Kernels for manipulating images. More...

Functions
kernel void	separateRGBChannels_Float2Float (global float AoS, global float r, global float g, global float b, local float *data)
	Separates the 3 channels of an RGB image. More...

kernel void	separateRGBChannels_Uchar2Float (global uchar AoS, global float r, global float g, global float b, local uchar *data)
	Separates the 3 channels of an RGB image. More...

kernel void	combineRGBChannels_Float2Float (global float r, global float g, global float b, global float AoS, local float *data)
	Combines the 3 channels of an RGB Image. More...

kernel void	combineRGBChannels_Float2Uchar (global float r, global float g, global float b, global uchar AoS, local float *data)
	Combines the 3 channels of an RGB Image. More...

kernel void	depth_Ushort2Float (global ushort4 depth, global float4 fDepth, float scaling)
	Converts a buffer from type `uchort` to `float`. More...

kernel void	depthTo3D (global float depth, global float4 pCloud, float f, float scaling)
	Transforms a depth image to a point cloud. More...

kernel void	rgbNorm (global float in, global float out)
	Performs RGB color normalization. More...

kernel void	rgbdTo8D (global float depth, global float r, global float g, global float b, global float8 p8D, local float data, uint cols, float f, float scaling, int rgb_norm)
	Fuses 3-D space coordinates and RGB color values into 8-D feature points (homogeneous coordinates + RGBA values). More...

kernel void	splitPC8D (global float8 pc8d, global float4 pc4d, global float4 *rgba, unsigned int offset)
	Splits an 8-D point cloud into 4-D geometry points and RGBA color points. More...

Detailed Description

Kernels for manipulating images.

Author: Nick Lamprianidis

Version: 1.1

Date: 2015

Copyright: The MIT License (MIT)

: Copyright (c) 2015 Nick Lamprianidis

: Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

: THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Function Documentation

kernel void combineRGBChannels_Float2Float	(	global float *	r,
		global float *	g,
		global float *	b,
		global float *	AoS,
		local float *	data
	)

Combines the 3 channels of an RGB Image.

Performs a matrix transposition on an RGB image (SoA -> AoS). For avoiding alignment restrictions, the SoA structure is broken out to the individual channels, R, G, B.

Note: The global workspace should be one-dimensional (= # pixels in the input buffer). The global and local workspaces should be multiples of 3.

Parameters

[in]	r	input buffer with all the pixel values in channel R.
[in]	g	input buffer with all the pixel values in channel G.
[in]	b	input buffer with all the pixel values in channel B.
[out]	AoS	output buffer with the following (logical) arrangement: float[total-pixels][3]. Each row contains the RGB values of a pixel.
[in]	data	local buffer with size `3 x (# work-items in work-group) x sizeof (float)` bytes.

kernel void combineRGBChannels_Float2Uchar	(	global float *	r,
		global float *	g,
		global float *	b,
		global uchar *	AoS,
		local float *	data
	)

Combines the 3 channels of an RGB Image.

Performs a matrix transposition on an RGB image (SoA -> AoS), demotes the float type to uchar, and scales the data to 255. For avoiding alignment restrictions, the SoA structure is broken out to the individual channels, R, G, B.

Note: The global workspace should be one-dimensional (= # pixels in the input buffer). The global and local workspaces should be multiples of 3.

Parameters

[in]	r	input buffer with all the pixel values in channel R.
[in]	g	input buffer with all the pixel values in channel G.
[in]	b	input buffer with all the pixel values in channel B.
[out]	AoS	output buffer with the following (logical) arrangement: uchar[total-pixels][3]. Each row contains the RGB values of a pixel.
[in]	data	local buffer with size `3 x (# work-items in work-group) x sizeof (float)` bytes.

kernel void depth_Ushort2Float	(	global ushort4 *	depth,
		global float4 *	fDepth,
		float	scaling
	)

Converts a buffer from type uchort to float.

Note: The global workspace should be one dimensional and equal to the number of elements in the image divided by 4.

Parameters

[in]	depth	depth image (for Kinect, type: uint16, unit: mm).
[out]	fDepth	depth image with type `float`.
[in]	scaling	factor by which to scale the depth values in the output array.

kernel void depthTo3D	(	global float *	depth,
		global float4 *	pCloud,
		float	f,
		float	scaling
	)

Transforms a depth image to a point cloud.

Note: The global workspace should be equal to the dimensions of the image.

