Adaptive Neighbor Templates for TIMD Intra Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current implementations of Template-based Intra Mode Derivation (TIMD) tools in video coding do not fully consider the features of the target block, leading to suboptimal intra prediction efficiency.
Innovation Solution
A method for determining a predictor block by minimizing distortion between a neighbor template and a reference array of samples, using a neighbor template size based on the target block dimensions, and refining initial intra prediction modes to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current TIMD tool implementations are used, then decoder complexity is kept at minimum, but intra prediction efficiency is suboptimal due to not considering target block features
Solution Approach 1:
The patent changes parameters of the neighbor template (size, shape, orientation) based on the target block dimensions and characteristics. Specifically, the neighbor template width is set to one-quarter of the target block width, and height is set to one-quarter of the target block height, creating an adaptive template that improves prediction efficiency without requiring complex decoder operations
Solution Approach 2:
The neighbor template is derived automatically from already decoded causal samples in the reconstructed neighborhood, allowing the system to self-generate prediction data without external intervention or complex processing. The template-based derivation inherently adapts to the target block features using only previously decoded information
2Measurement precision
If a fixed-size neighbor template is used in TIMD, then processing is simplified, but prediction accuracy is reduced for blocks of varying dimensions
Solution Approach 1:
The neighbor template dimensions are made dynamic and adaptive rather than fixed. The template width is dynamically set to one-quarter of the target block width, and the height is dynamically set to one-quarter of the target block height, allowing the template to automatically adapt to blocks of varying dimensions while maintaining simple processing rules
Solution Approach 2:
The patent applies different template sizes and orientations based on the local characteristics of each target block. Each block receives a customized neighbor template derived from its specific reconstructed neighborhood, ensuring optimal prediction accuracy for local variations in block dimensions and content characteristics
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for encoding comprising: obtaining (700) a target block; obtaining a neighbor template of samples in a reconstructed neighborhood of the target block, a width of the neighbor template being based (701, 702, 703) on a height of the target block and a height of the neighbor template being based (704, 706, 705) on a width of the target block; determining a predictor block for the target block based on a minimization of a distortion between a prediction of the neighbor template from a reference array of samples in a reconstructed neighborhood of the neighbor template and the neighbor template.