = [3 4 2 1]
    array cFilter(numDims[1], &(in[1].front()));
    // coloumn filter dims = [2 1 1 1]
    array rFilter(numDims[2], &(in[2].front()));
    // row filter dims = [3 1 1 1]
 
    array output = convolve(cFilter, rFilter, signal, AF_CONV_DEFAULT);
    // output signal dims = [3 4 2 1] - same as input since 'expand = false'
    // notice that the input signal is 3d array, therefore
    // batch mode will be automatically activated.
    // output will be 3d array with result of each 2d array convolution(with
    // same filter) stacked along the 3rd dimension

Parameters

[in]	signal	is the input signal
[in]	col_filter	is the signal that shall be along coloumns
[in]	row_filter	is the signal that shall be along rows
[in]	mode	indicates if the convolution should be expanded or not(where output size equals input)

Returns

the convolved array

Note

The default parameter of domain, AF_CONV_AUTO, heuristically switches between frequency and spatial domain.

Separable convolution only supports two(ONE-to-ONE and MANY-to-ONE) batch modes from the ones described in the detailed description section.