The Study On Color And Distinctiveness Information Based Local Stereo Matching

Stereo image system is a newly emerging media system and can largely enhance our watching experience. Disparity information is the basis of stereo image processing. As an important method to obtain disparity information, stereo matching becomes a hotspot in computer vision area, and can be widely applied in many kind of situations such as robot vision, medicine, military affairs, amusement, etc. Stereo matching algorithm can be classified in two classes-local algorithm and global algorithm. Our research is targeted at three matching algorithm in local method-adaptive window matching algorithm, adaptive support weight algorithm and distinctive similarity matching algorithm.The traditional fixed window matching algorithm use fixed support window for all pixels, and it can not achieve accurate matching result for both low textured area and discontinuity area. Adaptive window algorithm choose different support window for pixels in different region, and can achieve more accurate disparity map than fixed matching method. How to select appropriate matching window effectively for each pixel is the most important problem of adaptive window algorithm. We proposed a color grouping based adaptive window method, which can insure all pixels in the same matching window has strong correlation. The experiment result shows this algorithm can get accurate disparity map with only one time iteration.Different from adaptive window based algorithm, adaptive support weight algorithms use fixed square window, but allocate different support weight for each pixel considering the disparity similarity with the pixel in the center of support window. Based on the assumption that pixels with similar color and small distance has strong correlation, adaptive color weight algorithm allocate different support weight for each pixel in support window according to the color similarity and space distance compared with the center pixel. We give detailed analyze to the shortage of adaptive color weight method and improve the support weight allocate process with color segment, to enhance the matching accuracy in region border, high textured area and low textured area.The distinctive similarity matching algorithm extracts each pixel's distinctive information, and based on the hypothesis that pixels with larger distinctiveness value have higher matching possibility, this algorithm select matching pixel according to the pixel's distinctiveness and the similarity with the target pixel. This algorithm can improve the matching accuracy in region border and high textured area, because these areas have large distinctiveness value. The key problem of distinctive similarity matching algorithm is how to extract the distinctive information effectively in both reference and target image. We propose a color segment based distinctive similarity method to handle this problem, and use the color region segmented by color segment algorithm to reduce the disparity search scope, to improve the matching efficiency. At last, we proposed a fast matching method combined adaptive window algorithm and distinctive similarity algorithm to further improve the disparity matching efficiency. This algorithm uses distinctive similarity matching method in picture areas which can't be easily matched correctly, but use adaptive window algorithm in other picture region. This algorithm has both good matching accuracy and efficiency.