Method and apparatus for dynamically optimizing gamma correction for a high dynamic ratio image
An information handling system operating a learning dynamic gamma-correction optimization system may execute a method including identifying a combination of red, green, and blue subpixel component values for a plurality of pixels, and determining optimal red, green, and blue subpixel component values for the plurality of pixels by running a gamma correction algorithm using a gamma correction factor associated with a maximum gradient energy between the plurality of pixels. A training session for a neural network may include associating each of the red, green, and blue subpixel component values with one of a plurality of input signal node values and forward propagating the input signal node values through the neural network to output preliminary output signal node values using initial weighting values, comparing the optimal red, green, and blue subpixel component values with the preliminary output signal node values to determine error signals for each output value, determining an error signal associated with each node of the neural network by back propagating the determined error signals through the neural network, and determining a plurality of corrected weighting variables that would result in output of the optimal red, green, and blue subpixel component values.
1. An information handling system operating a dynamic gamma-correction optimization training system comprising:
a processor, executing code instructions of the dynamic gamma-correction optimization training system, to identify a combination of red, green, and blue subpixel component values for a selected pixel and a plurality of surrounding adjacent pixels around the central, selected pixel that determine a limited pixel window having a preset width and a preset length as a portion a high dynamic ratio (HDR) image;
the processor to perform a neural network training session for a neural network applying the red, green, and blue subpixel component values for the arrangement of the central, selected pixel and surrounding pixels in the limited pixel window as inputs into the neural network and train the neural network to convert input signal node values to preliminary output signal node values using an initial weighting value for each of a plurality of weighting values representing preliminary relatedness of nodes through the neural network;
the processor to determine optimal red, green, and blue subpixel component values for the central, selected pixel and the plurality of surrounding adjacent pixels by running a gamma correction algorithm using a gamma correction factor associated with a maximum gradient energy to maintain variation between intensity of light emitted by the selected pixel and the plurality of surrounding adjacent pixels of the limited pixel window in the HDR image;
the processor to further apply the neural network assessment to:
compare the optimal red, green, and blue subpixel component values with the preliminary output signal node values to determine an error associated with each of the preliminary output node values; and
determine a plurality of corrected weighting variables based on an array of intensity values for the central, selected pixel and plurality of surrounding adjacent pixels arranged around the central, selected pixel for the limited pixel window such that the neural network is trained to forward propagate input signal node values through the plurality of layers of the neural network using the plurality of corrected weighting variables to output of the optimal adjusted red, green, and blue subpixel component values to provide gamma correction for such input signal node values relating to other pixel windows similar to the limited pixel window, wherein the plurality of corrected weighting variables are determined using a gradient-descent method.
2. The information handling system of claim 1 further comprising:
the processor executing code instructions to perform the training session on each possible combination of red, green, and blue subpixel component values for the central, selected pixel and the plurality of surrounding adjacent pixels around the central, selected pixel within the limited pixel window.
3. The information handling system of claim 1 , wherein the preset width is three or more pixels and the preset length is three or more pixels.
4. The information handling system of claim 1 , wherein the initial weighting value is one.
5. The information handling system of claim 1 further comprising:
the processor storing the plurality of corrected weighting variables in a neural network file associated with the identified combination of red, green, and blue subpixel component values for input signal node values for the limited pixel window.
6. The information handling system of claim 5 further comprising:
transmitting the neural network file to a remote information handling system for remote execution of the trained pixel window gamma correction neural network.
7. The information handling system of claim 6 , wherein the neural network file may be used to form a trained neural network to yield optimal red, green, and blue subpixel component values for gamma correction for pixel windows across a plurality of portions of the high dynamic ratio (HDR) image.
