Watershed-based stereo matching algorithm for depth map generation

In Binocular depth generation algorithm, the depth value can be found by the disparity estimation by taking advantage of the similarity in these two views which be captured at the same time. However, it's not easy to find the accurate depth value if there is high similarity in color space around the area. To reduce the matching error in the regions with high similarity, the pixel information on the edge of the objects in color space can be used to predict the depth value of the object. This paper proposes an object region determination based on watershed segmentation algorithm to generate a matching window for stereo matching. To reduce the depth discontinuity and the computation time for disparity estimation on all pictures, this paper uses the motion vector information to compensate depth value for the successive depth video. Thus, depth map generated by the proposed algorithm can produce more suitable stereo images. Experimental results show that the stereo scopic reliability can be improved by the proposed algorithm in high stereoscopic sequence. The SSIM is increased about 0.34 and the PSNR is increased about 0.55dB when the proposed algorithm is used to compare the other approaches.