Research On Depth Information Optimization And Sprite Coding In 3D Video Coding

Technologies of communication, multimedia, and computer vision have been developed rapidly. Videos are essential for people’s life, especially 3D videos. 3D videos are preferred due to the reality and immersive. Most of 3D videos are represented using the multi-view texture image plus depth map(MVD) format. Uncompressed 3D videos cannot be transferred by constrained or low bandwidth and efficiently stored. It is necessary to compress 3D videos efficiently using MVD format. In this paper, we propose two methods to improve performance of 3D video coding. The main researches are shown as following:1. A quadtree-based partitioning methods for motion vector inheritance between depth and texture was proposed。Currently, in 3D-AVS 2.0 video coding standard, texture image and its corresponding depth map are encoded independently. However, the motion parameters of the texture image and its associated depth map should be similar, since they are both the same scenery which is captured by the same camera from the same viewpoint at the same time instant. The current 3D-AVS 2.0 encoding platform does not consider the correlation of them, resulting lots of wasted bits. Therefore, as to shortcoming of 3D-AVS 2.0 encoding platform, a quadtree-based partitioning method for Motion vector inheritance between depth and texture was proposed in this paper, reducing the data storage redundancy and improving the video coding efficiency. Experiments show that our proposed method achieves a better coding performance compared with the original 3D-AVS 2.0 encoding platform.2. A SPRITE aided video compression and segmentation method was proposed.Today, 3D-HEVC coding standard platform HTM 10.0 uses a multi-view texture image plus depth map coding format. But, the sequences of multi-view texture image plus depth map coding format were captured for the same scene by a number of cameras from different positions.Therefore, the texture images of adjacent viewpoints have a large number of overlapping background information. That is to say, there are much serious background redundancy information among the texture images of adjacent viewpoints.In order to reduce the redundant information of the backgrounds among all texture images of adjacent viewpoints, we proposed a SPRITE aided video compression and segmentation method based on 3D-HEVC coding platform in this paper. That is to say, the foreground and background were separated by the corresponding depth map information. A ccording to the camera parameters and the corresponding depth map of all the viewpoints, A SPRITE which is a Synthetic large background image by all backgrounds of all theviewpoints is created, eliminating the redundant information of the original backgrounds of Synthetic background image and three separated foreground images were coded instead of multiple viewpoints original texture images. Compared with the original reference software, the average coding bitrate of BD-rate respectively increases by 2.5%, 0.73%, 0.4%. The reason is that our method can reduce coding bits for nearly half of all, but the PSNR of every viewpont decreases about 2d B. In the later work, we will apply more viewpoints, such as 10 viewpoints or 18 viewpoints, to our SPRITE coding framework to further reduce the coding bits and enhance coding efficiency.