| Video data is characterized by high dimensionality,large traffic,difficult compres-sion,and limited transmission bandwidth,especially in the 5G era,which poses challenges to efficient and high-performance coding.How to promote the transmission quality of pic-tures in the process of high-definition video coding and ensure efficient compression effi-ciency,how to balance the bit-rate and distortion through the rate-distortion optimization technology as much as possible to choose a better prediction mode so as to obtain better coding performance,all are the key issues of the study in the field of video coding.Driven by these problems,this paper analyzes and establishes a temporal-domain redundancy model,and provides a picture-level and LCU(Largest Coding Unit)-level rate-distortion optimization strategies,respectively.The specific work is divided into three aspects.(1)Aiming at the temporal redundancy which accounts for the largest proportion in coding,a temporal redundancy model is proposed and established to predict the infor-mation entropy of uncoded LCUs.Then the information entropy generated by coding of benchmark test sequences is analyzed and counted.The motion vector and residual de-composition are performed on the source pictures,then the prediction complexity infor-mation of each coding LCU is obtained,and the linear mapping between the information entropy and the complexity of the block is established.The LCU-level and picture-level temporal redundancy models,LTDR-Model and PTDR-Model,are respectively proposed.Those models are applied to predict the size of information entropy before coding,pro-viding a model basis for bit redistribution and Rate-distortion optimization.(2)Under the temporal-domain hierarchical reference structure,an algorithm of picture-level coding parameter adjustment is proposed.According to the reference relationships and reference probability of pictures in different layers,the reference value and informa-tion entropy predicted of the pictures in GOP are analyzed to obtain the reference energy factors.Thus the Lagrange multiplier of the reference pictures are adjusted by simultane-ous energy factors and hierarchical values,and the corresponding quantization parameters are updated by feedback.On the random access structure configuration,the test results are compared with the AVS3 reference software HPM,where an average BDBR performance improvement of 0.79%and a maximum BDBR gain of 2.03%are obtained.(3)According to strong correlation properties of adj acent bits of LCUs in the temporal-domain,the adaptive adjustment of the LCU-level for optimization parameters is realized.Collecting all the distortions and bit costs of the current picture,this paper discovers that there is a Markov property between adjacent LCUs in the temporal-domain.Furthermore,bit-rate of current LCU and average bit-rate of current picture are derived from the neigh-boring LCU to find a better bit allocation for current LCU.The rate-distortion model is used to allocate better Lagrangian multiplier for each LCU.For AVS3 standard,the exper-imental tests on the low-delay structure configuration possesses an average improvement of BDBR performance by 0.23%compared with the benchmark configuration.And there is a maximum BDBR gain of 2.03%.The three aspects of work have contributed a series of video coding optimization tech-nologies to AVS3,which have achieved improvements in subjective and objective picture quality.Particularly,these methods significantly improved the overall performance of the latest Chinese audio and video standard.Related work has submitted technology proposals to the AVS working group. |
PDF Full Text Request