Affine Motion Vector Derivation Using Extended Control Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding schemes face limitations in deriving motion vectors for affine prediction, particularly when the interval between control points is equal to the block side length, leading to decreased precision and increased decoding complexity, especially in cases where affine prediction is used.
Innovation Solution
A prediction image generation device that computes motion vectors for prediction blocks within a target block by referencing motion vectors at control points with a longer interval than the block side, and uses relative coordinates from these points to generate a prediction image, thereby improving precision and reducing decoding complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If control points are selected at corners of the decoding target block with interval equal to block side length, then the device complexity is reduced, but the motion vector derivation precision deteriorates
Solution Approach 1:
The patent changes the parameter of control point selection by extending the interval between control points beyond the block side length to specific distances (e.g., 64 pixels or 2 times the block width/height). This parameter change improves motion vector precision by capturing broader motion patterns while maintaining manageable device complexity through standardized distance calculations.
Solution Approach 2:
The patent introduces a new dimension to control point selection by using points that are not just at block corners but at extended positions relative to the block. This dimensional extension allows for better motion representation by referencing a wider area, improving precision without proportionally increasing device complexity.
2Adaptability or versatility
If two-step processing is used to derive motion vectors from prediction units to sub-blocks, then the adaptability to different block sizes is improved, but the motion vector precision deteriorates due to cumulative errors
Solution Approach 1:
The patent segments the motion vector derivation process by directly calculating motion vectors for each sub-block independently using affine transformation formulas, rather than using a two-step hierarchical approach. This segmentation eliminates cumulative errors while maintaining adaptability to different block sizes through the flexible application of the affine formula to various partitioning schemes.
Solution Approach 2:
The patent introduces affine transformation parameters (a, b, c, d) as intermediaries that directly relate control point motion vectors to sub-block motion vectors. This intermediary approach allows for precise direct calculation without the cumulative errors of multi-step processing, while the parameters can be adapted to any block size configuration.
3Speed
If the size of prediction unit containing reference point is not considered, then the processing speed is improved, but the motion vector derivation precision deteriorates
Solution Approach 1:
The patent changes the approach by using fixed affine transformation parameters and standardized control point positions that are independent of prediction unit size variations. This allows for consistent precision across different block sizes while maintaining processing speed through efficient direct calculation formulas that don't require dynamic parameter adjustment based on prediction unit dimensions.
Data Source
AI summary
At least one of a vector candidate derivation unit and a merge candidate derivation unit derives a motion vector of each of multiple sub-blocks contained in a decoding target block by referring to motion vectors at multiple control points including two points at an interval which is longer than one side of the target block, and a prediction image generation unit refers to the motion vector of each sub-block to generate a prediction image.