Template-Matching Motion Vector Refinement With Partial Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The conventional template matching (TM) process in video coding systems experiences significant latency due to the need to access fully reconstructed neighboring blocks for motion vector refinement, which hampers processing efficiency.
Innovation Solution
The proposed methods utilize predefined regions, parent nodes, adaptive template selection, and the use of prediction samples to generate templates for motion vector refinement, reducing latency by using partially reconstructed or predicted samples and allowing for flexible template combinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fully reconstructed neighboring blocks are used for template matching, then motion vector refinement accuracy is improved, but processing latency increases
Solution Approach 1:
The patent performs preliminary reconstruction of neighboring blocks before template matching operations. By reconstructing blocks in advance and making them available in a buffer, the system eliminates waiting time during template matching while maintaining accuracy. The reconstruction is done beforehand so that when template matching needs to access neighboring block samples, they are already ready in memory.
Solution Approach 2:
The patent divides the video picture into multiple blocks and processes them in a specific scanning order (e.g., raster scan). By segmenting the processing into discrete block units and managing their reconstruction and availability independently, the system can optimize the timing of reconstruction operations for each block, reducing overall latency while maintaining the accuracy needed for template matching.
2Measurement precision
If neighboring blocks are reconstructed before template matching, then template matching accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent employs in-loop filtering that automatically processes reconstructed blocks to improve their quality. The filtering operations (deblocking, SAO, ALF) are applied automatically during the reconstruction process itself, improving template matching accuracy without requiring additional separate processing stages. The system serves its own need for high-quality reference blocks through built-in filtering capabilities.
Solution Approach 2:
The patent integrates multiple functions into the reconstruction process: blocking compensation, in-loop filtering, and template preparation all occur within the same processing pipeline. The reference picture buffer serves both as a storage mechanism and as an active participant in the filtering and reconstruction processes, reducing the need for separate dedicated structures for each function.
Data Source
AI summary
A method and apparatus for video coding system that utilizes low-latency template-matching motion-vector refinement are disclosed. According to this method, a current template for a current block is determined, where the current template includes an inside current template including inside prediction samples or inside partially reconstructed samples inside the current block. The inside partially reconstructed samples are derived by adding a DC value of the current block to the inside prediction samples. Corresponding candidate reference templates associated with the current block are determined at a set of candidate locations. A location of a target reference template among the candidate reference templates that achieves a best match between the current template and the candidate reference templates is determined. An initial motion vector (MV) is then refined according to the location of the target reference template.


