Template Matching With Flipping for Video Block Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems face challenges in efficiently compressing digital video signals, particularly in optimizing prediction blocks for improved compression efficiency.
Innovation Solution
Implementing template-based prediction methods that involve determining a template orientation for a current block through template matching, adjusting the prediction block based on the orientation, and performing template matching searches in multiple orientations to enhance compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template-based prediction with multiple orientations is implemented, then video coding efficiency is improved, but device complexity increases
Solution Approach 1:
The template matching process is segmented into multiple orientation-specific searches (upright, horizontally flipped, vertically flipped, diagonally flipped). Each orientation has its own search range and matching criteria, allowing the system to handle different spatial patterns separately and efficiently without overwhelming complexity in a single unified process.
Solution Approach 2:
The system dynamically selects which template orientations to search based on the current block characteristics and previously determined search ranges. The search ranges are adjusted adaptively based on matching results, allowing the system to optimize complexity for each specific coding situation rather than using a fixed complex approach for all cases.
2Measurement precision
If multiple template orientations are searched, then prediction accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary template matching searches in multiple orientations to identify candidate prediction blocks before final decoding. By pre-determining which orientations are most likely to yield accurate matches based on block characteristics, the system reduces the number of full processing cycles needed while maintaining high prediction accuracy.
Solution Approach 2:
The system performs template matching searches in all four orientations (upright, horizontal flip, vertical flip, diagonal flip) which is more than the minimum required for many cases. This excessive action ensures that no potentially optimal prediction is missed, and the system can then select the best match from these comprehensive searches, balancing thoroughness with practical time constraints.
3Productivity
If template matching search range is expanded, then compression efficiency is improved, but computational load increases
Solution Approach 1:
The search range for template matching is not uniformly applied across all regions. Instead, the system determines different search ranges based on the specific characteristics of each block and its surrounding context. Blocks with higher complexity or more distinct patterns receive larger search ranges, while simpler blocks use smaller ranges, optimizing computational load distribution.
Solution Approach 2:
The system changes the search range parameter dynamically based on block size, position, and previously identified matching patterns. By adjusting the search range as a variable parameter rather than using a fixed value, the system can maximize compression efficiency for each block while keeping overall computational load manageable through adaptive parameter selection.
Data Source
AI summary
Systems, methods, and instrumentalities are disclosed herein for the field of video encoding and decoding. In examples, a video decoder or encoder may determine that a template-based prediction is enabled for a current block. A prediction block and a template orientation for the current block may be determined based on template matching. The decoder or encoder may decode or encode the current block based on the prediction block and the template orientation. In examples, the prediction block may be adjusted (e.g., horizontally flipped, vertically flipped, diagonally flipped, and/or rotated) based on the determined template orientation and the current block may be decoded or encoded based on the adjusted (e.g., reoriented) prediction block.