Cross-Component Colour Prediction with Reference Pixel Subsets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for predicting color components in video coding, particularly in the VTM of H.266/VVC, suffer from high computational complexity and memory bandwidth due to the large number of samples used for model parameter derivation, which can lead to inaccurate prediction models, especially when nonlinear or multi-models are employed.
Innovation Solution
A method is introduced to reduce the number of adjacent reference pixels by constructing a subset of these pixels, using them to calculate model parameters for N prediction sub-models, where N is greater than or equal to 2, thereby improving prediction accuracy and reducing computational complexity and memory bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of adjacent reference pixels are used for model parameter derivation, then the prediction model can capture more comprehensive color component variations, but the computational complexity and memory bandwidth requirements increase significantly
Solution Approach 1:
The patent extracts only the necessary subset of reference pixels from the complete set of adjacent reference pixels. By identifying and removing unimportant or abnormal pixels, the method retains only those pixels that contribute meaningfully to prediction accuracy, thereby reducing computational complexity while preserving essential prediction information
Solution Approach 2:
The patent changes the parameter of reference pixel selection from using all adjacent pixels to using a filtered subset based on specific criteria. This parameter change in the selection process allows the system to achieve comparable or better prediction accuracy with fewer pixels, thus reducing computational burden
2Reliability
If a large number of adjacent reference pixels are used for model parameter derivation, then more complete color information can be obtained, but the memory bandwidth consumption increases
Solution Approach 1:
The patent extracts a subset of reference pixels that are most relevant for reliable prediction, removing pixels that do not contribute to model reliability. This extraction process reduces the quantity of pixel data that needs to be accessed from memory, thereby reducing memory bandwidth consumption while maintaining prediction reliability
3Measurement precision
If all adjacent reference pixels are used for prediction, then comprehensive color component data is available, but abnormal samples can degrade prediction accuracy
Solution Approach 1:
The patent converts the potential harm of abnormal pixels into a benefit by using their absence in the subset. By deliberately excluding pixels that may be abnormal or unimportant, the method ensures that only reliable pixels contribute to prediction, thereby improving accuracy while avoiding the degrading effect of abnormal samples
4Measurement precision
If multiple prediction sub-models are constructed, then better prediction accuracy can be achieved through model diversity, but the computational complexity increases
Solution Approach 1:
The patent segments the prediction task into multiple prediction sub-models, each handling different aspects or regions of the color component prediction. By dividing the overall prediction problem into smaller sub-problems, the system can achieve better accuracy through model diversity while managing computational complexity through organized segmentation
Data Source
AI summary
Disclosed are a colour component prediction method, an encoder, a decoder, and a storage medium. The method includes: determining adjacent reference pixels of a current block in a picture; constructing a subset of adjacent reference pixels according to the adjacent reference pixels, wherein the subset of adjacent reference pixels contains a part of the adjacent reference pixels; and calculating model parameters of a prediction model according to the subset of adjacent reference pixels, wherein the prediction model includes N prediction sub-models, the N prediction sub-models correspond to N groups of model parameters, and the prediction sub-models are used to perform, through corresponding model parameters, cross-component prediction of colour components to be predicted.