Method and data processing system for lossy image or video encoding, transmission and decoding

1. A method for lossy image or video encoding, transmission and decoding, the method comprising the steps of:

receiving an input image at a first computer system;

encoding the input image using a first trained neural network to produce a latent representation;

identifying one or more regions of the input image associated with high visual sensitivity;

encoding the one or more regions of the input image associated with high visual sensitivity using a second trained neural network to produce one or more region latent representations;

performing a quantization process on the latent representation and the one or more region latent representations;

transmitting the result of the quantization process to a second computer system; and

decoding the result of the quantization process to produce an output image, wherein the output image is an approximation of the input image.

2. The method of claim 1 , wherein the one or more regions of the input image associated with high visual sensitivity are identified by a third trained neural network.

3. The method of claim 1 , wherein the location of the areas of the one or more regions of the input image associated with high visual sensitivity are stored in a binary mask.

4. The method of claim 1 , further comprising the step of transmitting location information identifying the location of the one or more of the regions of the input image associated with high visual sensitivity to the second computer system.

5. The method of claim 1 , wherein the quantization process comprises quantizing the first latent representation to produce a quantized latent and quantizing the one or more region latent representations to produce one or more quantized region latents; and

the step of transmitting the result of the quantization process comprises transmitting the quantized latent and the one or more quantized region latents.

6. The method of claim 5 , wherein the step of decoding the result of the quantization process comprises:

decoding the quantized latent using a fourth trained neural network to produce an intermediate image;

decoding the one or more quantized region latents using a fifth trained neural network to produce one or more output region images; and

combining the intermediate image and the one or more output region images to produce the output image.

7. The method of claim 6 , wherein the intermediate image and the one or more region images are combined using Poisson Blending.

8. The method of claim 6 , wherein the intermediate image and the one or more region images are combined using a sixth trained neural network.

9. The method of claim 5 , wherein the step of decoding the result of the quantization process comprises:

decoding the quantized latent and the one or more quantized region latents using a seventh trained neural network to produce the output image.

10. The method of claim 9 , wherein the quantized latent and the one or more quantized region latents are input into separate layers of the seventh trained neural network; and

information corresponding to each of the quantized latent and the one or more quantized region latents are both input into a subsequent layer of the seventh trained neural network.

11. The method of claim 5 , wherein the method further comprises:

combining the quantized latent and the one or more quantized region latents to produce a quantized combined latent;

wherein the step of transmitting the result of the quantization process comprises transmitting the quantized combined latent.

12. The method of claim 11 , wherein the step of decoding the result of the quantization process comprises decoding the quantized combined latent using an eighth trained neural network.

13. The method of claim 1 , further comprising the step of merging the latent representation and the one or more region latent representations to produce a merged latent representation; and

quantizing the merged latent representation to produce a quantized merged latent;

wherein the step of transmitting the result of the quantization process comprises transmitting the quantized merged latent.

14. The method of claim 13 , wherein the step of decoding the result of the quantization process comprises decoding the quantized merged latent using a ninth trained neural network.

15. The method of claim 1 , wherein the one or more regions of high visual sensitivity includes a plurality of categories of region; and

an additional trained neural network is used to encode each category of region during the encoding of the one or more regions of high visual sensitivity.

16. A method of training one or more neural networks, the one or more neural networks being for use in lossy image or video encoding, transmission and decoding, the method comprising the steps of:

receiving a first input training image;

encoding the first input training image using a first neural network to produce a latent representation;

identifying one or more regions of the input image associated with high visual sensitivity;

encoding the one or more regions of the input image associated with high visual sensitivity using a second trained neural network to produce one or more region latent representations;

performing a quantization process on the latent representation and the one or more region latent representations;

decoding the result of the quantization process to produce an output image, wherein the output image is an approximation of the input training image;

evaluating a loss function based on differences between the output image and the input training image;

evaluating a gradient of the loss function;

back-propagating the gradient of the loss function through the first neural network to update the parameters of the first neural network; and

repeating the above steps using a first set of training images to produce a first trained neural network.

17. The method of claim 16 , further comprising, prior to the step of encoding the first input training image, performing the steps of:

receiving a second input training image;

encoding the second input training image using a second neural network to produce a latent representation;

performing a quantization process on the latent representation;

decoding the result of the quantization process to produce an output image, wherein the output image is an approximation of the second input training image

evaluating a loss function based on differences between the output image and the second input training image;

evaluating a gradient of the loss function;

back-propagating the gradient of the loss function through the second neural network to update the parameters of the second neural network; and

repeating the above steps using a second set of training images to produce a second trained neural network.

18. The method of claim 17 , wherein the same set of training images is used to produce the first trained neural network and the second trained neural network.

19. The method of claim 16 , wherein the difference between the output image and the input training image is determined by a neural network acting as a discriminator; and

back-propagation of the gradient of the loss function is additionally used to update the parameters of the neural network acting as a discriminator.

20. A data processing system configured to perform the method of claim 1 .