A Daptive D Isparity E Stimation for a Uto C Onvergence of R Egion of I Nterest in a V Ideo

Recently, various devices for three-dimensional (3-D) effect have been developed. For producing 3-D effect of the scene or the region of interest (ROI), disparity should be accurately estimated. People watching 3-D video feel visual fatigue if magnitude of parallax for the ROI is excessively large because a convergence point is not accurately put on the ROI. For producing 3-D effect, a 3-D formatter overlaps left and right images by shifting horizontally the right image by the estimated disparity of the ROI. In this paper, an adaptive disparity estimation algorithm for auto convergence of the ROI in a video is proposed using the first-order Taylor series expansion of disparity and adaptive disparity search range prediction in a stereoscopic video. First, a stereo video that consists of a number of pairs of left and right images is captured in parallel stereo camera configuration. A window in each frame is selected within the ROI and tracked. Then, for automatically adjusting a convergence point on the ROI, two steps are needed with the previously estimated disparities. The first-order Taylor series expansion is used to approximate disparity of the current frame of a video. Then, a moving average filter is used to adaptively determine disparity search range in similarity measure computation. Subjective evaluation such as visual fatigue, comfort, and 3-D effect of the proposed algorithm and existing algorithms is performed. Experimental results with four test videos and subjective evaluation show that the proposed algorithm gives 3-D effect with visual comfort.