Intra-Prediction Template Matching With Geometry-Transformed Reference Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression systems face challenges in efficiently utilizing template matching for intra-prediction due to inherent complexity, limiting the matching possibilities within the defined target and candidate reference block neighborhoods.
Innovation Solution
The method involves determining a candidate reference block neighborhood based on the availability of reconstructed samples and applying a geometry transform to these samples, comparing them with target template patches, and selecting the candidate reference block with the minimum cost to generate an intra-prediction block using the associated geometry transform.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template matching is applied to intra-prediction mode coding, then compression performance is improved, but device complexity increases
Solution Approach 1:
The patent divides the template matching process into distinct segments: target template patch extraction from the current block, candidate reference template patch extraction from neighboring blocks, and cost calculation through comparison. This segmentation allows for systematic optimization of each step while maintaining overall compression performance.
Solution Approach 2:
The patent applies partial template matching by using only the necessary neighborhood samples (target block neighborhood and candidate reference block neighborhood) rather than the entire block. This reduces the complexity of comparison while maintaining sufficient compression performance through selective matching of relevant regions.
2Device complexity
If the target block neighborhood and candidate reference block neighborhood are limited to 4 samples, then device complexity is reduced, but matching possibilities are limited
Solution Approach 1:
The patent introduces dynamic adjustment of the candidate reference block neighborhood based on the geometry transform type. When a geometry transform is applied to improve matching, the neighborhood definition adapts accordingly, allowing more flexible matching possibilities while maintaining manageable complexity through conditional logic.
Solution Approach 2:
The patent changes the parameters of the candidate reference block neighborhood by defining it according to the geometry transform type. This allows the neighborhood to be dynamically adjusted based on the specific transform being applied, expanding matching possibilities without requiring a fixed large search area.
3Productivity
If geometry transforms are applied to expand search possibilities, then compression efficiency is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary determination of the candidate reference block neighborhood based on the geometry transform type before actual template matching. This preliminary action allows the system to prepare the appropriate reference samples in advance, reducing the computational burden during the actual matching and compression process.
Solution Approach 2:
The patent uses copying of reconstructed samples to create the candidate reference template patch from the candidate reference block neighborhood. This copying approach is computationally simpler than performing full geometry transforms on all candidate blocks, maintaining compression efficiency while reducing computational complexity.
Data Source
AI summary
A method of intra-predicting a target block of a video picture based on template matching that compares a target template patch with a candidate reference template patch, the method including: determining a candidate reference block neighborhood according to target block neighborhood reconstructed samples availability and according to a geometry transform type; obtaining the candidate reference template patch by applying a geometry transform, identified by the geometry transform type, on the candidate reference block neighborhood; selecting a candidate reference block with a minimum cost obtained by comparing the candidate reference template patch and the target template patch; and generating an intra-prediction block for the target block by applying the geometry transform, identified by the geometry transform type associated with the selected candidate reference block, to the selected candidate reference block.