Design And Quality Analysis Of Deep Learning Coding Architecture Based On VVC

Video is not only an important part of human life,but also plays a key role in applications in various fields.Facing the urgent need for video compression efficiency,the next-generation video coding standards VVC and AVS3 have been born and achieved excellent compression benefits.On the other hand,the unique advantages of deep learning in various fields have made it widely used in video coding,such as inter-frame prediction,loop filtering,etc.In recent years,there is no lack of excellent end-to-end deep learning image compression algorithms which have achieved unprecedented results in image compression.The intra-frame prediction module in the video coding framework is to compress the current frame.The current frame can be regarded as a picture.In principle,this picture can be compressed in an end-to-end deep learning method,but there is currently no end-to-end deep learning image compression is directly applied to the precedent of video coding framework,so its impact on the efficiency and quality of video coding needs to be studied.Based on the replacement feasibility of intra-frame compression coding and end-to-end deep learning image compression coding in the video coding framework,this paper designs the use of end-to-end image compression algorithms to complete the intra-frame compression work in the video coding framework.The whole coding framework is integrated,this article will explore its implementation and practice,and analyze the coding benefits and quality of the new framework.This article will select the latest end-to-end deep learning image compression algorithm,including VAE image compression with hyperprior,auto-regressive hierarchical prior image compression,image compression algorithm based on context convolution network entropy model,and on this basis,according to the characteristics of each algorithm,the feasibility of completing intra-frame coding is studied,and the video coding framework is designed based on the VVC video coding standard,respectively,and according to the design of this article,the principle will be implemented on the VVC corresponding software VTM6.0In addition,this article will evaluate the quality of each proposed new video codec framework.After being implemented on VTM6.0,all videos in Class C,Class D,and Class E of HEVC test video sequences are selected for encoding and decoding,and PSNR alignment and bit rate alignment will be compared with the original VTM6.0.In addition,the results of all comparative experiments will be analyzed and summarized from an objective perspective and a subjective perspective.From an objective point of view,according to the internal data rules of the new framework and the comparison rules with VTM6.0,from a subjective point of view,it is analyzed by multiple observations and repeated comparisons.In summary,this paper proposes a new concept of using end-to-end deep learning image compression algorithm to achieve intra-frame compression in video coding,and proves the feasibility in rich experiments.A new type of video coding is designed for different image compression algorithms and the framework was implemented.A breakthrough from principle to practice was completed.Afterwards,the new framework was subjected to rich objective and subjective tests.The experimental results show that different image compression algorithms bring different benefits to the video encoding framework,and some even surpass VTM6.0 in subjective effects,proving that end-to-end deep learning image compression can be applied in the video encoding framework.