Research On Video Inter-component Predictive Coding Algorithm

To effectively compress video content,the universal video coding standard H.266/VVC(Versatile Video Coding,VVC)introduces the Cross-Component Linear Model(CCLM)prediction between components.The algorithm establishes a linear model based on the correlation between the components in the local area of the image,and predicts the chrominance components from the luminance components based on the linear model,which effectively makes up for the shortcomings of the existing chrominance prediction algorithms and greatly improves the chrominance coding performance.It should be noted that there is still room for improvement of the existing linear model used in CCLM,and a more efficient prediction model can be designed to further improve the coding performance.To address the above problems,on the one hand,this paper improves the CCLM algorithm by deeply studying the implementation principle of the CCLM algorithm and proposes two algorithms to optimize the coding performance;on the other hand,by analyzing the correlation between the components,we propose a chroma prediction algorithm based on a lightweight fully connected network to improve the coding performance further significantly.The specific research contents and innovation work of this article are as follows:1.In terms of improving the coding performance of CCLM technology,this paper presents an in-depth analysis of the linear model construction process and the reference pixel selection process of CCLM,and proposes two CCLM coding performance optimization algorithms:(1)A weighted least mean square error-based CCLM algorithm(W-CCLM))is proposed.This algorithm uses the luminance characteristics between the current coding block and the reference pixel to assign weights to the reference pixel and implements the weighted least mean square error method for linear model parameter derivation.The correlation of luminance between the selected reference sample and the coding block is used to improve the matching between the linear model and the content of the current coding block,thus effectively improving the prediction accuracy.(2)Based on(1),a W-CCLM algorithm based on different reference pixel sets is proposed.Based on the above algorithm,this paper proposes a W-CCLM algorithm based on different number of reference pixel sets by analyzing the relationship between the number of reference pixels and the prediction error.The algorithm increases the flexibility of the reference pixels involved in the prediction,which can further improve the compression efficiency.2.Aiming at the problem that the local area luminance and chrominance components are correlated but difficult to model,this paper proposes a neural network-based intercomponent prediction algorithm:(1)A lightweight neural network-based intercomponent prediction algorithm is designed.The algorithm selects the reference pixels with strong correlation based on the luminance difference between the pixels to be predicted and the reference pixels to form a reference subset,and then feeds the reference subset into a lightweight fully connected network to obtain the chromaticity prediction values.Experiments show that it can save 0.27%,1.54%and 1.84% of code rate in Y,Cb and Cr components,respectively,compared with the H.266/VVC test model version 10.0,which is an effective complement to the chromaticity prediction in H.266/VVC.(2)A data preprocessing module is designed for the flexible block partitioning mechanism in H.266/VVC.For any size of coding blocks,the data preprocessing module can select a reference subset of fixed size with strong correlation as the input of the unified network.This means that there is no need to design separate network structures for different size coding blocks,which greatly improves the generality of the proposed algorithm.(3)By analyzing the complexity of the proposed algorithm,a simplification scheme for the proposed algorithm operations is proposed.Specifically,the proposed algorithm is integrated into the VVC reference software by storing network weights and using C++ language,and the data pre-processing step is performed quickly by the dictionary method combined with the windowing method.The proposed algorithm achieves a significant reduction in complexity without degrading the coding performance and meets the demand for low complexity in video coding.