Research On CU Partitioning And Inter-frame Prediction Based On AVS3

With the advancement of communication network technology and the rapid development of various terminal devices,video has gradually become an important carrier for people to record and exchange information,and an important part of people’s daily lives.In this context,video technology has developed a variety of scenarios,such as 4K,8K,short video,surveillance and virtual reality,etc.,which urges the original video coding technology to have stronger compression and decompression capabilities.Faced with these problems,the AVS working group has finally formulated a new generation of audio and video codec standard AVS3 after several years.Compared with the previous generation AVS2,the bit rate has been reduced by about 30%.However,due to the huge computational complexity brought by the introduction of many new technologies in AVS3,the video compression time has also been greatly improved compared with the past.Therefore,based on the super compression level of AVS3,in order to reduce its encoding time,this paper mainly studies the relationship between DCT coefficient,image texture and block division,obtains an ideal fast block division algorithm in intra mode,and then explores the possible application of DCT coefficient in inter prediction mode.The specific work is as follows:1.First,research on the rules and processes of CU partitioning in the new generation of AVS3,the main points are two kinds of binary trees and two extended quad trees which are newly introduced in the previous generation of video coding standard.Then,the relationship between different basis function coefficients and image content after DCT is studied,and it is found that for images with relatively flat content,the coefficients in the first row and first column of the DCT coefficients matrix have a strong correlation with the texture direction of the images(especially horizontal and vertical images).In addition,the high-frequency part of the DCT coefficients also has an intuitive correspondence with the texture complexity of the image.Finally,according to these two researches,a fast block division algorithm of AVS3 based on DCT coefficients is proposed.According to the relationship between the original block division results and the image block texture,the DCT coefficients can be used to indicate the characteristics of the image texture.By reducing the block division decision.In this case,the amount of computation is reduced to reduce the coding time complexity.Applying the algorithm to the reference code,in AI mode,according to the experimental results,can reduce the coding time complexity at a lower loss of compression performance.2.This paper explores the research of DCT coefficients combined with historical decision information for inter prediction mode selection.Through the observation of the division results of coding inter prediction modes,we find the similarity between the prediction mode of the CU in the LCU and the image content in time and space,and use this similarity to reduce the decision mode of inter prediction,so as to reduce the selection The calculation amount of the mode is to achieve the purpose of reducing the time complexity.The algorithm is applied to the reference code.In LDP mode,according to the experimental results,the encoding time is reduced,and the BD-rate is improved very little.