8. A method for operating a dynamic gamma-correction optimization training system comprising:
identifying, for a display device, a combination of red, green, and blue subpixel component values for a central, selected pixel and a plurality of surrounding adjacent pixels around the central, selected pixel within a limited pixel window having a preset width and preset length as a portion of a high dynamic ratio (HDR) image;
performing, via a processor executing code instructions of a dynamic gamma-correction optimization training system, a neural network training session by applying the red, green, and blue subpixel component values for the arrangement of the central, selected pixel and surrounding pixels in the limited pixel window as inputs into the neural network and to train the neural network to convert input signal node values to preliminary output signal node values using an initial weighting value for each of a plurality of weighting values representing preliminary relatedness of nodes through the neural network;
determining optimal red, green, and blue subpixel component values for the central, selected pixel and the plurality of surrounding pixels by running a gamma correction algorithm using a gamma correction factor associated with a maximum gradient energy to maintain variation between intensity of light emitted by the central, selected pixel and the plurality of surrounding adjacent pixels arranged in the limited pixel window of the HDR image;
comparing the optimal red, green, and blue subpixel component values with the preliminary output signal node values to determine an error associated with each of the preliminary output node values; and
determining a plurality of corrected weighting variables based on an array of intensity values for the central, selected pixel and plurality of surrounding adjacent pixels around the selected pixel for the limited pixel window for use with a trained neural network, such that forward propagating the input signal node values of a selected pixel and pixel window through the plurality of layers of the neural network using the plurality of corrected weighting variables results in output of the optimal adjusted red, green, and blue subpixel component values to provide gamma correction for the input signal node values relating to other pixel windows similar to the limited pixel window, wherein the plurality of corrected weighting variables are determined using a gradient-descent method.
9. The method of claim 8 further comprising:
performing the training session on each possible combination of red, green, and blue subpixel component values for the central, selected pixel and the plurality of surrounding adjacent pixels around the selected pixel within the limited pixel window.
10. The method of claim 8 , wherein the preset width is three or more pixels and the preset length is three or more pixels.
11. The method of claim 8 , wherein the initial weighting value is one.
12. The method of claim 8 , wherein the preset width is five pixels and the preset length is five pixels.
13. The method of claim 8 , wherein the neural network includes a number of hidden layers equivalent to a number of pixels included in the preset width of the pixel window.
14. The method of claim 8 , wherein the neural network includes three hidden layers.
15. A method for operating a dynamic gamma-correction optimization training system comprising:
identifying, for a display device, a combination of red, green, and blue subpixel component values for a central, selected pixel and a plurality of surrounding adjacent pixels around the selected pixel within a limited pixel window having a preset width and preset length as a portion of a high dynamic ratio (HDR) image;
performing, via a processor executing code instructions of a dynamic gamma-correction optimization training system, a neural network training session by applying the red, green, and blue subpixel component values for the arrangement of the central, selected pixel and surrounding pixels in the limited pixel window as inputs into the neural network and to train the neural network to convert input signal node values to preliminary output signal node values using an initial weighting value for each of a plurality of weighting values representing preliminary relatedness of nodes through the neural network, wherein the initial weighting value is one;
determining optimal red, green, and blue subpixel component values for the central, selected pixel and the plurality of surrounding pixels by running a gamma correction algorithm using a gamma correction factor associated with a maximum gradient energy to maintain variation between intensity of light emitted by the central, selected pixel and the plurality of surrounding adjacent pixels arranged in the limited pixel window of the HDR image;
comparing the optimal red, green, and blue subpixel component values with the preliminary output signal node values to determine an error associated with each of the preliminary output node values;
determining a plurality of corrected weighting variables based on an array of intensity values for the central, selected pixel and plurality of surrounding adjacent pixels around the central, selected pixel for the limited pixel window for use with a trained neural network, such that forward propagating the input signal node values of a selected pixel and pixel window through the plurality of layers of the neural network using the plurality of corrected weighting variables results in output of the optimal adjusted red, green, and blue subpixel component values to provide gamma correction for the input signal node values relating to other pixel windows similar to the limited pixel window; and
determining output of the optimal adjusted red, green, and blue subpixel component values to provide gamma correction via the trained neural network for the input signal node values of a second limited pixel window adjacent to the first limited pixel window arranged with a second central, selected pixel and second plurality of surrounding adjacent pixels around the second central, selected pixel.
16. The method of claim 15 , wherein the plurality of corrected weighting variables are determined using a gradient-descent method.
17. The method of claim 15 , further comprising:
storing the plurality of corrected weighting variables in a neural network file associated with the identified combination of red, green, and blue subpixel component values for input signal node values for the limited pixel window; and
transmitting the neural network file to a remote information handling system for remote execution of the trained pixel window gamma correction neural network.
18. The method of claim 15 , wherein the preset width is three or more pixels and the preset length is three or more pixels.
19. The method of claim 15 , wherein the neural network includes a number of hidden layers equivalent to a number of pixels included in the preset width of the limited pixel window.
20. The method of claim 15 , wherein the neural network includes three hidden layers.