Research On Low Complexity Depth Intra Coding In 3D-HEVC

After achieving a great success in the film industry,3D video has been developing to the free view video for 3D family interactive entertainment.In multi-view video,the quality of free view video depends on the number of views.Although,multi-view video can be efficient-ly compressed by MVC(Multi-view video coding)encoder,its data is linearly proportional to the numbers of views,which can generate huge data for high quality 3D application.In order to address this issue,a high efficient 3D video format called multi-view plus depth(MVD)is developed because of its capability of providing the depth perception and flexibly employing virtual view generation via Depth Image-Based Render technology.MVD contains several views and their corresponding depth maps in general.The most important challenge issue is how to compress their huge data efficiently.To address this challenge,the standardization of the depth enhanced 3D video coding standard,3D-HEVC,was formally launched by Joint Col-laborative Team on 3D Video Coding Extension.When applying the progressive coding tools,3D-HEVC yields significant coding efficiency and provides the outstanding perceptual quali-ty of synthesized views compared with HEVC.Some state-of-the-art depth tools are adopted by 3D-HEVC for depth coding.Simultaneously,the coding complexity of depth map is also increased drastically,which prevents the 3D-HEVC from practical applications.Hence,a more efficient depth coding process of 3D-HEVC is highly demanded for prac-tical applications.In this context,this thesis introduces depth intra coding and analyzes its complexity,then proposes some algorithms to speed up depth intra coding.Main works and innovations of this paper are as follows:·This thesis analyzes the complexity of depth modelling mode(DMM)decision and proposes two fast algorithms:region-based wedgelet pattern decision algorithm and low com-plexity rate-distortion metric.In region-based wedgelet pattern decision algorithm,the piece smooth of depth block is investigated and the most probable wedgelet patterns can be predicted by the piece-smooth of current block.By investigating the statistical characteristics of variance distributions in the two partitions of DMM,a simple but efficient criterion based on the squared Euclidean distance of variances is suggested to evaluate RD costs of the DMM candidates in the low complexity rate-distortion metric.·This thesis analyzes the complexity of depth intra coding and proposes two fast algo-rithms.First,By investigating the spatial correlation of depth map and analyzing the reference pixels,we have found that not all intra modes are necessary to be involved in most cases.Hence,an adaptive intra mode skipping strategy is proposed in this paper.More specifically,the most probable mode candidates are predicted according to the neighboring reference pix-els.Consequently,some intra mode candidates are out of Rate-distortion calculation as well as mode signaling for smooth and piece-wise smooth blocks.Second,a probability-based ear-ly depth intra mode decision is proposed to select only the most promising mode and make the early determination of using segment-DC depth coding based on the low complexity RD-Cost in rough mode decision.Since this method uses the intermediate coding parameter as the feature,it is easier to be implemented in hardware and has good compatibility..3D-HEVC adopts the brute-force method to determine the optimal CU partition,which increases the computational complexity dramatically.Hence,by investigating the relationship between intra modes and CU partition,an efficient CU decision algorithm is proposed for depth intra coding,in which the depth level of CU is predicted by corner-point(CP)and the co-located texture CU.More specially,firstly,the CPs are obtained by corner detector in junction with the quantization parameter,which are further used to pre-allocate the depth level.After that,the refinement of pre-allocation depth level is performed by considering the block partition of the co-located texture.Finally,different depth search range is selected based on the final pre-allocation depth levels.In order to verify the validity of the proposed algorithm,experiments were performed using the 3D-HEVC reference software HTM and the state-of-the-art algorithms.The simu-lation results show that the proposed algorithm can achieve significantly complexity reduction without incurring noticeable performance degradation for synthesized views.