Blending-Mask IntraTMP Fusion for Spatial Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently utilizing intra template matching prediction to reduce redundancy and improve compression efficiency, particularly in handling spatial correlations within video frames.
Innovation Solution
The implementation of intra template matching prediction (intraTMP) using blending masks, which involves determining candidate prediction blocks, applying weighting matrices to these blocks, and fusing them through a sum of weighted prediction blocks, with filtering based on specific format rules to enhance prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If intra template matching prediction is applied to compress video data, then compression efficiency is improved, but prediction accuracy and handling of spatial correlations deteriorate
Solution Approach 1:
The patent combines multiple candidate prediction blocks into a single fused prediction block by computing a weighted sum of the candidates. This merging approach integrates information from multiple sources to produce a more accurate prediction, resolving the contradiction between compression efficiency and prediction accuracy by maintaining high fidelity while still achieving compression through selective candidate usage.
Solution Approach 2:
The patent introduces blending weights as adjustable parameters that control the contribution of each candidate prediction block to the final fused prediction. By optimizing these weight parameters, the system can adaptively balance between using simpler prediction methods (for compression) and more complex fused predictions (for accuracy), thus resolving the technical contradiction.
2Measurement precision
If multiple candidate prediction blocks are fused with weighting matrices, then prediction accuracy is improved, but computational complexity and processing overhead increase
Solution Approach 1:
The patent applies different blending weights to different regions or positions within the prediction block, allowing locally optimized prediction quality. This local quality approach improves overall prediction accuracy without requiring uniformly high complexity across the entire block, as simpler regions can use fewer candidates or simpler weighting schemes.
Solution Approach 2:
The system computes a fused prediction block using multiple candidates and weighting matrices, which represents an excessive action compared to simple single-candidate prediction. However, this partial fusion approach (using a limited number of carefully selected candidates rather than all possible candidates) achieves sufficient accuracy improvement while controlling the increase in processing complexity.
3Reliability
If blending masks and weighting matrices are applied to candidate blocks, then spatial correlation utilization is improved, but encoding/decoding overhead increases
Solution Approach 1:
The patent performs preliminary selection of candidate prediction blocks based on spatial correlation metrics before the actual fusion process. By pre-identifying and ranking candidates according to their spatial correlation with the current block, the system can utilize strong spatial correlations effectively while limiting the number of candidates that require full fusion processing, thus reducing bitstream overhead.
Solution Approach 2:
The patent uses reference blocks from previously decoded regions as candidate prediction blocks, effectively copying spatial patterns that have already been processed. This copying approach leverages spatial correlation without requiring complete re-processing of reference data, as the reference blocks are already available in the decoded picture buffer, thereby minimizing additional overhead.
Data Source
AI summary
An apparatus includes processing circuitry. The processing circuitry is configured to determine a plurality of candidate prediction blocks for a current block based on a cost value between a template of each of the plurality of candidate prediction blocks and a template of the current block according to intraTMP. The processing circuitry is configured to determine a plurality of masks for the plurality of candidate prediction blocks. The respective mask for each of the plurality of candidate prediction blocks includes a respective weighting matrix with a plurality of weighting values. The processing circuitry is configured to reconstruct the current block based on a fused prediction block that is a sum of weighted prediction blocks. Each weighted prediction block is equal to a product of the mask of a respective candidate prediction block of the plurality of candidate prediction blocks and the respective candidate prediction block.