Research On Fast Algorithm Of Depth Video For 3D-HEVC

Three-dimensional(3D)video system has become a hot research field in video coding and communication because it can provide realistic stereoscopic visual perception experience and support users to choose the viewing angle freely.Compared with 2D video system,3D video system needs to transmit more video data to provide stereoscopic visual experience in the application side,which brings a great burden to transmission network.In order to reduce the data amount required for 3D video,multi-view video plus depth(MVD)has become the most popular format of representing 3D scenes.One of the reasons is that depth video represents the distance information from the camera to the objects in the scene,the ultimate goal is to assist in generating arbitrary intermediate views for 3D video system by using the depth image based rendering technique.Compared with the conventional single-view color video,the MVD signal is consisted of multi view color video and corresponding depth video.Besides the spatial-temporal correlation,inter-view correlation also can be employed.In order to improve the compression efficiency of the MVD signal,the 3D video coding reference platform test model of 3D-HEVC(HTM)adopts the following technologies including: the mode full search,the quad-tree recursive partition,the rate distortion optimization,the motion estimation and the disparity estimation,et al.Different from the natural videos or images,depth maps generally have more homogeneous regions and a small amount of sharp edge areas.In addition,the quality of the depth image affects the quality of the rendered virtual view video.Therefore,in order to improve the compression efficiency and the quality of virtual view,HTM also adopts these technologies including: the depth model models,the inheritance of motion parameters and the rate distortion optimization criterion based on virtual viewpoint distortion.Compared with the color video coding,these technologies result in the increase of computational complexity of depth video.Therefore,it is of great significance to develop real-time applications of 3D-HEVC with high quality of virtual view and high coding compression efficiency.In this paper,three creative and effective algorithms are proposed:(1)A multi-class support vector machine-based(MSVM)fast coding unit(CU)size decision algorithm for 3D-HEVC depth video intra-coding is proposed.The algorithm included two steps: off-line training and fast CU size and mode decision.In the process of off-line training,a MSVM model is constructed,where the texture complexity of current LCU,the optimal partition depth of its spatial neighboring LCUs and co-located LCU in texture video are treated as feature vectors,and the optimal partition depth of LCU is utilized as corresponding class label.In the process of fast CU size and mode decision,features of LCU are extracted before coding a LCU,then,a MSVM model is used to predict the class label.Finally,the class label that represents the largest partition depth of the current LCU is employed to terminate the CU recursive splitting process and CU mode search process.Experimental results show that the proposed algorithm saves the encoding time of 3D-HEVC by 35.91% on average,and the encoding time of depth video by 40.04% on average,with negligible rendered virtual views image degradation.(2)A fast CU size decision and CU mode decision algorithm for 3D-HEVC depth video intercoding is proposed.First,an off-line trained Bayesian model is used to predict the class label with a feature vector consisting of the partition depth of the corresponding spatial,temporal,and intercomponent(texture-depth)neighboring LCUs as the input.Then,the class label that represents the largest partition depth of the current LCU is used to terminate the CU recursive splitting process.Finally,the CU mode search process is terminated early via mode correlation of its spatial,intercomponent(texture-depth),and inter-view neighboring CUs.Compared with the 3D-HEVC reference software HTM-10.0,the proposed algorithm reduces the encoding time of depth video by 65.03%,on average over different test sequences,with negligible quality degradation of the synthesized virtual views.(3)Aiming at the high complexity problem in 3D-HEVC depth video inter-coding,a fast depth video inter coding algorithm is proposed by motion consistency.Firstly,each frame of depth video is divided into four regions: texture move and depth move region,texture move and depth static region,texture static and depth move region and texture static and depth static region.Then,different partition depth levels and prediction modes of LCU are designed for encoding different regions.Finally,compared with HTM-10.0,experimental results show that the proposed algorithm can save 40.13% entire time and 63.58% depth time over various test sequences with the little impact on the quality of virtual views and coding performance increased by 2.78%,respectively.