Multi-view Video Coding Technology Optimization Based On Rate Distortion Theory

With the rapid development of multimedia applications and stereo display technology,3D video is becoming more and more popular.The new generation of multi-view texture plus depth(MVD)video has also become the mainstream 3D format.In order to efficiently compress 3D video in MVD format,the 3D-HEVC standard was formulated on the basis of the latest generation of high-efficiency video coding(HEVC)standard.Due to the in-troduction of multi-view and depth maps taken from multiple different perspectives,many new inter-view prediction technologies and depth map coding tools have been developed in the 3D-HEVC standard.These new technologies and tools have greatly improved the cod-ing efficiency of multi-view video,but also brought new challenges to the optimization of rate-distortion performance and reduction of coding complexity in multi-view video coding.Rate-distortion theory is the theoretical basis of video coding and has an important guiding role for the design of efficient multi-view encoder.Based on the rate-distortion optimization theory,this paper has carried out technical optimization research on the 3D-HEVC standard.The inter-view bit allocation of the rate control and the distortion-rate trade-off of the depen-dent views is optimized in the multi-view texture video coding,and multi-view depth video coding is simplified.Firstly,a novel bit allocation method for multi-view texture video coding is proposed,which brings about the optimal inter-view bit allocation in the rate control of multi-view coding,and improves the rate-distortion(RD)performance of the rate control of multi-view coding.Firstly,considering that the distortion in the base view(BV)is directly transmit-ted to the dependent view(DV)by inter-view prediction technique,a joint multi-view RD model is built.Based on the proposed joint multi-view RD model,a precise power model is derived to represent the target bitrates relationship between the BV and the DV.Secondly,by analyzing the relationship between the ratio of the average bitrates of the P-DV to the corresponding I-BV and the optimal ratio of the total bitrates of the DV to the BV,a lin-ear model is developed to assign the target bitrates of the P-DV.Finally,considering the spatio-temporal correlation,a new parameter prediction method based on Pearson correla-tion coefficient weight is proposed for the R-?model in CTU level.Secondly,the distortion-rate trade-off of the dependent views in multi-view texture video coding is studied.First of all,by investigating the sources of the distortion in the DV,a new distortion model for the DV is established.In addition,based on the proposed distortion model,an efficient Lagrangian multiplier decision for B frame is proposed by considering the inter-view dependency.Finally,the optimized Lagrangian multiplier for P frame is de-signed using the scaling factor,and a linear model is established to represent the relationship between the optimal scaling factor and the disparity of I-P frame.Finally,fast depth intra coding algorithm is studied,which significantly reduces the com-plexity of depth intra coding.Firstly,using the RDcost feature based on view synthesis opti-mization,a maximum depth layer decision algorithm based on RDcost is proposed to accu-rately predict the maximum splitting depth layer(MSDL)of each CTU.By studying the fre-quency distribution characteristics(FDC)of the RDcost J₀of the CTU level in depth layer 0is investigated under different MSDL,the initial threshold of the MSDL is constructed under the tolerable false rate(FR).An exponential model is established to represent the relationship between the designed initial threshold and QP,to finally determine the MSDL threshold of the CTU.Secondly,a fast prediction mode decision algorithm based on RDcost and spatial correlation is proposed.Considering the spatial correlation,the FDC of the RDcost J_ORMof the optimal rough mode(ORM)in rough mode decision(RMD)is analyzed under differ-ent optimal prediction mode.Based on the FDC of the J_ORM,the mode-skipping threshold T H_stthat has an exponential relationship with QP is designed to skip angular modes and depth modeling modes(DMMs).Finally,a fast wedgelet pattern decision method based on K-Means is proposed to speed up DMM1.In summary,based on the rate-distortion optimization theory,the new RD model in de-pendent view is derived using inter-view distortion transmission in this paper,and then the joint multi-view RD model is constructed.Based on this,the optimal inter-view bit alloca-tion is realized in the rate control and the distortion-rate trade-off of the dependent views is optimized,resulting on the improvement of the rate-distortion performance in multi-view video coding.At the same time,the research of RDcost based on view synthesis optimization reduces the complexity of depth map coding.The research results of this paper improve the coding performance of the existing 3D-HEVC standard and reduce its coding complexity,and can be applied to real-time coding and transmission of multi-view video.