Overlapped Motion Refinement for Smoother Video Subblock Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing decoder-side motion vector refinement (DMVR) scheme in video coding introduces inconsistencies and large motion vector differences between subblocks, leading to noticeable subblock boundaries in reconstructed pictures.
Innovation Solution
Extend the size of reference blocks during decoder-side motion search to increase overlap and correlation between neighboring subblocks, using larger reference blocks to calculate cost and refine motion vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If independent motion vector refinement is applied to each subblock, then motion vector precision is improved, but subblock boundary consistency deteriorates
Solution Approach 1:
The patent merges the motion vector refinement process across adjacent subblocks by having each subblock use reference blocks that overlap with neighboring subblocks. This combining approach ensures that motion vectors at subblock boundaries are consistent while still allowing precise independent refinement within each subblock's region.
Solution Approach 2:
The patent applies local quality by allowing different refinement strategies in different regions. Interior subblocks can use standard independent refinement for maximum precision, while boundary subblocks use overlapping reference blocks to ensure consistency with neighbors, creating locally optimized quality throughout the block.
2Stability of the object's composition
If reference block size is increased to improve motion vector correlation, then subblock boundary consistency is improved, but computational complexity increases
Solution Approach 1:
The patent applies partial action by using overlapping reference blocks only where necessary - specifically at subblock boundaries - rather than uniformly across all subblocks. This selective approach improves motion vector correlation where needed while avoiding unnecessary computational complexity in interior regions.
3Stability of the object's composition
If overlapping reference blocks are used for neighboring subblocks, then motion vector consistency is improved, but processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-identifying and marking boundary subblocks that require overlapping reference blocks before the main motion vector refinement process. This allows the system to prepare the necessary reference block overlaps in advance, reducing the processing time penalty during the actual refinement stage.
Data Source
AI summary
Methods and apparatus for encoding or decoding a first structure within a picture in a video sequence are provided. The method comprises, for a second structure within the first structure, determining an initial motion vector, MV, and MV candidates. The method comprises, for each of the MV candidates, using the MV candidate to derive first and second reference blocks, and a size of the first and second reference blocks may not be equal to a size of the second structure. The method comprises deriving a cost value based on samples from the first and second reference blocks. The method comprises using the cost values from the MV candidates to derive a best MV candidate and using the best MV candidate to derive prediction samples for the second structure.