Convolutional neural network filter stacker for colorization preprocessing

1. A computer-implemented method comprising:

receiving, at a computing device, an image pair, wherein a first image of the image pair comprises a respective initial training image and wherein a second image of the image pair comprises a respective ground truth training image;

providing, by the computing device, instances of the first image to a plurality of image filtering modules, wherein each image filtering module is configured to apply an image filter on an input image and output a filtered representation of the input image;

determining, by the computing device, respective filtered representations of the first image using the plurality of image filtering modules;

indirectly training, by the computing device, an adaptor artificial neural network (ANN) by at least:

applying the adaptor ANN on at least the respective filtered representations of the first image to produce an adapted representation of the first image;

determining, using a trained colorization ANN, a colorized image from the adapted representation; and

computationally updating weights of the adaptor ANN based on a loss function that comprises a difference between the colorized image and the second image of the image pair; and

providing, using the computing device, the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN.

2. The computer-implemented method of claim 1 , wherein the adaptor ANN comprises an input layer that is configured to receive and concatenate the respective filtered representations of the first image.

3. The computer-implemented method of claim 2 , wherein the input layer is further configured to receive an instance of the first image and concatenate the respective filtered representations with the instance of the first image.

4. The computer-implemented method of claim 1 , wherein the filtered representation of the input image comprises a difference between the input image and the input image after having been transformed by the image filter.

5. The computer-implemented method of claim 1 , further comprising:

after indirectly training the adaptor ANN, further training the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN by at least:

receiving a second image pair, wherein a first image of the second image pair comprises a respective second initial training image and wherein a second image of the second image pair comprises a respective second ground truth training image;

providing instances of the first image of the second image pair to the plurality of image filtering modules;

determining respective filtered representations of the first image of the second image pair using the plurality of image filtering modules;

applying the trained adaptor ANN on at least the respective filtered representations of the first image of the second image pair to produce a second adapted representation;

determining, using the trained colorization ANN, a second colorized image from the second adapted representation; and

computationally updating weights of the plurality of image filtering modules, the indirectly trained adaptor ANN, and the trained colorization ANN based on a loss function that comprises a difference between the second colorized image and the second image of the second image pair.

6. The computer-implemented method of claim 1 , wherein the image pair is part of a plurality of image pairs, and wherein the receiving, providing, determining, and indirect training are also applied to each of the plurality of image pairs.

7. The computer-implemented method of claim 1 , wherein the difference between the colorized image and the second image of the image pair comprises a pixel-wise difference between the colorized image and the second image of the image pair.

8. The computer-implemented method of claim 1 , wherein the trained colorization ANN is trained to receive a colorless image and correspondingly output a colorized version of the colorless image.

9. The computer-implemented method of claim 1 , wherein the providing comprises providing the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN to a printing device.

10. The computer-implemented method of claim 1 , wherein the second image of the image pair comprises a colorized version of the first image of the image pair.

11. The computer-implemented method of claim 1 , wherein each image filtering module from the plurality of image filtering modules is configured to apply a different image filter.

12. The computer-implemented method of claim 1 , wherein the plurality of image filtering modules comprise at least one of: a morphological filter, a blur filter, or a median filter.

13. A computing device, comprising:

one or more processors; and

non-transitory data storage storing at least computer-readable instructions that, when executed by the one or more processors, cause the computing device to perform operations comprising:

receiving an image pair, wherein a first image of the image pair comprises a respective initial training image and wherein a second image of the image pair comprises a respective ground truth training image;

providing instances of the first image to a plurality of image filtering modules, wherein each image filtering module is configured to apply an image filter on an input image and output a filtered representation of the input image;

determining respective filtered representations of the first image using the plurality of image filtering modules;

indirectly training an adaptor artificial neural network (ANN) by at least:

applying the adaptor ANN on at least the respective filtered representations of the first image to produce an adapted representation of the first image;

determining, using a trained colorization ANN, a colorized image from the adapted representation; and

computationally updating weights of the adaptor ANN based on a loss function that comprises a difference between the colorized image and the second image of the image pair; and

providing the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN.

14. The computing device of claim 13 , wherein the adaptor ANN comprises an input layer that is configured to receive and concatenate the respective filtered representations of the first image.

15. The computing device of claim 14 , wherein the input layer is further configured to receive an instance of the first image and concatenate the respective filtered representations with the instance of the first image.

16. The computing device of claim 13 , wherein the filtered representation of the input image comprises a difference between the input image and the input image after having been transformed by the image filter.

17. The computing device of claim 13 , wherein the operations further comprise:

after indirectly training the adaptor ANN, further training the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN by at least:

receiving a second image pair, wherein a first image of the second image pair comprises a respective second initial training image and wherein a second image of the second image pair comprises a respective second ground truth training image providing instances of the first image of the second image pair to the plurality of image filtering modules;

determining respective filtered representations of the first image of the second image pair using the plurality of image filtering modules;

applying the trained adaptor ANN on at least the respective filtered representations of the first image of the second image pair to produce a second adapted representation;

determining, using the trained colorization ANN, a second colorized image from the second adapted representation; and

computationally updating weights of the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN based on a loss function that comprises a difference between the second colorized image and the second image of the second image pair.

18. The computing device of claim 13 , wherein the difference between the colorized image and the second image of the image pair comprises a pixel-wise difference between the colorized image and the second image of the image pair.

19. The computing device of claim 13 , wherein the trained colorization ANN is trained to receive a colorless image and correspondingly output a colorized version of the colorless image.

20. An article of manufacture comprising non-transitory data storage storing at least computer-readable instructions that, when executed by one or more processors of a computing device, cause the computing device to perform operations comprising:

receiving an image pair, wherein a first image of the image pair comprises a respective initial training image and wherein a second image of the image pair comprises a respective ground truth training image;

providing instances of the first image to a plurality of image filtering modules, wherein each image filtering module is configured to apply an image filter on an input image and output a filtered representation of the input image;

determining respective filtered representations of the first image using the plurality of image filtering modules;

indirectly training an adaptor artificial neural network (ANN) by at least:

applying the adaptor ANN on at least the respective filtered representations of the first image to produce an adapted representation of the first image;

determining, using a trained colorization ANN, a colorized image from the adapted representation; and

computationally updating weights of the adaptor ANN based on a loss function that comprises a difference between the colorized image and the second image of the image pair; and

providing the plurality of image filtering modules, the trained adaptor ANN, and the trained colorization ANN.