Research On The Loop Filter And Rate Distortion Optimization Of Versatile Video Coding

In the era of "Big Data",there is a growing demand for video compression efficiency,which is mainly caused by two reasons: one is that videos are now available and can be viewed through a variety of display devices,which increasingly challenges the existing network channel capacity;another reason is that video users have higher and higher requirements for video viewing experience,which leads the needs of higher spatial and temporal resolution,higher dynamic range and wider color gamut.In other words,the video format which meets the requirements from users needs a higher data rate,which puts forward higher requirements for video coding efficiency.Therefore,a new generation of international video coding standard VVC(Versatile Video Coding)is being developed and is expected to be released officially in the summer of 2020.This thesis aims at the next-generation international video coding standard VVC,deeply studies in-loop filtering and rate-distortion optimization techniques and proposes improvements.The specific work is as follows:1.Aiming at in-loop filtering technology in VVC,a self-guided filtering-based loop filtering technology is proposed,which uses the local structure information of the reconstructed image to adjust the regularization parameters of the self-guided filter adaptively.We also divide the reconstructed images into non-overlapping regions,iteratively update the region-level subspace mapping,which makes the mapping result closer to the original image.The experiment is implemented based on VTM3.0 reference software,which improved the coding performance by 0.23% on average.At the same time,considering that deep learning has made breakthroughs in many fields,this thesis proposes an in–loop filtering algorithm based on CNN,using edge information of the reconstructed image to further improve the network's ability to recover image detail information.Experimental results show that compared with the VTM3.0 reference software,the algorithm can achieve 0.75% code rate savings with the same encoding quality.2.Aiming at the rate-distortion optimization technology in VVC,we have studied the problem of temporal dependent rate-distortion optimization under Low-Delay structure from the perspective of distortion.Firstly,a temporal distortion propagation chain is established based on the relationship of temporal references in the Low-Delay structure by using forward motion vector search.Based on the analysis of the distortion propagation probability of skip mode and inter mode,the temporal distortion dependent rate-distortion optimization problem is reintroduced.Secondly,we calculate the propagation factor by estimating the distortion of all subsequent coding units affected by the current coding unit.The CTU-level Lagrangian multipliers and the quantization parameters can be adjusted by the propagation factor adaptively.At the same time,the I frame is subjected to two-pass coding,the I frame distortion obtained from the first-pass coding is used to estimate the impact of the I frame on all subsequent B frames,then the QP of I frame can be adjusted adaptively.The experiment is implemented based on VTM5.0 reference software,which improved the encoding performance by 2.57% on average.