Research On Efficient Depth Coding In3D Video

With the rapid developing of stereoscopic display and interactive multimediasystem, Three-Dimensional video (3DV) has drawn increasing attention as a newmultimedia technique for capturing real-world scenes. Three-Dimensional television(3DTV) aims to provide the viewers with depth perception of the scene bysimultaneously rendering multiple views from different viewing angles.Free-viewpoint television (FTV) allows interactive selection of the viewpoint anddirection in the scene within a certain operating range. In order to promote3DTVand FTV applications, multi-view acquisition, depth map coding, and virtual viewrendering based on depth are being developed as the key technologies. This paperresearched in deep around high efficiency depth map coding and the majorcontributions are as follows:A fast depth map compression method oriented to virtual view rendering isproposed in order to speed up the compression of depth data while ensure the imagequality of the synthesized virtual view. According to the characteristics of depth dataand the impacts of depth distortion on virtual view synthesis, depth map is dividedinto edge regions and smooth regions, and correspondingly two search strategies aredesigned for the mode decision in depth map coding. Full search is for the relativelycomplexity regions and edge regions, while SKIP, Inter16×16and all macroblockmodes of Intra are for smooth regions and low complex regions. The experimentalresults evaluated by objective and subjective methods indicate that compared toX264, the proposed algorithm not only accelerates the depth map encoding greatlywith2.06~2.29times but also maintains considerable virtual view rendering qualitywith constant bit rate. It can be applied in real time depth based3DV system.A JNDD based fast depth map coding method oriented to virtual view renderingis proposed to make further optimization of mode decision in depth map coding.Through analyzing the characteristics of depth data and the quality of virtual views impacted by depth distortions in depth coding, JNDD model is introduced. Depthmap is divided into vertical edge regions which are sensitive to rendering distortionsand smooth regions whose distortions are not noticed by human eyes, andcorrespondingly two search strategies are designed for mode decision in the codingprocess. Experimental results show that the proposed strategy saves55.05%~66.27%total coding computation for standard test sequences and57.69%~68.43%total coding computation for captured test sequences with a considerable virtual viewrendering quality and a constant bit rate, which is helpful to accelerate the depth mapcoding in3DV systems with good robustness.