Affine Motion Predictor Candidates for Better Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video encoding technologies, particularly in high compression formats like HEVC, face limitations in the set of control point motion vectors used as predictors in Affine Merge and Advanced Motion Vector Prediction (AMVP) modes, leading to suboptimal compression performance.
Innovation Solution
The method involves determining a set of predictor candidates for inter coding mode based on spatial neighboring blocks, using a motion model to derive motion fields from control point motion vectors, and selecting the best candidate through rate distortion optimization, which includes expanding the predictor set to include non-affine motion information and creating new candidates from neighboring block motion vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a limited set of control point motion vectors is used as predictors in Affine Merge and AMVP modes, then the encoding process remains simple, but compression performance deteriorates
Solution Approach 1:
The patent segments the predictor candidate set into multiple categories: spatial neighboring block predictors, temporal predictors, and affine motion model-based predictors. Each segment uses different motion vector derivation methods, allowing the system to explore multiple prediction approaches without overwhelming complexity in a single unified method.
Solution Approach 2:
The patent introduces a new dimension to motion vector prediction by incorporating affine motion models that use control point motion vectors from multiple spatial locations (top-left, top-right, bottom-left corners) rather than relying on a single motion vector. This multi-dimensional approach enriches the predictor candidate set with diverse motion information.
2Productivity
If more predictor candidates are generated from spatial neighboring blocks, then compression efficiency improves, but the number of processing steps increases
Solution Approach 1:
The patent performs preliminary actions by pre-determining spatial neighboring blocks and their motion vectors during the encoding process setup. The affine motion model parameters are pre-computed from control point motion vectors, so that when predictor candidates are needed, the system can quickly select from pre-processed information rather than computing everything from scratch.
Solution Approach 2:
The patent generates a comprehensive set of predictor candidates including spatial, temporal, and affine-based predictors, then selectively uses only the most suitable candidates for each block. This excessive generation of candidates followed by selective application ensures optimal compression while managing processing time through practical selection criteria.
3Measurement precision
If affine motion modeling with multiple control points is used, then motion field accuracy improves, but the complexity of motion estimation increases
Solution Approach 1:
The patent applies local quality by deriving motion vectors at different locations (top-left, top-right, bottom-left corners) using the affine motion model, then using these location-specific motion vectors as predictors for different regions of the block. Each control point motion vector is derived from local motion information, allowing the system to capture spatial variations in motion while maintaining overall coherence through the affine model.
Data Source
AI summary
Different implementations are described, particularly implementations for determining a set of predictor candidates for affine merge coding mode from neighboring blocks for motion compensation of a picture block based on a motion model. The motion model, may be, e.g., an affine model in a merge mode or AMVP mode for a video content encoder or decoder. The motion model, may be, e.g., an affine model based on top-left/top-right control point motion vectors or an affine model based on top-left/bottom-left control point motion vectors. Such affine model may be signaled by a flag. In an embodiment, predictor candidates are sorted in the set based on a criterion such as, e.g., a validity check or a vectors coherence cost. In an embodiment, a predictor candidate is selected from the set based on a motion model for each of the multiple predictor candidates, and may be based on a criterion such as, e.g., a rate distortion cost. The corresponding motion field is determined based on, e.g., one or more corresponding control point motion vectors for the block being encoded or decoded. The corresponding motion field of an embodiment identifies motion vectors used for prediction of sub-blocks of the block being encoded or decoded.