Method and electronic device for processing images that can be played on a virtual device by using a super-resolution deep learning network model

1. A method for processing an image that can be played on a virtual reality (VR) device, comprising:

obtaining a super-resolution deep learning network model, which is trained to learn to reconstruct an image from low resolution to high resolution; wherein the super-resolution deep learning network model comprises a plurality of feature filters to extract features of the image;

modifying the resolution of the feature filters from a preset value to an established value, wherein the established value is higher than the preset value;

inputting a low-resolution image into the super-resolution deep learning network model; and

increasing the resolution of the low-resolution image to become a high-resolution image through the super-resolution deep learning network model;

wherein the training of the super-resolution deep learning network model comprises:

selecting a training image from an image set;

randomly cropping a first sampled image with a resolution of 384*384 pixels from the training image;

decreasing the resolution of the first sampled image to 96*96 pixels to obtain a second sampled image;

utilizing a resolution enhancement algorithm in the super-resolution deep learning network model to increase the resolution of the second sampled image back to 384*384 pixels to obtain a virtual image;

comparing the virtual image with the first sampled image to calculate a degree of distortion in the virtual image relative to the first sampled image;

repeatedly adjusting the feature weight of the resolution enhancement algorithm until the degree of distortion in the virtual image relative to the first sampled image is lower than a threshold; and

selecting another training image from the image set and repeating the training.

2. The method as claimed in claim 1 , further comprising:

decreasing the resolution of the high-resolution image to generate a medium-resolution image;

inputting the medium-resolution image into the virtual reality device for display.

3. The method as claimed in claim 2 , wherein the field of view of the medium-resolution image is 360 degrees, and the resolution of the medium-resolution image is one-half of that of the high-resolution image.

4. The method as claimed in claim 1 , wherein the low-resolution image and the training image are images with a 360-degree field of view and 4K resolution, and the high-resolution image is an image with a 360-degree field of view and 16K resolution.

5. The method as claimed in claim 1 , wherein the preset value is 3*3 pixels, and the established value is 5*5 pixels.

6. The method as claimed in claim 1 , wherein the super-resolution deep learning network model is a super-resolution generative adversarial network (SRGAN).

7. An electronic device, configured to process an image that can be played on a virtual reality device, comprising:

a processor, configured to execute a super-resolution deep learning network model, wherein the super-resolution deep learning network model is trained to learn to reconstruct an image from low resolution to high resolution, the super-resolution deep learning network model comprises a plurality of feature filters to extract features of the image, the resolution of the feature filters is modified from a preset value to an established value, and the established value is higher than the preset value;

wherein the processor inputs a low-resolution image into the super-resolution deep learning network model, and increases the resolution of the low-resolution image to become a high-resolution image through the super-resolution deep learning network model;

wherein the processor executes the training of the super-resolution deep learning network model, comprising:

selecting a training image from an image set;

randomly cropping a first sampled image with a resolution of 384*384 pixels from the training image;

decreasing the resolution of the first sampled image to 96*96 pixels to obtain a second sampled image;

executing the resolution enhancement algorithm in the super-resolution deep learning network model to increase the resolution of the second sampled image back to 384*384 pixels to obtain a virtual image;

comparing the virtual image with the first sampled image to calculate the distortion degree of the virtual image relative to the first sampled image;

repeatedly adjusting the feature weight of the resolution enhancement algorithm until the distortion degree of the virtual image relative to the first sampled image is lower than a threshold; and

selecting another training image from the image set and repeating the training.

8. The electronic device as claimed in claim 7 , wherein the processor decreases the resolution of the high-resolution image to generate a medium-resolution image, and sends the medium-resolution image to the virtual reality device for display through wired or wireless communication.

9. The electronic device as claimed in claim 8 , wherein the field of view of the medium-resolution image is 360 degrees, and the resolution of the medium-resolution image is one-half of that of the high-resolution image.

10. The electronic device as claimed in claim 7 , wherein the low-resolution image and the training image are images with a 360-degree field of view and 4K resolution, and the high-resolution image is an image with a 360-degree field of view and 16K resolution.

11. The electronic device as claimed in claim 7 , wherein the preset value is 3*3 pixels, and the established value is 5*5 pixels.

12. The electronic device as claimed in claim 11 , wherein the super-resolution deep learning network model is a super-resolution generative adversarial network.