Systems and methods for motion compensated residual prediction
Systems and methods are disclosed for improving the prediction efficiency for residual prediction using motion compensated residual prediction (MCRP). Exemplary residual prediction techniques employ motion compensated prediction and processed residual reference pictures. Further disclosed herein are systems and methods for generating residual reference pictures. These pictures can be generated adaptively with or without considering in-loop filtering effects. Exemplary de-noising filter designs are also described for enhancing the quality of residual reference pictures, and compression methods are described for reducing the storage size of reference pictures. Further disclosed herein are exemplary syntax designs for communicating residuals' motion information.
1. A method comprising:
generating a residual reference picture comprising at least one block, the residual reference picture representing a difference between a corresponding block of a reference picture and a corresponding first-order prediction of the reference picture;
filtering the residual reference picture with a de-noising filter to generate a filtered residual reference picture; and
for at least a current block in a current picture:
forming a first-order prediction of the current block;
predicting residuals of the current block from a residual reference block in the filtered residual reference picture using motion-compensated prediction; and
combining the first-order prediction and the predicted residuals to generate a composite prediction signal for the current block.
2. The method of claim 1 , wherein a flag is signaled in a bitstream at the block level indicating whether to predict residuals of the current block from the filtered residual reference picture.
3. The method of claim 1 , wherein generating the first-order prediction comprises predicting the current block using intra prediction.
4. The method of claim 1 , wherein generating the first-order prediction comprises predicting the current block using inter prediction.
5. The method of claim 1 , wherein filtering the residual reference picture comprises taking an absolute value of at least some of the samples in the residual reference picture.
6. The method of claim 5 , wherein a sign value is signaled in a bitstream for the current block, and wherein predicting residuals of the current block comprises multiplying the residual reference block by the sign value.
7. The method of claim 5 , wherein a weighting factor is signaled in a bitstream for the current block, and wherein predicting residuals of the current block comprises multiplying the residual reference block by the weighting factor.
8. The method of claim 1 , wherein generating the residual reference picture comprises subtracting a first-order prediction of the reference picture from either a filtered or an unfiltered reconstruction signal of the reference picture, and wherein an indication of whether to use the filtered or the unfiltered reconstruction signal is signaled in a bitstream.
9. The method of claim 1 , wherein filtering the residual reference picture comprises applying a threshold to the residual reference picture such that residual values below the threshold in the residual reference picture are zero in the filtered residual reference picture.
10. The method of claim 9 , wherein the threshold is selected adaptively to substantially achieve a predetermined ratio k, where k is the ratio of the number of samples below the threshold to the number of samples above the threshold in the current picture.
11. The method of claim 1 , wherein filtering the residual reference picture comprises quantizing sample values in the residual reference picture.
12. The method of claim 1 , wherein filtering the residual reference picture comprises applying a linear filter to the residual reference picture, wherein coefficients of the linear filter are signaled in a bitstream.
13. The method of claim 1 , wherein a residual motion vector is signaled in a bitstream, and wherein predicting residuals includes identifying the residual reference block using the residual motion vector.
14. The method claim 1 , further comprising:
determining a second-order residual representing a difference between (i) an input video block corresponding to the current block and (ii) the composite prediction signal for the current block; and
encoding the second-order residual in a bitstream.
15. The method claim 1 , further comprising:
receiving, in a bitstream, a second-order residual for the current block; and
reconstructing the current block by adding the second-order residual to the composite prediction signal for the current block.
16. A system comprising:
a processor configured to perform at least:
generating a residual reference picture comprising at least one block, the residual reference picture representing a difference between a corresponding block of a reference picture and a first-order prediction of the reference picture;
filtering the residual reference picture with a de-noising filter to generate a filtered residual reference picture; and
for at least a current block in a current picture:
forming a first-order prediction of the current block;
predicting residuals of the current block from a residual reference block in the filtered residual reference picture using motion-compensated prediction; and
combining the first-order prediction and the predicted residuals to generate a composite prediction signal for the current block.
17. The system of claim 16 , wherein the processor is further configured to perform functions comprising:
determining a second-order residual representing a difference between (i) an input video block corresponding to the current block and (ii) the composite prediction signal for the current block; and
encoding the second-order residual in a bitstream.
18. The system of claim 16 , wherein the processor is further configured to perform functions comprising:
receiving, in a bitstream, a second-order residual for the current block; and
reconstructing the current block by adding the second-order residual to the composite prediction signal for the current block.
19. A video decoder for decoding a video from a bitstream, wherein the video comprises a plurality of pictures, including a current picture and at least one reference picture, each picture comprising blocks of samples, the video decoder being operative to perform functions comprising:
generating a residual reference picture comprising at least one block, the residual reference picture representing a difference between a corresponding block of the reference picture and a first-order prediction of the reference picture;
filtering the residual reference picture with a de-noising filter to generate a filtered residual reference picture; and
for at least a current block in the current picture:
forming a first-order prediction of the current block;
predicting residuals of the current block from a residual reference block in the filtered residual reference picture using motion-compensated prediction; and
combining the first-order prediction and the predicted residuals to generate a composite prediction signal for the current block.
20. The video decoder of claim 19 , wherein the decoder is further configured to perform functions comprising:
receiving, in a bitstream, a second-order residual for the current block; and
reconstructing the current block by adding the second-order residual to the composite prediction signal for the current block.