System, devices and/or processes for processing image pixel values
Example methods, apparatuses, and/or articles of manufacture are disclosed that may be implemented, in whole or in part, techniques to process pixel values sampled from a multi color channel imaging device. In particular, methods and/or techniques to process pixel samples for interpolating pixel values for one or more color channels.
1 . A method comprising:
determining coefficients for implementing a kernel to be applied to pixel values derived, at least in part, from pixel signals of an imaging array to interpolate pixel values for a multi color channel imaging system, the coefficients to being determined based at least in part, on training operations, wherein:
the kernel is partitionable into a plurality of sub arrays, each sub array to map pixel values to pixel values of a color channel associated with the sub array such that:
a sum of coefficients of a first sub array of the plurality of sub arrays to map pixel values of a particular color channel substantially equals a first predetermined value; and
a sum of coefficients of at least one other sub array of the plurality of sub arrays to map pixel values of at least one remaining color channel other than the particular color channel substantially equals a second predetermined value.
2 . The method of claim 1 , wherein the coefficients are determined based, at least in part, on a back propagation operation to adjust one or more weights associated with nodes in a neural network according to a loss function.
3 . The method of claim 1 , wherein:
the
the first predetermined value comprises a unit value; and
the second predetermined value comprises a null value.
4 . The method of claim 1 , and further comprising:
mapping input pixel signals of the imaging array to one or more arrays of pixel signal values spatially set off from positions of the input pixel signals;
applying the kernel to the input pixel signals to provide a first component of demosaiced pixel signal values for the particular color channel;
applying the kernel to at least one of the one or more arrays of pixel signal values that are spatially set off from positions of the input pixel signals to provide at least a second component of demosaiced pixel signal values for the particular color channel; and
combining the first and second components of the demosaiced pixel signal values for the particular color channel.
5 . The method of claim 1 , wherein the coefficients are determined based at least in part, on machine learning operations applied to a neural network (NN), and wherein the NN comprises a convolutional neural network (CNN) comprising:
one or more filter paths to detect and/or classify image features other than pixel values for color channels; and
wherein coefficients of the kernel are further determined based, at least in part, on a back propagation operation to adjust one or more weights associated with one or more nodes of the CNN in at least one of the one or more filter paths according to a loss function.
6 . The method of claim 5 , wherein the kernel is adapted to map input pixel values at a first resolution to output pixel values at a second resolution higher than the first resolution.
7 . The method of claim 1 , wherein:
determining the coefficients for implementing the kernel comprises determining coefficients to be applied to pixel values at two or more filter stages of stages of a filtering network.
8 . The method of claim 7 , wherein
the first predetermined value comprises a unit value; and
the second predetermined value comprises a null value.
9 . A computing device, comprising:
one or more memory devices; and
one or more processors, the one or more processors to be coupled to the one or more memory devices, to:
determine coefficients to implement a kernel to be applied to pixel values derived, at least in part, from pixel signals of an imaging array for interpolation of pixel values for a multi color channel imaging system, the coefficients to be determined based at least in part, on training operations, wherein:
the kernel to be partitionable into a plurality of sub arrays, each sub array to map pixel values to pixel values of a color channel associated with the sub array such that:
a sum of coefficients of a sub array to map pixel values of a particular color channel substantially to equal a first predetermined value; and
a sum of coefficients of at least one of one or more sub arrays to map pixel values of at least one remaining color channel other than the particular color channel substantially to equal a second predetermined value.
10 . The computing device of claim 9 , the coefficients to be determined based, at least in part, on a back propagation operation to adjust one or more weights associated with nodes in a neural network according to a loss function.
11 . The computing device of claim 9 , wherein:
the
first predetermined value comprises a unit value; and
the second predetermined value comprises a null value.
12 . The computing device of claim 9 , wherein the one or more processors are further to:
map input pixel signals of the imaging array to one or more arrays of pixel signal values spatially set off from positions of the input pixel signals;
apply the kernel to the input pixel signals to provide a first component of demosaiced pixel signal values for the particular color channel;
apply the kernel to at least one of the one or more arrays of pixel signal values that are spatially set off from positions of the input pixel signals to provide at least a second component of demosaiced pixel signal values for the particular color channel; and
combine the first and second components of the demosaiced pixel signal values for the particular color channel.
13 . The computing device of claim 9 , wherein the coefficients to be determined based at least in part, on machine learning operations applied to a neural network (NN), and wherein the NN to comprise a convolutional neural network (CNN) to have:
one or more filter paths to detect and/or classify image features other than pixel values for color channels; and
wherein coefficients of the kernel are further to be determined based, at least in part, on a back propagation operation to adjust one or more weights associated with one or more nodes of the CNN in at least one of the one or more filter paths according to a loss function.
14 . The computing device of claim 13 , wherein the kernel to be adapted to map input pixel values at a first resolution to output pixel values at a second resolution higher than the first resolution.
15 . The computing device of claim 9 , wherein the coefficients for implementing the kernel to be applicable to pixel values at two or more filter stages of stages of a filtering network.
16 . The computing device of claim 15 , wherein
the first predetermined value comprises a unit value; and
the second predetermined value comprises a null value.
17 . A method comprising:
applying coefficients of a kernel to pixel values derived, at least in part, from pixel signals of an imaging array for interpolation of pixel values for a multi color channel imaging system, the coefficients being determined based, at least in part, on machine learning operations applied to a neural network (NN), wherein:
the kernel to is partitionable into a plurality of sub arrays, each sub array to map pixel values to pixel values of a color channel associated with the sub array such that:
a sum of coefficients of a sub array to map pixel values of a particular color channel substantially to equal a first predetermined value; and
a sum of coefficients of at least one of one or more sub arrays to map pixel values of at least one remaining color channel other than the particular color channel substantially to equal a second predetermined value.
18 . The method of claim 17 , wherein:
the
first predetermined value comprises a unit value; and
the second predetermined value comprises a null value.
19 . The method of claim 17 , wherein the coefficients are determined based at least in part, on machine learning operations applied to a neural network (NN), and wherein the NN to comprise a convolutional neural network (CNN) to have:
one or more filter paths to detect and/or classify image features other than pixel values for color channels; and
wherein coefficients of the kernel are further to be determined based, at least in part, on a back propagation operation to adjust one or more weights associated with one or more nodes of the CNN in at least one of the one or more filter paths according to a loss function.
20 . The method of claim 17 , wherein the coefficients are determined based at least in part, on machine learning operations applied to a neural network (NN), and wherein the NN to comprise a convolutional neural network (CNN) to have:
one or more filter paths to detect and/or classify image features other than pixel values for color channels; and
wherein coefficients of the kernel are further to be determined based, at least in part, on a back propagation operation to adjust one or more weights associated with one or more nodes of the CNN in at least one of the one or more filter paths according to a loss function.