Research On Fast And Efficient Algorithm For Depth Video Coding

With the development of three-dimensional (3D) display and interactive multimedia systems,3D video applications, such as3DTV and free viewpoint TV (FTV), have been booming up recently. Multiview plus depth (MVD) is the main3D representation format. As the important part of MVD, multiview depth video is used to synthesize new virtual view at the display side. Compared with single view video, MVD includes more information and the amount of data is enormous. In the case of existing network environment, the primary problem is how to efficiently compress the huge data. Moreover, the depth video and the texture video use the same video coding standard. This standard searches several modes and reference frames to improve compression performance and obtain a higher quality of virtual view, which makes the coding complexity exponentially. This paper presents three possible algorithms against the problem of low compression performance and high complexity in the coding process of depth video. Specific algorithms are as follows:(1)Most regions of the depth video are relatively flat and less texture. The kinds of mode are relatively small and some mode has a high proportion. The pixel values of motion and boundary regions change relatively fiercely, which plays the important role to the quality of the virtual view. The optimal modes in these regions are complex. Considering the above factors, the depth video is divided into four different regions. Different modes are adopted in different regions. Experimental results show that the proposed algorithm saves encoding time ranging from85.73%to91.06%while it maintains virtual view quality and bit rate.(2)Firstly, according to the difference of pixel values in texture video within the range of time domain, the depth video is segmented into two regions:motion regions and static regions. The blocks in static regions are skipped without encoding. The skipped blocks are predicted from the neighboring depth images. Experimental results show that the proposed algorithm not only saves the BDBR up to56.83%and ameliorates the BDPSNR up to1.93dB but also improves virtual view quality and saves the encoding time.(3)Combine the correlation between the current macroblock and the corresponding macroblock of the reference frames, the adjacent macroblock and the affect of the boundary regions on the quality of the virtual view, the depth video is divided into five regions. The candidate modes in different regions and the rate distortion cost of the main candidate modes are analyzed. Then an early termination method of the mode selection for some regions is proposed. Experimental results show that the proposed algorithm saves encoding time ranging from83.60%to93.41%while it maintains virtual view quality and bit rate.