Research On Video Coding Optimization Algorithm For SCC And VVC

With the rapid development of digital video capture and display technologies,videos has gradually replaced text and images as an important medium for daily information and social networking,which also makes video contents and applications more abundant.Screen content video produced by desktop sharing,distance education and other applications has its unique content characteristics,which are different from traditional natural content video.Meanwhile,the resolution,frame rate and dynamic range of video become larger and larger due to the demand for video reality.The unique screen content videos and increasing data valume makes the coding efficiency of HEVC can not satisfy the demand of video storage and transmission.In order to improve the compression efficiency of screen content video,MPEG and VCEG propose the HEVC screen video coding SCC(Screen Content Coding)extension,which adopts Intra Block Copy and Palette mode for screen content to improve its coding efficiency.Similarly,to obtain more compression gain for Ultra-High-Definition Videos,MPEG and VCEG develope the VVC coding standard,which introduces the more flexible QTMT(Quad Tree plus Multi-type Tree)CU partition structure and many other coding technologies.It is reported that VVC achieves 40% BD-rate reduction compared to HEVC at the same coding quality.However,the extended prediction modes and more flexible block CU partition structure also dynamically increase the complexity of the encoder,which seriously limits the practical application of SCC and VVC,especilally in weak computing terminals and real-time scenarios.Therefore,this paper conductes an in-depth exploration and research on the encoder complexity optimization for SCC and VVC.The main contents and contributions are as follows:(1)To reduce the complexity of SCC intra-coding complexity,this paper explores the relationship among the video contents,coding mode and coding size,and proposes a fast screen screen video coding algorithm based on machine learning.In particular,prediction mode selection is modeled as a video content classification problem,and quadtree partition problem is modeled as a binary classification problem.By considering the features of the screen video texture,such as limited noise,distinct color number and shape edges,decision tree classifiers are trained to early select coding unit size and prediction mode.Experimental results show that the algorithm achieves a 45% complexity reduction with negligible encoding performance loss.(2)To reduce the complexity of VVC intra-coding,this paper first analyses the characteristics of the QTMT partition structure and the optimal intra-prediction mode distribution.Then,cascading partition decision framework is proposed to solve the QTMT partition mode selection problem.Based on the more effective local texture description features,data-driven statistical learning method is used to purn the complex QTMT partition tree structure.In addition,one-dimensional gradient descent search algorithm based on Hadamard cost is proposed to reduce the number of VVC intra prediction modes.The initial search point and search steps are also explored to improve the coding efficiency.Experimental results show that the algorithm has achieved a 62% complexity reduction,which is significantly outperform previous algorithms.(3)To reduce the high computational complexity of inter-frame coding in VVC,a fast partition mode selection algorithm based on multi-level decision framework is proposed by analyzing the complexity distribution of inter prediction modes.Specifically,the QTMT partition mode selection problem is decomposed into five independent sub classification problems through multi-level decision framework.Considering the influence of temporal layer on inter frame block size,the encoded information of AF＿Merge mode,motion and texture information,context information and prediction error information are used to build multiple partition decision models for different temporal layer to early predict the partition mode of QTMT.In addition,in order to reduce the impact of the misclassification of the classifier on the coding performance,the coding performance protection mthod based on soft decision mechanism is proposed.