Research On Prediction Teahniques Of Multiview Video Plus Depth Coding

In recent years, with the rapid development of stereoscopic display and interactive multimedia technology, people are no longer satisfied with the visual perception brought by the traditional two-dimensional video, but eager to the colorful video which can record the natural scene more realistically. Three-dimensional display video gradually enters people’s vision. Three-dimensional display video, just as its name implies, it is a set of signal collection which can provide depth perception for a given scenario. Multi-view video plus depth, as the mainstream data format of three-dimensional display video, becomes a research hotspot in the current field of stereo video.Multi-view video plus depth format consists of multi-channel videos and its corresponding sequences of depth map, so the amount of data is usually several times of traditional two-dimensional video. Furthermore, the collected video resolution is more and more big causing further surge of the data. A substantial amount of data makes it difficult to achieve the requirement of storage and transmission so that efficient compression coding technology is needed urgently to reduce the amount of video data. In 2013, a new generation of highly efficient video coding standard(High Efficiency Video Coding, HEVC) was published, and extended towards multi-view video plus depth coding to form a 3D video extension of High Efficiency Video Coding standard known as 3D-HEVC. Complying with the development demand of epoch, this paper aims at the study on the prediction technologies in 3D-HEVC. Prediction is the key link in stereo video coding whose quality directly affects the amount of transmission data and the final quality of display video, so this study has important application value and practical significance. From the perspective of improving coding efficiency, reducing coding complexity and enhancing coding capacity, this paper mainly includes the following the innovations and contributions:1. Depth map prediction: In depth modelling mode of 3D-HEVC standard, the reference pixel selection algorithm of constant partition value prediction does not well consider the boundary characteristics of depth map. For this problem, combined with the actual object boundary information, this paper proposes an improved constant partition value prediction algorithm. Meanwhile, considering the spatial edge similarity, an enhanced Contour segmentation algorithm is put forward.2. Inter-view prediction: In Sub-PU level inter-view motion prediction of 3D-HEVC standard, motion hole filling algorithm cannot pipeline execution and parallel processing, seriously affect its engineering applications. Aiming at this problem, this paper puts forward four motion hole filling schemes to realize pipeline execution and parallel processing.3. Texture prediction: In the depth-based block partitioning prediction of 3D-HEVC standard, great redundancy exists between the partition derivation process and the segmentation mask process. To address this problem, a modified partition derivation algorithm is proposed in this paper, skillfully using the mask value information obtained in the segmentation mask process, to simplify the partition derivation and eliminate internal process redundant.