Bi-predicted Motion Vector Refinement Using Gradient Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding techniques face challenges in efficiently deriving motion vectors for bi-predicted blocks, leading to suboptimal compression and decoding performance, particularly in complex video sequences.
Innovation Solution
The proposed solution involves refining bi-predicted motion vector predictors using gradient information to identify refined reference blocks within a search range, enabling more accurate motion vector derivation and improved predictive block generation for efficient video decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional motion vector derivation is used for bi-predicted blocks, then decoding complexity is reduced, but compression efficiency and prediction accuracy deteriorate
Solution Approach 1:
The patent applies preliminary action by performing gradient calculation and motion vector refinement in advance during the decoding process. The gradient information is computed beforehand for the current block and reference blocks, and motion vector candidates are refined using this pre-computed gradient data before final prediction. This preliminary processing improves motion vector accuracy without significantly increasing overall decoding complexity.
Solution Approach 2:
The patent implements local quality by applying gradient-based refinement selectively to motion vector derivation in bi-predicted blocks. The gradient calculation and refinement process focuses specifically on the local region of the current block and its corresponding reference blocks, rather than processing the entire picture. This localized approach improves prediction accuracy for specific blocks while maintaining overall decoding efficiency.
2Reliability
If gradient-based refinement is applied to motion vectors, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent applies partial action by implementing gradient-based refinement only for bi-predicted blocks where it provides the most benefit, rather than applying it uniformly to all blocks. The refinement process uses a limited number of gradient calculations and motion vector candidate evaluations, performing just enough refinement to improve prediction accuracy without excessive computational overhead. This selective application balances reliability improvement with computational efficiency.
Data Source
AI summary
A method of decoding video data includes determining, by a video decoder implemented in circuitry, a bi-predicted MV predictor for a block of video data. The bi-predicted MV predictor indicates a first input reference block and a second input reference block. The method further includes refining, by the video decoder, the bi-predicted MV predictor using gradient information to determine a refined bi-predicted MV predictor indicating a first refined reference block that is within a search range from the first input reference block and a second refined reference block that is within the search range from the second input reference block. The method further includes generating, by the video decoder, a predictive block for the block of video data based on the refined bi-predicted MV predictor, and decoding, by the video decoder, the block of video data based on the predictive block.