Research On3D Video Coding Based On Depth Down-sampling

In the MVD format3D video coding, in order to reduce the bit-rate, the depth map is down-sampled at the encoder. But the process of down-sampling will increase the depth map distortion and how to effectively up-sample the depth map at the decoder is the key technique for this scheme. Different from the conventional methods, the proposed method in this paper takes advantage of spatial correlation and temporal consistency in depth map, as well as the correlation between the depth map and its corresponding texture image. This scheme improves both depth map coding performance and synthesized quality.In addition, the purpose of increasing the depth coding performance is to improving the virtual synthesized quality, so during the depth coding the coding performance should be measure based on virtual synthesized view distortion and the conventional rate-distortion function model is not suitable for the depth coding we proposed. For this reason, we optimize the rate-distortion function model in the depth coding scheme that the depth was down-sampled. This new rate-distortion function highly improves the coding performance and decreases the synthesized view distortion. The main contents are as follows:(1) A new depth up-sampling algorithm that the spatial correlation and consistency are jointly employed in depth up-sampling to achieve efficient depth coding is proposed. The spatial correlation among the depth map and the temporal consistency between the adjacent depth frames, as well as the correlation between the depth map and its corresponding texture image is utilized to decrease the depth maps distortion.(2) Depth up-sampling algorithm based on texture and spatial-temporal gradient of depth is proposed, during the depth up-sampling, the time gradient, horizontal gradient and vertical gradient of the depth map help to decide the intensity of the spatial correlation and temporal consistency, and then determine the way of up-sampling the pixels. This scheme makes full use of the character of the depth maps and improves the depth map coding performance.(3) The rate-distortion function model is optimized, in the down-sampled depth coding, an efficient distortion function measuring the virtual synthesized view distortion by analyzing depth down sampling, view warping and rendering process is proposed. The results prove that the proposed scheme much improves the coding efficiency and the synthesized view quality with carefully considering depth down sampling, view warping and rendering process.