Iterative disparity estimation for multiview video applications

In multiview video applications the disparity is a important feature, being able to be used in coding techniques, in synthesis of views, among others. The estimation process, however, is a very complex task which requires high processing capabilities and can lead to imprecise results. This paper presents an estimation method that uses segmentation and disparity map information in an iterative cycle. In each cycle the frames are segmented and then applied in a disparity estimation process. Next an error analysis is made, correcting inaccurate values. The result map is then used to improve the segmentation, initiating another cycle. Experimental results indicate that the proposed method works diminishing up to 25% the mean error presented in the maps.