Nonlinear processing method and apparatus for image compression

Abstract

This application provides a nonlinear processing method and apparatus for image compression. The nonlinear processing method for image compression in this application includes the steps of acquiring a first image feature to be processed and performing a nonlinear transformation process on the first image feature to acquire a processed image feature, wherein the nonlinear transformation process sequentially includes a first convolution process, a second convolution process, and element-wise multiplication operations. This application can expand the receptive field and reduce the computational load, thereby further enhancing local attention, maintaining model performance, and effectively balancing computational power.

Claims

1. A nonlinear processing method for image compression, The steps include: obtaining the first image features to be processed, A step of obtaining processed image features by performing a nonlinear transformation process on the first image feature, wherein the nonlinear transformation process sequentially includes a first convolution process, a second convolution process, and element-wise multiplication operations. Methods that include...

2. The aforementioned nonlinear transformation process further includes element-wise addition operations after the element-wise multiplication operations, The method according to claim 1.

3. The nonlinear transformation process further includes a first activation process after the first convolution process. The method according to claim 1 or 2.

4. The aforementioned nonlinear transformation process further includes a second activation process after the second convolution process. The method according to claim 3.

5. The first activation process and the second activation process each include an identity operation, a rectifier linear unit ReLU process, a leaky rectifier linear unit LeakyReLU process, or a parametric rectifier linear unit PReLU process, The method according to claim 4.

6. The step of performing the nonlinear transformation process on the aforementioned image feature to obtain the processed image feature is: The steps include: performing the first convolution operation on the first image feature to obtain a second image feature; The steps include: performing the two convolution operations on the two image features to obtain a third image feature; The steps include: performing the element-wise multiplication operation on the first image feature and the third image feature to obtain the processed image feature; The method according to claim 1, including the method described in claim 1.

7. The step of performing the nonlinear transformation process on the aforementioned image feature to obtain the processed image feature is: The steps include: performing the first convolution operation on the first image feature to obtain a second image feature; The steps include: performing the two convolution operations on the two image features to obtain a third image feature; The steps include: performing the element-wise multiplication operation on the first image feature and the third image feature to obtain a fourth image feature; The steps include: performing the element-wise addition operation on the first image feature and the fourth image feature to obtain the processed image feature; The method according to claim 2, including the method described in claim 2.

8. The first and second convolution processes are performed under the following conditions, namely: 1 <= mid_chs <= chs The following conditions are met, where mid_chs represents the number of output channels in the first convolution process or the number of input channels in the second convolution process, and chs represents the number of input channels in the first convolution process or the number of output channels in the second convolution process. The method according to any one of claims 1 to 7.

9. chs = mid_chs or chs = 2 × mid_chs. The method according to claim 8.

10. The convolution kernel size of the first convolution operation is the same as or different from the convolution kernel size of the second convolution operation. The method according to any one of claims 1 to 9.

11. After the step of performing the nonlinear transformation process on the aforementioned image feature to obtain the processed image feature, A step of obtaining a bitstream by performing encoding based on the processed image features, The method according to any one of claims 1 to 10, further comprising:

12. After the step of performing the nonlinear transformation process on the aforementioned image feature to obtain the processed image feature, A step of obtaining a reconstructed image based on the processed image features, The method according to any one of claims 1 to 10, further comprising:

13. A nonlinear processing device for image compression, An acquisition module configured to acquire the first image features to be processed, A transformation module configured to obtain processed image features by performing a nonlinear transformation process on the first image feature, wherein the nonlinear transformation process sequentially includes a first convolution process, a second convolution process, and an element-wise multiplication operation. A device equipped with the following features.

14. The aforementioned nonlinear transformation process further includes element-wise addition operations after the element-wise multiplication operations, The apparatus according to claim 13.

15. The nonlinear transformation process further includes a first activation process after the first convolution process. The apparatus according to claim 13 or 14.

16. The aforementioned nonlinear transformation process further includes a second activation process after the second convolution process. The apparatus according to claim 15.

17. The first activation process and the second activation process each include an identity operation, a rectifier linear unit ReLU process, a leaky rectifier linear unit LeakyReLU process, or a parametric rectifier linear unit PReLU process, The apparatus according to claim 16.

18. The aforementioned conversion module specifically, The system is configured to perform the first convolution operation on the first image feature to obtain a second image feature, perform the second convolution operation on the second image feature to obtain a third image feature, and perform the element-wise multiplication operation on the first and third image features to obtain the processed image feature. The apparatus according to claim 13.

19. The aforementioned conversion module specifically, The system is configured to obtain a second image feature by performing the first convolution operation on the first image feature, obtain a third image feature by performing the second convolution operation on the second image feature, obtain a fourth image feature by performing the element-wise multiplication operation on the first and third image features, and obtain the processed image feature by performing the element-wise addition operation on the first and fourth image features. The apparatus according to claim 14.

20. The first and second convolution processes are performed under the following conditions, namely: 1 <= mid_chs <= chs The following conditions are met, where mid_chs represents the number of output channels in the first convolution process or the number of input channels in the second convolution process, and chs represents the number of input channels in the first convolution process or the number of output channels in the second convolution process. The apparatus according to any one of claims 13 to 19.

21. chs = mid_chs or chs = 2 × mid_chs. The apparatus according to claim 20.

22. The convolution kernel size of the first convolution operation is the same as or different from the convolution kernel size of the second convolution operation. The apparatus according to any one of claims 13 to 21.

23. It is an encoder, One or more processors, A non-temporary computer-readable storage medium coupled to the processor and storing a program to be executed by the processor, An encoder comprising the above, wherein when the program is executed by the processor, the decoder becomes capable of performing the method according to any one of claims 1 to 11.

24. It is a decoder, One or more processors, A non-temporary computer-readable storage medium coupled to the processor and storing a program to be executed by the processor, A decoder comprising the above, wherein when the program is executed by the processor, the decoder becomes capable of performing the method according to any one of claims 1 to 11 and 12.

25. A computer program product comprising program code, wherein when the program code is executed on a computer or processor, the method according to any one of claims 1 to 12 is executed.

26. A computer-readable storage medium containing an instruction, wherein when the instruction is executed on a computer, the computer is able to perform the method according to any one of claims 1 to 12.

Images