Research On Fast Depth Map Coding For 3D-HEVC

3D High Efficiency Video Coding(3D-HEVC)is proposed by Joint Collaborative Team on 3D Video Coding Extension Development(JCT-3V),which is composed of MPEG and ITU-T for compressing 3D video data efficiently.A coding format of Multiview Video Coding plus Depth(MVD)is introduced to reduce the number of directly coding views in 3D-HEVC.With depthimage-based rendering(DIBR),virtual views can be generated and provide threedimensional visual feeling.There are some coding techniques introduced in depth map coding for high coding performance,which accelerates the computational load.Two fast algorithms for 3D-HEVC are proposed to reduce the coding complexity while maintain the coding performance.For intra prediction,a fast depth map intra coding algorithm based on sumof-gradient and segment-wise DC Coding(SOG-SDC)is proposed according to properties of depth map.In the algorithm,a detection of gradient is exploited to judge whether the region is homogeneous or not.For smooth area,some timeconsuming unlikely modes would be skipped,and the CU segmentation is no further splitting.In addition,since a coding tool of SDC is checking for all prediction mode,ignore some modes checking SDC calculation process in homogeneous region is reasonable.As experimental results shown,compared with original 3D-HEVC,the algorithm can save 26.03% coding time in average in all intra mode,with 0.14% increment of BDBR.For inter prediction,a Bayesian-decision-based fast depth map inter coding algorithm is proposed,including fast CU decision and DIS&&SKIP mode decision.The algorithm has two steps.First,the RDcost of DIS(Depth Intra Skip)is taken as the feature to decide whether current CU needs split or not,and then through modeling with Gaussian model,a Bayesian probability value is calculated to determine the CU splitting.Second,the RDcost of DIS and SKIP mode is picked as the Bayesian feature when they are the optimal modes.After offline training,the minimum error rate Bayesian decision is used to judge whether to skip the prediction mode except DIS and SKIP modes.As experimental results shown,compared with the original 3D-HEVC,the proposed algorithm saves 45.7% coding time of depth map in average,while maintaining the coding performance with 0.62% increment of BDBR.