The Research Of High Efficiency Depth Estimation On3D Video

Along with the development of the3D technology, people realize the importance of3DTV and it does give the users a sense of reality. However, compare3DTV with traditionalTV programme, the data size of3DTV is much larger than the traditional ones, which bringsus a big problem in3D data translation. In order to meet the requirements of the limitedbandwidth, it is not convenient to transport each view points of the3D video, thus it isnecessary to use an immediate3D representation of view scene system. Using depth mapsand synthesizing video point at the decoders can be much more efficient than transportingmulti-view videos. Firstly this paper introduced the principle of stereo matching, thus raisesthe concept of the stereo matching and depth map, and then from the aspects of accuracy andthe speed of the algorithm do some research, respectively, the results have achieved goodperformance.1) Stereo matching is one of the key techniques in depth estimation. Traditional stereomatching methods will destroy the edge of an image partly. To improve the depth estimationprecision, an image enhancement based stereo matching algorithm is proposed that utilizesimage enhancement on the reference image and the target image and the scene is enhanced bya gradient method which makes the latter segmentation step more robust on edges, then thedisparity plane is assigned to each segment. Experiment results demonstrate the superiorperformance of the proposed algorithm.2) A fast depth estimation algorithm based on the K-means image segmentation isproposed in this paper, aiming to reduce time cost and at the same time maintain or evenimprove the performance of depth map estimation. Firstly, the reference image is segmentedby K-means clustering algorithm method and then mark each segment as different types.Secondly, apply different matching methods for each kind of segment to get the initialmatching cost. Thirdly, correct the depth values of unreliable pixels in each segments. Finally,depth values of each segments are determined by using the final matching cost. The results show that the proposed method present a high quality depth map with a faster speed.The above algorithms are global based and local based algorithm respectively, we canget a high quality and stable depth map through the global based method, and the fast depthmap algorithm can reduce the time cost significantly but the image quality stability is low.The two kinds of method suit for different fields.