Region-based 3D Video Coding Research

Along with the development of science and technology, multi-media technology has played a more and more important role. To be a rich content represent form, three-dimension video(3DV) has been achieved more and more attention. Multi-view plus depth(MVD) format is the most common format to express 3DV, which formed by several pictures of camera and these corresponding depth pictures. As 3DV has a better tradeoff in the relationship between the amount of videos' data and the number of views, and we can use texture pictures and these corresponding depth images to render virtual views in receiver side by taking advantage of depth-image based rendering(DIBR) technology, therefore increase the number of views. However, with the increasing of original views, the data size has a drastically increase. The texture images have a higher significance as they are watched by user directly, and a higher proportion in code stream. Therefore this article is mainly do research in the compression and coding of texture views, and the main work and innovations are as follows:1. A region-based 3D video coding scheme is proposed. In this scheme, according to virtual view synthesis regular and combining original depth maps, we do region division on original texture views by simulating the procedure of virtual views render. Then divide the regions into important ones and unimportant ones according to its contribution to the virtual view, and different importance has a different BCS coefficients. The important regions have a higher coefficient to protect these regions and the unimportant ones have a lower coefficient to decrease the proportion in code stream. In our experiment we do BCS measurement and reconstruction on original views using eight sets of coefficients, and synthesize virtual view of each set, then compare the PSNR of virtual views. The final results indicate that in the equivalent measuring coefficients condition, compare to measure the original views directly, and region divided scheme has a higher virtual view quality.2. A region-division JPEG based 3D video coding scheme is proposed. In this scheme, on the basis of region division of original views, JPEG coding scheme is used to compress and coding the different regions, different quantization precision is used to quantify different regions. The important regions have a smaller quantization step to protect them and unimportant regions have a larger quantization step to waken them. Finally virtual view is synthesized using decoded video sequence. Experiment result indicate that, compare to the undivided scheme, on the condition of equal bpp(bits per pixel) the region division scheme has a promotion of virtual views' quality.