With the development of the times,video has been widely used in people’s life and work,and has become an essential way of transmitting and obtaining information in people’s daily life.However,the amount of information contained in the video itself is enormous and the transmission process not only consumes a lot of time but also places increasingly high demands on the communication equipment.Therefore,video compression is required.And with the rapidly development of science and technology and the Internet,the variety of video terminals is increasing,which requires that the video stream can be adapted to the different terminals.The spatial scalable high efficiency video coding(SHVC)proposed based on high efficiency video coding(HEVC)can effectively solve this problem.However,the coding process of this standard is very complicated,which seriously affects its wide application.Therefore,the study of spatial SHVC fast algorithms is of great importance.In this thesis,the spatial SHVC intra-frame fast algorithm is investigated.Firstly,a fast algorithm for inter_layer reference(ILR)mode selection based on the distribution of RD costs is proposed.Since the RD costs of both the ILR mode and the Intra mode used by coding unit(CU)obey a Gaussian distribution,and the RD costs of the two modes are very different.Based on this feature,Bayesian decision making is used to determine whether the ILR mode is the best mode to encode,thus skipping the encoding of the Intra mode.The experimental results show that the bj(?)ntegaard delta bit rate(BDBR)increases by 0.07% on average,the peak signal noise ratio(PSNR)decreases by 0.08 d B on average,and the enhancement layer and overall coding time are saved by 59.48% and 45.15% on average,respectively.Secondly,a fast algorithm for directional mode selection based on the distribution characteristics of the residual coefficients is proposed.The number of corresponding non-zero coefficients is studied according to the distribution characteristics of the residual coefficients.The hadamard costs and the number of non-zero coefficients are combined to predict the optimal directional mode among the RD optimization(RDO)candidate modes.Experimental results show that the BDBR is reduced by 0.028% on average,the PSNR is reduced by 0.02 d B on average,and the enhancement layer and overall coding time are saved by 29.61% and 22.56% on average,respectively.Finally,a fast algorithm for early termination of the coding depth based on the residuals of the CU is proposed.Eigenvalues are selected based on the residuals of the current CU,and the eigenvalues and Liblinear are combined to predict whether the depth of the current CU is terminated early or not.The experimental results show that the BDBR increases by 0.01% on average,the PSNR decreases by 0.02 d B on average,and the enhancement layer and overall coding time are saved by 41.24% and 31.36% on average,respectively. |