Parameters

[in]	depth	depth image.
[out]	pCloud	point cloud (in 4D homogeneous coordinates with $ w=1 $).
[in]	f	focal length (for Kinect: 595.f).
[in]	scaling	factor by which to scale the depth values before building the point cloud.

kernel void rgbdTo8D	(	global float *	depth,
		global float *	r,
		global float *	g,
		global float *	b,
		global float8 *	p8D,
		local float *	data,
		uint	cols,
		float	f,
		float	scaling,
		int	rgb_norm
	)

Fuses 3-D space coordinates and RGB color values into 8-D feature points (homogeneous coordinates + RGBA values).

Gathers the R, G, B channel values and assembles them into a pixel with opacity set to 1.0. It can optionally perform RGB normalization. Transforms the depth values into 4-D homogeneous coordinates, setting $ w=1 $.

Note: The global workspace should be one-dimensional (= # elements in the input buffers). Both global and local workspaces should be multiples of 3.

Parameters

[in]	depth	depth array.
[in]	r	channel R array.
[in]	g	channel G array.
[in]	b	channel B array.
[out]	p8D	array of 8D elements (homogeneous coordinates + RGBA values).
[in]	data	local buffer with size `3 x (# work-items in work-group) x sizeof (float)` bytes.
[in]	cols	number of columns (width) in the input images.
[in]	f	focal length (for Kinect: 595.f).
[in]	scaling	factor by which to scale the depth values before building the point cloud.
[in]	rgb_norm	flag to indicate whether to perform RGB normalization.

kernel void rgbNorm	(	global float *	in,
		global float *	out
	)

Performs RGB color normalization.

That is $$ \hat{p}.i = \frac{p.i}{p.r + p.g + p.b},\ \ i=\{r,g,b\} $$

Note: The global workspace should be one-dimensional (= # pixels in the input buffer).

Parameters

[in]	in	original frame.
[out]	out	processed frame.

kernel void separateRGBChannels_Float2Float	(	global float *	AoS,
		global float *	r,
		global float *	g,
		global float *	b,
		local float *	data
	)

Separates the 3 channels of an RGB image.

Performs a matrix transposition on an RGB image (AoS -> SoA). For avoiding alignment restrictions, the SoA structure is broken out to the individual channels, R, G, B.

Note: The global workspace should be one-dimensional (= # pixels in the input buffer). The global and local workspaces should be multiples of 3.

Parameters

[in]	AoS	input buffer with the following (logical) arrangement: float[total-pixels][3]. Each row contains the RGB values of a pixel.
[out]	r	output buffer with all the pixel values in the first channel, R.
[out]	g	output buffer with all the pixel values in the second channel, G.
[out]	b	output buffer with all the pixel values in the third channel, B.
[in]	data	local buffer with size `3 x (# work-items in work-group) x sizeof (float)` bytes.

kernel void separateRGBChannels_Uchar2Float	(	global uchar *	AoS,
		global float *	r,
		global float *	g,
		global float *	b,
		local uchar *	data
	)

Separates the 3 channels of an RGB image.

Performs a matrix transposition on an RGB image (AoS -> SoA), and promotes the uchar type to float while normalizing the values to one. For avoiding alignment restrictions, the SoA structure is broken out to the individual channels, R, G, B.

Note: The global workspace should be one-dimensional (= # pixels in the input buffer). The global and local workspaces should be multiples of 3.

Parameters

[in]	AoS	input buffer with the following (logical) arrangement: uchar[total-pixels][3]. Each row contains the RGB values of a pixel.
[out]	r	output buffer with all the pixel values in the first channel, R.
[out]	g	output buffer with all the pixel values in the second channel, G.
[out]	b	output buffer with all the pixel values in the third channel, B.
[in]	data	local buffer with size `3 x (# work-items in work-group) x sizeof (uchar)` bytes.

kernel void splitPC8D	(	global float8 *	pc8d,
		global float4 *	pc4d,
		global float4 *	rgba,
		unsigned int	offset
	)

Splits an 8-D point cloud into 4-D geometry points and RGBA color points.

Note: The global workspace should be one-dimensional. The x dimension of the global workspace, $ gXdim $, should be equal to the number of points in the point cloud. The local workspace is irrelevant.

Parameters

[in]	pc8d	array with 8-D points (homogeneous coordinates + RGBA values).
[out]	pc4d	array with 4-D geometry points.
[out]	rgba	array with 4-D color points.
[in]	offset	number of points to skip in the output arrays. The kernel will write in the output arrays starting at position `offset`.

Functions

Detailed Description

Function Documentation