Affine Motion Template Matching for Control Point Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently handling complex motions such as zoom, rotation, and perspective transformations, leading to suboptimal compression and reconstruction quality in affine motion models.
Innovation Solution
Implementing template matching-based motion refinement for affine coded blocks, which involves determining candidate reference templates and selecting the best match based on template matching cost to refine motion vectors, using decoder-side motion vector refinement techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template matching-based motion refinement is implemented for affine coded blocks, then motion compensation accuracy is improved, but computational complexity increases
Solution Approach 1:
The current block is divided into multiple affine subblocks, and template matching is performed separately for each subblock rather than for the entire block. This segmentation reduces the computational complexity of template matching while maintaining motion compensation accuracy for different regions with different motion characteristics
Solution Approach 2:
Different template matching refinement processes are applied to different affine subblocks based on their local motion characteristics. Each subblock can have its own motion vector refinement, allowing localized optimization of motion compensation accuracy without uniformly increasing complexity across the entire block
2Measurement precision
If multiple candidate reference templates are evaluated for each control point, then motion vector accuracy is improved, but processing time increases
Solution Approach 1:
Candidate reference templates are pre-identified and prepared before the actual template matching process. The search range and candidate templates are determined in advance based on initial motion vectors, reducing the real-time processing burden during motion compensation
Solution Approach 2:
Instead of exhaustively searching all possible reference templates, a limited number of candidate reference templates are evaluated. This partial action approach provides sufficient motion vector accuracy while significantly reducing processing time compared to exhaustive search methods
Data Source
AI summary
A current block is coded by an affine mode and includes a first control point at a first corner of the current block. A current template associated with the first control point is determined. A plurality of candidate reference templates is determined in a reference picture for the current template. A reference template is selected from the plurality of candidate reference templates for the current template based on a template matching (TM) cost. The TM cost indicates a respective difference between each candidate reference template and the current template of the first control point. A first control point motion vector (CPMV) is determined based on the selected reference template, where the first CPMV indicates an offset between the selected reference template in the reference picture and the current template associated with the first control point. The current block is reconstructed based at least on the first CPMV.