Research On Depth Coding Of 3D Video

With the development of multimedia communication business,three-dimensional video has greatly enriched the existing forms of multimedia.3D video brings a new stereoscopic experience to users and is the developmental direction of multimedia technology in the future.The data format of 3D video is the multi-view video with depth map plus texture map.Due to the increasing of the viewpoints and the depth map,the data volume of 3D video grows by geometric multiples.Facing with the pressure of 3D video in the process of compression,transmission and storage,it is urgent to efficiently compress and reconstruct the 3D video in a high-quality.And depth map can effectively express the depth information of the objects in 3D scene and occupy an important proportion in the 3D video data volume.Based on the characteristics of depth image and the difficulties encountered in the development of 3D video,the paper carry out the relevant research on the subject and achieve some work:(1)Considering that the depth map adopts the down-sampling/up-sampling frame as the pre-processing and post-processing in the most applications,we propose block-level entropy-based adaptive sampling framework for depth map.Based on the characteristics of the depth map and the essence of the information entropy,the scheme adaptively allocates the sampling rate according to the entropy value of the image block to protect the boundary of the depth map from damaging.The scheme can improve the reconstruction quality of the map as much as possible in the case of low sampling rate,so as to ensure that the reconstructed depth map can provide a reliable three-dimensional stereoscopic effect.(2)Aimed at the resistance of standard video coding in the development of wireless sensor network,a entropy-based classified dictionary learning for distributed compression sensing video coding scheme is proposed based on the structural features of video frames.In the scheme,DCVS can resolve the complex operations,and dictionary learning can sparsely represent the signal effectively to improve the signal reconstruction accuracy.The proposed classified dictionary learning algorithm makes the sparse representation coefficients of each type of image blocks to reach a higher degree of sparseness,so as to reconstruct the video more accurately and improve the quality of reconstructed video.(3)In the DCVS framework,GOP grouping directly affects the samples of dictionary training,which means GOP grouping indirectly effects the reconstruction quality of the video sequence.It will waste the compression resources if the group is too small while it will cause the low quality of reconstruction frames if the group is too big.In view of this contradiction,a dynamic GOP grouping for distributed compression sensing video coding algorithm is proposed,which employs mutual information to group dynamically.The grouping result is more reasonable and the extracted Key frames is more representative so that the trained dictionary can improve the reconstruction accuracy.The scheme is especially suitable for the video with rather more scene changes.