Stereoscopic Video Objects Coding Based DT Mesh

As to the monocular videos have been far from enough to satisfy human visualdemand with the rapid development of multimedia technologies nowadays.3Dtechnology is appeared in many fields, for example, Communication, network, videoconference and so on. Binocular stereo video which is related to theories of computervision is the most widely used currently. Owing to binocular stereo videotransmission need two channels, its most significant difference with the traditionalvideo transmission is its e large amount of coding data is double of traditional videocoding. Obviously, audience doesn’t care about the whole details while enjoying the3D video but focus on person character or moving field. Therefore, coding in thispaper is established on object and the background segmentation and separate coding.Receiver can receive interested part selectively.Objects and the background segmentation in this paper is based on combinationof disparity map and frame difference. According to result of stereo matching forleft-right view, we extract object and background for the first time. Stereo matching isbased-on one-dimensional window matching and reducing stereo matching calculatedtimes effectively. Then make the object of the first extraction combined with framedifference. The experiences proved that extraction effect is more accurate.Stereoscopic video objects coding algorithm in this paper is based-on objectssegmentation and DT mesh. Receiver can decide whether to receive video backgroundand the content they are interested or not. Making use of iterative fast generatingalgorithm for generation of DT grid nodes, for the point utilizing Harris cornerdetection for input stereo image as initial node and generating initial Delaunaytriangle. The following mesh Detailing is based-on the insertion point by point.finally，make generated DT mesh appear on the original stereo view. Owing to takeinto account to maintain topological structure of DT mesh, we regard the object as thecomplete rectangle in the appropriate area to stereo matching for getting disparityestimation when we achieve to gain disparity estimation of objects, and not mesh matching which is border restricted. Finally, we transmit data structure coding to thereceiver. It turned out that we have well reconstructed view.