Affine Merge Candidate Derivation Using Virtual Blocks in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards face challenges in accurately deriving motion vector candidates for affine merge mode, leading to inefficiencies in video compression and quality degradation.
Innovation Solution
The proposed methods involve obtaining candidate positions for affine and translational motion vectors from non-adjacent neighbor blocks, constructing virtual blocks, and projecting control point motion vectors to improve motion prediction accuracy, thereby enhancing the derivation of motion vectors for both regular inter and affine modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only adjacent neighboring blocks are used for affine merge candidate derivation, then the derivation process is simple, but the motion prediction accuracy is limited
Solution Approach 1:
The patent segments the candidate derivation process into two distinct parts: adjacent neighbor block derivation (maintaining simplicity) and non-adjacent neighbor block derivation (improving accuracy). By dividing the derivation into optional stages, the system can achieve higher accuracy when needed while keeping the basic process simple for常规 cases.
Solution Approach 2:
The patent implements partial action by making the non-adjacent neighbor block derivation optional rather than mandatory. This allows the system to apply enhanced derivation methods selectively when improvement is needed, without always incurring the full complexity cost. The encoder can choose to use only adjacent neighbors or both adjacent and non-adjacent neighbors based on content requirements.
2Measurement precision
If non-adjacent neighboring blocks are utilized for affine merge candidate derivation, then motion prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent introduces dynamic adaptability by allowing the encoder to switch between different derivation modes (adjacent-only or both adjacent and non-adjacent) based on content characteristics. This dynamic approach enables the system to adjust computational effort according to actual needs, using more complex non-adjacent derivation only when it provides measurable improvement over simpler adjacent-only derivation.
Solution Approach 2:
The patent changes the parameter of candidate block selection by introducing non-adjacent neighboring blocks as an additional option beyond the traditional adjacent blocks. This parameter change expands the search space for motion candidates, allowing the system to find more accurate motion vectors when non-adjacent blocks provide better prediction, while maintaining the original adjacent block approach as a baseline.
3Productivity
If a broader range of neighboring blocks is used for motion vector derivation, then coding efficiency improves, but the processing time increases
Solution Approach 1:
The patent applies preliminary action by first deriving candidates from adjacent neighboring blocks (quick and simple), then optionally adding non-adjacent neighbor block candidates if improvement is needed. This staged approach ensures that the basic, faster derivation is always completed first, with the more time-consuming non-adjacent derivation applied only as a supplementary enhancement when beneficial.
Data Source
AI summary
Methods for video decoding and encoding, apparatuses and non-transitory storage media are provided. In one decoding method, the decoder obtains a first candidate position and a second candidate position. The decoder obtains a third candidate position based on the first and second candidate positions and obtains a virtual block based on the first, the second, and the third candidate positions. The decoder may obtain a plurality of CPMVs for the virtual block based on translational MVs at the first, second, and third candidate positions; and project, the plurality of CPMVs for the virtual block to a current block to obtain a translational MV based on a specific position within the current block or a second plurality of CPMVs for the current block.