Research On Video Coding Optimization For Real-time Communication

With the development and popularization of high-definition and ultrahigh-definition video,the storage and transmission of massive video data has become a huge challenge in the industry.In order to further reduce the transmission pressure of video streaming media,new-generation video coding standards such as VP9 and HEVC have emerged.Because the VP9 standards is open source and freely licensed,it has received more support and promotion in the streaming media industry.Compared with its previous generation coding standard VP8,due to the introduction of a large number of new technologies,VP9 can improve nearly 50% coding performance of VP8.However,its coding complexity has also increased with the improvement of coding performance,limiting the application of the VP9 video coding standard in real-time communication scenarios.Reducing the encoding complexity becomes an urgent problem to be solved in the current real-time audio and video industry.Therefore,this paper optimizes the superblock partitioning process in VP9 encoder,which consumes a lot of coding complexity,and proposes two fast algorithms based on different improvement directions.Aiming at the problem of high coding complexity when superblock partitioning,this paper analyzes the characteristics of superblock partitioning in VP9,and builds a large data set,which contains a large number of superblock partitioning results.Through the statistical analysis of the data set,this paper abstracts the quad-tree partitioning process of the superblock into a three-layer three-class model SHRN,and predicts the superblock by building a hierarchical deep residual network.The encoder uses the predicted superblock partition structure to directly divide,which can skip a large number of unnecessary search processes and greatly reduce the coding complexity.The experimental results show that the fast algorithm proposed in this paper can reduce the coding time by 60% on the premise that the coding performance is basically unchanged.Aiming at the fact that the existing fast algorithms do not consider the characteristics of the human visual system,this paper proposes a superblock fast algorithm based on visual perception characteristics.This paper first systematically analyzes and summarizes the characteristics of the human visual system,and then proposes a multidimensional JND perception threshold model based on visual characteristics.According to the JND model,a perceptual changing factor with superblock partitioning can be calculated to pre-process the superblock partitioning depth,which can early terminate the superblock partitioning,and reduce a large amount of coding complexity.Moreover,by replacing the rate-distortion optimization process in the partitioning process with a fast algorithm based on SHRN,the coding complexity can be further reduced.The experimental results show that the superblock fast algorithm based on visual perception characteristics can reduce the coding time by more than 70% on the premise that the coding performance remains basically unchanged.