Unified Cross-Component Convolution Model for Chroma Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards like HEVC and VVC face challenges in efficiently utilizing inter-channel dependencies for predicting chroma components from luma samples, leading to suboptimal compression performance.
Innovation Solution
A unified convolution model is developed that is configurable for different coding tools, using a higher-degree model with a k-tap spatial term, non-linear term, and bias term to predict chroma samples from luma samples, with parameters derived through a unified solver for matrix equations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a unified convolution model with higher-degree terms is used to predict chroma samples from luma samples, then cross-component prediction accuracy is improved, but model complexity and computational load increase
Solution Approach 1:
The unified convolution model is segmented into distinct components: spatial terms (k-tap filters), non-linear terms, and bias terms. This segmentation allows the model to process different aspects of the luma-chroma relationship separately, improving prediction accuracy while maintaining manageable complexity through modular computation.
Solution Approach 2:
The model uses configurable parameters including the degree of spatial terms (k-tap), the order of non-linear terms, and bias terms. By adjusting these parameters, the model can adapt its complexity to match the specific coding tool and content characteristics, optimizing the balance between prediction accuracy and computational load.
2Adaptability or versatility
If a unified convolution model is used to support multiple coding tools, then adaptability is improved, but device complexity increases
Solution Approach 1:
The unified convolution model is designed with a universal architecture that can support multiple coding tools (e.g., various prediction modes and transform types) through a single model structure. The model uses configurable parameters and a unified solver that works across different coding scenarios, eliminating the need for separate models for each tool and reducing overall device complexity.
3Stability of the object's composition
If regression data is selected from a unified range for multiple coding tools, then data consistency is improved, but memory requirements increase
Solution Approach 1:
The regression data from the unified range is prepared and organized in advance before the actual prediction process. By pre-processing and structuring the data consistently across different coding tools, the model ensures data stability and consistency while reducing the computational overhead and memory access requirements during real-time encoding or decoding operations.
Data Source
AI summary
The video coder receives a selection of a coding tool among a plurality of coding tools. The video coder specifies a set of regression data based on the selection of the coding tool. The specified set of regression data is selected from component samples within or neighboring a reference block or a current block. The video coder specifies a configuration of a convolution model according to the selection of the coding tool. The convolution model is configurable to support each of the plurality of coding tools. The video coder derives parameters of the convolution model by applying the set of regression data according to the specified configuration. The video coder applies the convolution model with the generated parameters to a set of reference component samples to obtain a set of predictor component samples. The video coder encodes or decodes the current block by using the predictor component samples.