TIMD Merge Prediction Using Adaptive Weights and Block Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing template-based intra prediction modes in video encoding and decoding are sub-optimal due to the lack of strong correlation between neighboring blocks and the current block content, leading to inefficient prediction and increased overhead in signaling.
Innovation Solution
Recomputing blending weights based on template distortions, using block vectors from neighboring blocks, and partitioning templates into sub-templates to improve intra-prediction accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If template-based intra mode derivation is used to derive prediction modes from neighboring blocks, then the prediction process is simplified and overhead is reduced, but the prediction accuracy deteriorates due to lack of strong correlation between neighboring blocks and current block content
Solution Approach 1:
The patent dynamically adapts the prediction process by computing blending weights based on template distortions specific to each current block. Instead of using fixed template-based modes from neighboring blocks, the system dynamically evaluates multiple prediction modes and assigns weights based on their performance for the specific current block, making the prediction process adaptive rather than static.
Solution Approach 2:
The patent changes the parameters of the prediction process by computing blending weights based on template distortions. The distortion values serve as parameters that determine the optimal combination of prediction modes. By adjusting the blending weights according to computed distortions, the system optimizes prediction accuracy for each block while maintaining a simplified overall process.
2Measurement precision
If blending weights are computed based on template distortions for each current block, then prediction accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent applies partial action by computing blending weights only when template-based prediction is used, rather than for all prediction modes. The system selectively applies the distortion-based weighting approach to specific cases where it provides benefit, avoiding unnecessary computational overhead in other scenarios.
Solution Approach 2:
The system performs self-service by using the template distortion computations that are already necessary for mode selection to also derive the blending weights. The same template matching process that identifies the best prediction modes also provides the distortion values needed for weight computation, eliminating the need for separate weight calculation procedures.
3Measurement precision
If multiple prediction modes are blended to improve accuracy, then prediction quality increases, but the overhead for signaling and processing increases
Solution Approach 1:
The patent merges the mode selection and weight determination processes into a unified framework. By combining multiple prediction modes with dynamically computed blending weights, the system achieves high prediction quality while signaling only the final blended result rather than separate mode indicators, reducing overall signaling overhead.
Data Source
AI summary
Multiple techniques are presented for template-based intra mode derivation merge mode improvements. Three main examples are outlined: (I) Recomputing blending weights of TIMD merge candidates; (II) Using block vectors from neighboring blocks to replace one or more TIMD Merge predictors; and (III) Partitioning a template into sub-templates.