Granted Patent B2

US 12,542,923 · App. 18/402,569 · Granted Feb 3, 2026

Inventors: Saifeng Ni (Santa Clara, CA); Madhukar Budagavi (Plano, TX); Rajan Laxman Joshi (San Diego, CA)

Assignee: Samsung Electronics Co., Ltd.

H04N19/59G06T5/10H04N19/117H04N19/159H04N19/184H04N19/80

Abstract

Claims (74)

1 . An apparatus comprising:

a communication interface configured to receive a bitstream for a compressed video; and

a processor operably coupled to the communication interface, the processor configured to:

decode the bitstream for the compressed video and obtain reference samples associated with a block of a video frame;

select a down-sampling filter to use in down-sampling the reference samples;

apply the selected down-sampling filter to the reference samples to obtain reduced reference samples;

obtain a reduced intra prediction associated with the block using the reduced reference samples;

select an up-sampling filter to use in up-sampling the reduced intra prediction,

wherein a sum of losses, determined using a loss function, for a plurality of blocks with a same size as a block size of the block of the video frame, is minimized to obtain the at least one of the down-sampling and the up-sampling filter;

apply the selected up-sampling filter to the reduced intra prediction to obtain a full intra prediction of the block; and

output the full intra prediction of the block.

2 . The apparatus of claim 1 , wherein, to select the down-sampling filter, the processor is further configured to select the down-sampling filter from a set of filters based on a decimation factor.

3 . The apparatus of claim 1 , wherein the selection of at least one of the down-sampling filter and the up-sampling filter is based on block features including a variation metric of the reference samples.

4 . The apparatus of claim 3 , wherein, to select the down-sampling filter, the processor is further configured to:

determine whether the variation metric is above a variation threshold; and

if the variation metric is above the variation threshold, select the down-sampling filter from a set of filters based on a decimation factor, or

if the variation metric is below the variation threshold, select an averaging filter as the down-sampling filter.

5 . The apparatus of claim 3 , wherein, to select the up-sampling filter, the processor is further configured to:

determine whether the variation metric is above a variation threshold; and

if the variation metric is above the variation threshold, select a discrete cosine transform interpolation filter (DCTIF) as the up-sampling filter, or

if the variation metric is below the variation threshold, select a linear interpolation filter as the up-sampling filter.

6 . The apparatus of claim 5 , wherein the selected DCTIF filter is a discrete cosine transform type II filter.

7 . The apparatus of claim 3 , wherein the reference samples include left reference samples and top references samples relative to the block, and wherein the processor is further configured to:

determine the variation metric for each of the left reference samples and the top reference samples;

select a first down-sampling filter for the left reference samples and a second down-sampling filter for the top reference samples; and

select a first up-sampling filter for the left reference samples and a second up-sampling filter for the top reference samples.

8 . The apparatus of claim 1 , wherein the at least one of the down-sampling filter and the up-sampling filter are optimized using the loss function, wherein the loss function is based on a prediction residual, and wherein the prediction residual is a difference between an original block and a predicted block.

9 . A method comprising:

receiving and decoding a bitstream for a compressed video;

obtaining reference samples associated with a block of a video frame;

selecting a down-sampling filter to use in down-sampling the reference samples;

applying the selected down-sampling filter to the reference samples to obtain reduced reference samples;

obtaining a reduced intra prediction associated with the block using the reduced reference samples;

selecting an up-sampling filter to use in up-sampling the reduced intra prediction,

wherein a sum of losses, determined using a loss function, for a plurality of blocks with a same size as a block size of the block of the video frame, is minimized to obtain the at least one of the down-sampling and the up-sampling filter;

applying the selected up-sampling filter to the reduced intra prediction to obtain a full intra prediction of the block; and

outputting the full intra prediction of the block.

10 . The method of claim 9 , wherein selecting the down-sampling filter includes selecting the down-sampling filter from a set of filters based on a decimation factor.

11 . The method of claim 9 , wherein the selection of at least one of the down-sampling filter and the up-sampling filter is based on block features including a variation metric of the reference samples.

12 . The method of claim 11 , wherein selecting the down-sampling filter includes:

determining whether the variation metric is above a variation threshold; and

if the variation metric is above the variation threshold, selecting the down-sampling filter from a set of filters based on a decimation factor, or

if the variation metric is below the variation threshold, selecting an averaging filter as the down-sampling filter.

13 . The method of claim 11 , wherein selecting the up-sampling filter includes:

determining whether the variation metric is above a variation threshold; and

if the variation metric is above the variation threshold, selecting a discrete cosine transform interpolation filter (DCTIF) as the up-sampling filter, or

if the variation metric is below the variation threshold, selecting a linear interpolation filter as the up-sampling filter.

14 . The method of claim 13 , wherein the selected DCTIF filter is a discrete cosine transform type II filter.

15 . The method of claim 11 , wherein the reference samples include left reference samples and top references samples relative to the block, and wherein the method further comprises:

determining the variation metric for each of the left reference samples and the top reference samples;

selecting a first down-sampling filter for the left reference samples and a second down-sampling filter for the top reference samples; and

selecting a first up-sampling filter for the left reference samples and a second up-sampling filter for the top reference samples.

16 . The method of claim 9 , wherein the at least one of the down-sampling filter and the up-sampling filter are optimized using the loss function, wherein the loss function is based on a prediction residual, and wherein the prediction residual is a difference between an original block and a predicted block.

17 . An apparatus comprising:

a communication interface; and

a processor operably coupled to the communication interface, the processor configured to:

subdivide a video frame into a plurality of blocks of a plurality of block sizes;

obtain reference samples associated with a block of the plurality of blocks;

select a down-sampling filter to use in down-sampling the reference samples;

apply the selected down-sampling filter to the reference samples to obtain reduced reference samples;

obtain a reduced intra prediction associated with the block using the reduced reference samples;

select an up-sampling filter to use in up-sampling the reduced intra prediction,

wherein a sum of losses, determined using the loss function, of multiple blocks with a same size as a block size of the block, is minimized to obtain the at least one of the down-sampling filter and the up-sampling filter;

apply the selected up-sampling filter to the reduced intra prediction to obtain a full intra prediction of the block; and

encode and transmit the prediction residual as part of a bitstream, wherein the prediction residual is a difference between an original block and a predicted block.

18 . The apparatus of claim 17 , wherein the at least one of the down-sampling filter and the up-sampling filter are optimized using the loss function, wherein the loss function is based on the prediction residual.

19 . The apparatus of claim 17 , wherein, to select the down-sampling filter, the processor is further configured to:

determine whether a variation metric of the reference samples is above a variation threshold; and

if the variation metric is above the variation threshold, select the down-sampling filter from a set of filters based on a decimation factor, or

if the variation metric is below the variation threshold, select an averaging filter as the down-sampling filter.

20 . The apparatus of claim 17 , wherein, to select the up-sampling filter, the processor is further configured to:

determine whether a variation metric is above a variation threshold; and

if the variation metric is above the variation threshold, select a discrete cosine transform interpolation filter (DCTIF) as the up-sampling filter, or

if the variation metric is below the variation threshold, select a linear interpolation filter as the up-sampling filter.

Assignments (1)

Continuity (4)

References Cited (16)