Predictive color image processing
Apparatuses, systems, and techniques to process image data. In at least one embodiment, a neural network is trained to perform demosaicing of two-dimensional image data obtained from an image sensor.
1. A processor comprising: one or more circuits to;
cause one or more neural networks to generate a plurality of missing color values that are missing from a first portion of an image and to remove artifacts from the plurality of missing color values to generate a refined set of missing color values; and
add the refined set of missing color values to a plurality of known color values of the first portion of the image to produce a final image.
2. The processor of claim 1 , wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values.
3. The processor of claim 1 , wherein:
the final image is an RGB color image.
4. The processor of claim 1 , wherein the first portion of the image is obtained using a two dimensional image sensor coupled to a Bayer filter.
5. The processor of claim 1 , wherein the image is a mosaic image.
6. The processor of claim 1 , wherein the plurality of missing color values is generated as eight missing color values in a 2×2 quad based on a mosaic image that includes a 5×5 array of incomplete quads around the 2×2 quad.
7. The processor of claim 6 , wherein the 2×2 quad includes two green color values, one red color value, and one blue color value.
8. The processor of claim 1 , wherein the one or more neural networks are trained by at least:
applying a filter to a training image to produce a filtered image;
calculating color values for the filtered image;
calculating a loss based on differences between the color values and the training image; and
adjusting the one or more neural networks based at least in part on the loss.
9. The processor of claim 1 , wherein the one or more neural networks includes a first neural network trained to generate the plurality of missing color values based at least on the first portion of the image and a second portion of the image and a second neural network trained to refine the missing color values based at least on the first portion of the image and the second portion of the image.
10. The processor of claim 1 , wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values based at least on the first portion of the image and another portion of the image.
11. A computer-implemented method, comprising:
using one or more neural networks to generate a plurality of missing color values that are missing from a first portion of an image and to remove artifacts from the plurality of missing color values to generate a refined set of missing color values; and
adding the refined set of missing color values to a plurality of known color values of the first portion of the image to produce a final image.
12. The computer-implemented method of claim 11 ,
wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values.
13. The computer-implemented method of claim 11 , wherein:
the final image is an RGB color image.
14. The computer-implemented method of claim 11 ,
wherein the first portion of the image is obtained using a two dimensional image sensor coupled to a Bayer filter.
15. The computer-implemented method of claim 11 ,
wherein the image is a mosaic image.
16. The computer-implemented method of claim 11 , wherein the plurality of missing color values is generated as eight missing color values in a 2×2 quad based on a mosaic image that includes a 5×5 array of incomplete quads around the 2×2 quad.
17. The computer-implemented method of claim 16 , wherein the 2×2 quad includes two green color values, one red color value, and one blue color value.
18. The computer-implemented method of claim 11 , wherein the one or more neural networks are trained by at least:
applying a filter to a training image to produce a filtered image;
calculating color values for the filtered image;
calculating a loss based on differences between the color values and the training image; and
adjusting the one or more neural networks based at least in part on the loss.
19. The computer-implemented method of claim 11 ,
wherein the one or more neural networks includes a first neural network trained to generate the plurality of missing color values based at least on the first portion of the image and a second portion of the image and a second neural network trained to refine the missing color values based at least on the first portion of the image and the second portion of the image.
20. The computer-implemented method of claim 11 ,
wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values based at least on the first portion of the image and another portion of the image.
21. A computer system comprising one or more processors and memory storing executable instructions that, as a result of being executed by the one or more processors, cause the computer system to:
use one or more neural networks to generate a plurality of missing color values that are missing from a first portion of an image and to remove artifacts from the plurality of missing color values to generate a refined set of missing color values; and
add the refined set of missing color values to a plurality of known color values of the first portion of the image to produce a final image.
22. The computer system of claim 21 , wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values.
23. The computer system of claim 21 , wherein:
the final image is an RGB color image.
24. The computer system of claim 21 , wherein the first portion of the image is obtained using a two dimensional image sensor coupled to a Bayer filter.
25. The computer system of claim 21 , wherein the image is a mosaic image.
26. The computer system of claim 21 , wherein the plurality of missing color values is generated as eight missing color values in a 2×2 quad based on a mosaic image that includes a 5×5 array of incomplete quads around the 2×2 quad.
27. The computer system of claim 26 , wherein the 2×2 quad includes two green color values, one red color value, and one blue color value.
28. The computer system of claim 21 , wherein the one or more neural networks are trained by at least:
applying a filter to a training image to produce a filtered image;
calculating color values for the filtered image;
calculating a loss based on differences between the color values and the training image; and
adjusting the one or more neural networks based at least in part on the loss.
29. The computer system of claim 21 , wherein the one or more neural networks includes a first neural network trained to generate the plurality of missing color values based at least on the first portion of the image and a second portion of the image and a second neural network trained to refine the missing color values based at least on the first portion of the image and the second portion of the image.
30. The computer system of claim 21 , wherein the one or more neural networks includes a first neural network that generates the plurality of missing color values and a second neural network that removes the artifacts from the plurality of missing color values based at least on the first portion of the image and another portion of the image.