The single-view depth map generation algorithm based on object motion information

To improve the 3D effect in the motion objects, to which are paid more attention by the viewers, this paper uses the motion information of objects to generate the depth map. However, traditional motion estimation generates too much motion vectors and inconsistent motion vectors in an object. The results will lead to the decline of the quality of the depth map, especially in the homogeneous regions. For solving this problem, this paper proposes a seed growing segmentation method for object segmentation and finds the corresponding feature points in each object. For the object motion estimation, this paper uses an adaptive cross base matching window to achieve the motion value for each feature points of objects. Then the depth value can be obtained by the motion value. In addition, to solve the discontinuing depth value in successive frames and to reduce the viewers´ discomfort on viewing 3D content, the same depth value is given to the same object in the previous frame by the proposed object detection algorithm. In the experimental result, the proposed algorithm proves the performance is better by comparing the efficiency of algorithms, e.g. from 0.06 to 1.82dB in PSNR and from 0.06 to 0.18 in SSIM, respectively.