Template-Matched MRL Intra Prediction for Efficient Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face inefficiencies in intra prediction due to the use of less likely prediction directions requiring more bits, which increases redundancy and reduces compression efficiency.
Innovation Solution
Implementing template matching based multiple reference line (MRL) intra prediction mode and template-based intra mode derivation (TIMD) to determine the most likely intra prediction directions by comparing cost values across reference regions, allowing for more efficient reconstruction of video data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional intra prediction with fixed prediction directions is used, then the coding structure is simple, but more bits are required to represent less likely prediction directions, reducing compression efficiency
Solution Approach 1:
The patent implements dynamic intra prediction mode selection where the prediction mode is adaptively chosen based on template matching results. The system dynamically switches between different prediction modes (planar, DC, angular) and adjusts prediction directions based on the actual content characteristics of the template region, rather than using fixed prediction directions. This resolves the contradiction by making the coding structure flexible enough to handle different content types efficiently while maintaining simplicity through automated mode selection.
Solution Approach 2:
The patent changes the prediction parameters (mode and direction) based on template matching costs. When template matching indicates a strong correlation, the system uses fewer bits to signal the prediction mode. The prediction angle and reference line selection are adjusted according to the template region characteristics, allowing optimal parameter selection that reduces bit consumption for less likely directions while maintaining accuracy.
2Measurement precision
If more prediction directions are considered to improve accuracy, then reconstruction quality improves, but the number of bits required to signal the prediction mode increases
Solution Approach 1:
The patent performs preliminary template matching and cost calculation before finalizing the prediction mode selection. By pre-evaluating the template region against reference regions and calculating matching costs in advance, the system can determine the most likely prediction direction beforehand. This preliminary action allows the system to signal only the necessary mode information with fewer bits, as the template matching results guide the selection of the optimal prediction parameters before encoding.
Solution Approach 2:
The patent uses template matching cost as feedback to adaptively select the prediction mode. The cost calculation provides feedback on which prediction directions are most suitable for the current block, allowing the system to concentrate bits on the most likely directions while using fewer bits for less likely ones. This feedback mechanism ensures high prediction accuracy is achieved with optimized bit allocation rather than uniformly high precision for all directions.
3Productivity
If template matching based MRL intra prediction mode is implemented, then compression efficiency improves, but the computational complexity increases
Solution Approach 1:
The patent segments the prediction process into distinct stages: template region identification, reference region selection, cost calculation, and mode determination. By dividing the computational task into these segmented stages, the system can optimize each stage independently and reuse intermediate results (such as template matching costs) across multiple prediction modes, thereby reducing overall computational complexity while maintaining high compression efficiency.
Solution Approach 2:
The patent performs preliminary template matching and reference region identification before the actual prediction mode selection. This preliminary action computes the template costs once and reuses them for evaluating multiple prediction modes, avoiding redundant calculations. The reference regions are pre-identified based on template matching results, which streamlines the subsequent mode selection process and reduces the overall computational burden despite the enhanced prediction capability.
Data Source
AI summary
In a method of decoding, coded information of a coding unit (CU), a template region, and a plurality of reference regions is received. The coded information includes a first syntax element indicating whether the CU is predicted based on a template matching based multiple reference line (MRL) intra prediction mode. In response to the first syntax element indicating that the CU is predicted based on the template matching based multiple reference line (MRL) intra prediction mode, a plurality of cost values is determined between (i) respective prediction samples of the template region based on samples in each of the plurality of reference regions, and (ii) reconstructed samples of the template region corresponding to the respective prediction samples. A reference region from the plurality of reference regions is determined based on the plurality of cost values. Samples of the CU are reconstructed based on the samples in the determined reference region.