The Research On Multi-view Video Depth Extraction And Coding Technology

With the development of digital technology and communication technology,due to the lack of in-depth information,the original 2D video is gradually unable to meet the human visual needs.3D video can provide a natural visual experience and become the main video format for human needs and enjoyment.Multi-View Video(MVV)is one of the most important video representations of 3D video,however,the amount of Multi-View Video data the sheer size makes storage and transmission very difficult.At the same time,in practice,only a few fixed viewpoints can be taken,and scene information of all viewing angles cannot be collected,and the viewer can not provide any viewpoint video for selection.Therefore,the multi-view video compression coding problem and the synthesis of virtual viewpoints have become the main research contents of multi-view video development.Multi-View Video depth coding and virtual viewpoint synthesis are all based on depth images,and the encoding effect and virtual viewpoint synthesis quality largely depend on the obtained depth image quality.Therefore,this paper firstly studies the depth image extraction technology in detail,and focuses on the analysis to improve several typical depth extraction algorithms.Then,based on the traditional 2D texture image coding technology,the gray-scale distribution characteristics of the depth image are used first.The proposed edge detection method based on information fusion extracts the edge of the depth image,and then uses a multithreshold image segmentation method based on particle swarm optimization to segment the depth image and encode the segmented image separately to improve the depth map.The quality of the code.Major work and innovative research include:(1)Aiming at the problem that the fixed window matching in the regional stereo matching method is sensitive to the depth discontinuity and the weak texture region,an adaptive window stereo matching method based on pixel gradient is proposed.The method firstly sets a rectangular window with different sizes in advance,then selects the optimal window according to the gradient information of the pixels in the window,and finally calculates the best matching point using the normalized cross correlation(NCC)as the cost aggregation function.Higher depth map.(2)Because the feature matching uses the characteristic primitives with statistical special rules to perform the corresponding search of the left and right disparity maps,the influence of the depth discontinuity and weak texture regions is avoided.Combining the advantages of high matching precision and fast extraction speed based on feature matching and the advantages of scale,rotation and affine invariance of SIFT(Scale Invariant Feature Transformation),a stereo matching method based on SIFT features is proposed to achieve fast and accurate matching.(3)The method of stereo matching to extract depth images has the highest accuracy of depth map acquired by Graph Cut-based global stereo matching algorithm.For the multi-view video coding and virtual viewpoint rendering,the requirements of high-precision depth image are studied in detail.Based on the principle,the depth map extraction based on the solid matching of Graph Cut algorithm is realized.(4)For the depth image with clear contour and special texture,based on the traditional edge detection algorithm,an edge detection method based on multi-operator information fusion is proposed for edge extraction of depth image.(5)Propose a depth map coding based on image segmentation.Firstly,a multi-threshold image segmentation method based on particle swarm optimization is proposed.Then,this method is applied to the segmentation of depth images,and the segmentation blocks are separately encoded to achieve efficient coding of depth maps.