Adaptive Intra Prediction Templates for Position-Aware CCCM Chroma Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies fix the sets of reconstructed neighboring samples used for intra prediction, leading to suboptimal coding efficiency as they do not adapt to the location of the current coding unit within the picture.
Innovation Solution
Implement template selection in video coding, allowing the selection of different sets of reconstructed samples based on a comparison of a threshold and a quantity of reference samples, and applying a convolutional cross-component model (CCCM) mode to predict chroma samples using reconstructed luma samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed set of reconstructed neighboring samples is used for intra prediction, then the device complexity is reduced, but the coding efficiency deteriorates
Solution Approach 1:
The patent implements dynamic template selection where the template (set of reference samples) is chosen based on the location of the current coding unit within the picture. Different templates are selected for different CU positions, making the reference sample selection adaptive rather than fixed, thereby improving coding efficiency without significantly increasing complexity
Solution Approach 2:
The patent applies different templates for different locations of coding units within the picture. Each template is optimized for specific spatial positions, allowing the prediction to adapt to local characteristics of the video content at different positions, thus improving overall coding efficiency
2Productivity
If multiple templates are used for intra prediction, then the coding efficiency is improved, but the device complexity increases
Solution Approach 1:
The system dynamically selects from multiple templates based on the current coding unit's position and characteristics. The template selection is driven by position-based logic and convolutional neural network predictions, allowing the system to adapt to different content scenarios without requiring complex exhaustive searches, thus improving coding efficiency with controlled complexity increase
Solution Approach 2:
The patent changes the parameters of the prediction system by introducing multiple templates with different characteristics. Instead of using a single fixed template, the system varies the template selection based on position and content characteristics, allowing optimal parameters to be chosen for different coding scenarios, thereby improving coding efficiency
3Measurement precision
If template selection based on CNN predictions is implemented, then the prediction accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent uses convolutional neural networks to pre-compute and store template prediction information during an offline training phase. The trained model parameters are then used during actual video coding to guide template selection, reducing the computational burden during real-time encoding while maintaining high prediction accuracy
Solution Approach 2:
The patent introduces a convolutional neural network as an intermediary between the raw video data and the template selection process. The CNN processes the video content and generates predictions that guide template selection, acting as a smart mediator that improves prediction accuracy while managing computational complexity through learned patterns rather than exhaustive analysis
Data Source
AI summary
A method of coding video data comprises determining, based on a comparison of a threshold and a quantity of reference samples in a selected template, whether the selected template is allowed, wherein a mode index indicates which template from among a plurality of templates is the selected template, wherein each of the templates includes a different set of reconstructed samples that neighbor a current coding unit (CU) of a current picture of the video data; based on determining that the selected template is allowed and that a convolutional cross-component model (CCCM) mode is to be used, applying the CCCM mode to predict chroma samples of the current CU based on reconstructed luma samples of the current CU and the reference samples in the selected template; and; and encoding or decoding the current CU based on the predicted chroma samples of the current CU.