Intra Prediction Blending for Template Mismatch in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video codecs often result in sub-optimal intra-prediction due to the mismatch between the template region and the current block content, leading to inefficient signaling of intra-prediction modes.
Innovation Solution
Implement template-based intra mode derivation (TIMD) techniques that utilize reconstructed samples in the surrounding area of the current block to search for optimal intra-prediction modes, dividing the area into multiple templates and combining predictors using sample-based blending.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional intra-prediction modes are used with template region, then the signaling of intra-prediction modes is simplified, but the prediction accuracy deteriorates due to mismatch between template region and current block content
Solution Approach 1:
The current block is divided into multiple sub-blocks, and different intra-prediction modes are applied to different sub-blocks based on their local characteristics. This segmentation allows each sub-block to use the most appropriate prediction mode, improving overall prediction accuracy while maintaining efficient signaling through mode partitioning.
Solution Approach 2:
Different intra-prediction modes are selected for different regions (sub-blocks) of the current block based on local content characteristics. The encoder evaluates various prediction modes for each sub-block and selects the mode that best fits the local patterns, ensuring local optimization of prediction quality rather than applying a single global mode.
2Measurement precision
If multiple intra-prediction modes are searched and signaled for each block, then the prediction accuracy is improved, but the signaling overhead and complexity increase
Solution Approach 1:
Instead of searching for multiple modes within a single block, the block is divided into sub-blocks and only one mode is selected per sub-block. This reduces the complexity of mode selection while maintaining accuracy benefits through localized mode adaptation.
Solution Approach 2:
The full intra-prediction mode search is applied only to selected sub-blocks rather than the entire block. This partial application reduces computational complexity and signaling overhead while still capturing the essential benefits of adaptive mode selection in regions where it provides the most value.
Data Source
AI summary
There is disclosed a method for location dependent template-based intra mode derivation. In accordance with an embodiment, encoded video data is decoded to obtain reconstructed samples of a picture; the reconstructed samples are used to intra-predict a current block of the picture; an area in a surrounding of the current block is divided (701) into at least two templates formed of the already reconstructed samples at specific locations with respect to the current block; for each template one or more intra-prediction modes are determined (702) by template-based intra mode derivation; at least two predictors of the current block are computed (705) based on the determined intra-prediction modes; and the at least two predictors are combined together using blending by sample-based weights to form (705) a prediction block for the current block.