Research On Multi-Type Tree Partition Algorithm For Mixed Screen Content Video In VVA

With the rapid application of online education,video conferencing,screen sharing,etc,the use of screen content video is becoming more and more popular.On the one hand,the screen content video has special characteristics that are different from the natural content video,on the other hand,the partitioning structure of Quad-Tree plus Multi-type Tree(QTMT)introduced in the Versatile Video Coding(VVC)standard greatly increases the coding time.In this thesis,based on the characteristics of screen content video,the problem of reducing the coding complexity for multi-type tree partitioning and intra coding mode decision has been studied.The main contents of the work are as follows.(1)To reduce the computational complexity of multi-type tree partitioning in H.266/VVC,a fast decision algorithm is proposed based on the characteristics of screen content video.Firstly,using the degree of activity of horizontal and vertical direction in current CU,the traversal necessity of multi-tree division for blank area of screen content is determined.Secondly,by statistical analysis of the partition result of the border area and text area of screen content,it is found that the optimal partition mode for these kinds of screen content area shows certain regularity in selection of the horizontal and vertical directions.Based on this observation,a new metric named the amplitude of the change in the difference in brightness of equidistant sub-blocks(ESDA)is defined,by comparing the amplitude of change in the horizontal and vertical brightness values,the direction of partition is determined in advance.Finally,for other kinds of areas of screen content video,the average absolute error of brightness is used to reduce the number of candidate multi-tree partition modes.The experimental results show that,compared with the reference software VTM10.0,the algorithm saves 39.16% of coding time on average and the BD-rate only increases by1.45% under the AI configuration.(2)Furthermore,in the process of pattern selection for intra-frame prediction of screen content video,the behavior of pattern selection in different regions of screen content video is researched for the problem of high computational complexity caused by many candidate patterns.Firstly,according to the horizontal/vertical activity,ESDA in Algorithm(1),determine whether the current block belongs to one of blank area,fade area.Secondly,determine whether the current block belongs to the text area based on the high gradient pixel ratio.Finally,according to the characteristics of the above three types of video areas,targeted IBC,PLT,DC,Planar and 65 angular modes of advance decision.The experimental results show that,compared with the reference software VTM10.0,the algorithm saves 21.91% of coding time on average and the BD-rate only increases by 1.39% under the AI configuration.