Granted Patent B2

US 12,294,742 · App. 18/214,197 · Granted May 6, 2025

Inventors: Liang Wen (Beijing, CN); Weijing Shi (Beijing, CN); Lizhong Wang (Beijing, CN); Ying Zhang (Beijing, CN); Yinji Piao (Beijing, CN); Xiaoyan Lou (Beijing, CN)

Assignee: SAMSUNG ELECTRONICS CO., LTD.

H04N19/80H04N19/124H04N19/137H04N19/182H04N19/46H04N19/85H04N19/86

Abstract

Claims (54)

1. A video encoding method, comprising:

based on first distortion aware information about a current encoding unit and second distortion aware information about a corresponding reference encoding unit, determining a first predicted value of the current encoding unit using at least one interpolation filter, wherein the at least one type of interpolation filter comprises an artificial intelligence (AI) interpolation filter; and

encoding the current encoding unit based on the first predicted value of the current encoding unit.

2. The method according to claim 1 , wherein the first distortion aware information-comprises at least one of a quantization parameter of the current encoding unit and a quantization parameter difference between the current encoding unit and the corresponding reference encoding unit, and

wherein the second distortion aware information comprises at least one of a quantization parameter of the corresponding reference encoding unit, and the quantization parameter difference.

3. The method according to claim 2 , wherein the quantization parameter of the current encoding unit is determined based on the quantization parameter of the corresponding reference encoding unit and the quantization parameter difference, and

wherein the quantization parameter of the corresponding reference encoding unit is determined based on the quantization parameter of the current encoding unit and the quantization parameter difference.

4. The method according to claim 1 , wherein the AI interpolation filter comprises at least one convolution unit,

wherein the at least one convolution unit comprises at least one convolution layer and at least one nonlinear activation layer, and

wherein the at least one nonlinear activation layer is configured to, based on the first distortion aware information, process a convolution result output by a cascaded convolution layer.

5. The method according to claim 4 , wherein the at least one nonlinear activation layer is configured to process the convolution result output by the cascaded convolution layer based on an activation parameter determined based on the first distortion aware information.

6. The method according to claim 1 , wherein, based on at least two types of interpolation filters being used to determine the first predicted value of the current encoding unit, the method further comprises:

writing identification information which indicates an interpolation filter corresponding to the first predicted value into a code stream.

7. The method according to claim 6 , wherein the writing the identification information comprises:

writing the identification information into the code stream using a predetermined coding unit (CU)-level flag bit.

8. The method according to claim 1 , wherein the determining the first predicted value of the current encoding unit comprises:

based on the first distortion aware information and the second distortion aware information, determining a second predicted value corresponding to each sub-pixel position of the current encoding unit using the at least one type of interpolation filter; and

determining the first predicted value of the current encoding unit based on the second predicted value.

9. The method according to claim 8 , wherein the determining the second predicted value comprises:

obtaining a motion vector (MV) corresponding to an integer-pixel position of the current encoding unit; and

determining, based on a reference encoding unit corresponding to the MV, the second predicted value using the at least one type of interpolation filter.

10. The method according to claim 9 , wherein the determining the second predicted value of the current encoding unit further comprises:

based on at least one of a quantization parameter of the current encoding unit, a quantization parameter of the corresponding reference encoding unit, or a quantization parameter difference between the current encoding unit and the corresponding reference encoding unit, obtaining the second predicted value using the AI interpolation filter.

11. The method according to claim 10 , wherein the obtaining the second predicted value using the AI interpolation filter comprises:

expanding the reference encoding unit to a predetermined size to obtain a corresponding input matrix;

inputting, into at least one cascaded convolution unit of the AI interpolation filter, the input matrix and at least one of: the quantization parameter of the current encoding unit, the quantization parameter of the corresponding reference encoding unit, and the quantization parameter difference;

performing a convolution operation using a convolution layer of the convolution unit; and

processing the convolution result of the convolution layer using a nonlinear activation layer of the convolution unit to obtain a corresponding second predicted value which corresponds to a sub-pixel position of the current encoding unit.

12. The method according to claim 8 , wherein the determining the first predicted value of the current encoding unit further comprises:

obtaining a rate distortion cost based on each second predicted value; and

determining the first predicted value of the current encoding unit based on each rate distortion cost.

13. A video decoding method, comprising:

based on received code stream, determining a reference decoding unit corresponding to a current decoding unit;

based on first distortion aware information about the current decoding unit and second distortion aware information about the corresponding reference decoding unit, determining a predicted value of the current decoding unit using at least one interpolation filter, wherein the at least one type of interpolation filter comprises an artificial intelligence (AI) interpolation filter; and

decoding the current decoding unit based on the first predicted value of the current decoding unit.

14. The method according to claim 13 , further comprising:

obtaining, from the code stream, identification information which indicates an interpolation filter corresponding to the current decoding unit; and

based on the identification information, determining to use the AI interpolation filter to decode the current decoding unit.

15. The method according to claim 13 , wherein the first distortion aware information comprises at least one of: a quantization parameter of the current decoding unit, and a quantization parameter difference between the current decoding unit and the corresponding reference decoding unit, and

wherein the second distortion aware information comprises at least one of: a quantization parameter of the reference decoding unit, and the quantization parameter difference.

16. The method according to claim 15 , wherein the quantization parameter of the current decoding unit is determined based on the quantization parameter of the reference decoding unit and the quantization parameter difference, and

wherein the quantization parameter of the reference decoding unit is determined based on the quantization parameter of the current decoding unit and the quantization parameter difference.

17. The method according to claim 13 , wherein the AI interpolation filter comprises at least one convolution unit,

wherein the at least one convolution unit comprises at least one convolution layer and at least one nonlinear activation layer and

wherein the at least one nonlinear activation layer is configured to process a convolution result output by a cascaded convolution layer based on first the distortion aware information of the current decoding unit.

18. The method according to claim 17 , wherein the at least one nonlinear activation layer is configured to, using an activation parameter determined based on the distortion aware information of the current decoding unit, process the convolution result output by the cascaded convolution layer.

19. The method according to claim 13 , wherein the video decoding method further comprises:

obtaining motion vector precision information from the code stream,

wherein the predicted value of the current decoding unit is determined using the motion vector precision information and the at least one interpolation filter based on the first distortion aware information and the second distortion aware information.

20. An encoding device, comprising:

a memory configured to store instructions; and

a processor configured to execute the instructions to:

based on first distortion aware information about a current encoding unit and second distortion aware information about a corresponding reference encoding unit, determine a first predicted value of the current encoding unit using at least one interpolation filter, wherein the at least one type of interpolation filter comprises an artificial intelligence (AI) interpolation filter; and

encode the current encoding unit based on the first predicted value of the current encoding unit.

Assignments (1)

Priority Claims (2)

Continuity (2)

References Cited (22)