Research On Fast Intra Coding And Parallel Decoding For H.265/HEVC

A fundamental figure of merit for a video coding design is its coding efficiency. Because any improvement of coding efficiency will bring a great leap forward in video service. The High Efficiency Video Coding(HEVC) Standard was officially released in2013by the Joint Collaborative Team on Video Coding(JCT-VC). Compared with the premier video coding standard, HEVC reduce about50%bitrate with the negli-gible video quality loss. However its computational complexity is increased by2to4times. Higher computational complexity makes it impossible for HEVC achieving commercialization, unless the computational complexity is greatly reduced or HEVC is accelerated by parallelization.The increase of computational complexity is mainly caused by more complex cod-ing unit(CU) splitting algorithm. To reduce computational complexity, a fast Intra CU splitting algorithm is proposed. The fast algorithm is divided into three steps. First texture features are used to describe the characteristics of CU’s texture. Second, the decision functions are obtained by training texture features. At last, based on the deci-sion functions, CU splitting depth range is narrowed by skipping unnecessary splitting and early terminating splitting. The experimental results show that the fast CU split-ting algorithm obviously improve the coding efficiency with negligible loss of video quality. For HEVC’s parallel framework, three parallel proposals are introduced. After that Wavefront Pararllel Processing(WPP) is implemented and experiment results are analyzed.The main works of this paper are listed as follows.(1)First, based on the theory of intra angular prediction,3x3pixels block texture feature is defined. According to the statistic characteristics of texture features, strong textures, weak textures and flat textures features are extracted from CU. These multi-texture features reflect the complexity of CU’s texture. Second, Variance is employed to describe the distribution of CU’s texture. At last correlation coefficient is used to reflect the correlation between texture features and CU splitting depth.(2)In order to get the decision functions, support vector machine(SVM) is em-ployed to train3x3pixels block texture features. Then based on the decision functions, CU splitting depth range is narrowed by skipping unnecessary splitting and early ter-minating splitting. After that, recall rate and precision are used to evaluate the effects of the decision functions. At last, experiments results are analyzed.(3)Firstly HOG(Histogram of Oriented Gradient) descriptors are used to describe the gradient information of CU. Secondly AdaBoost algorithm is employed to obtain a strong classifier. Thirdly the strong classifier is used to optimize coding unit splitting algorithm. Finally three sets of comparative experiments have been done and the results are analyzed.(4) OpenMP is used to implement the Wavefront Parallel Processing of HEVC. Then the implementation details are explained. At last, experiment results are analyzed.