关键词: 局部立体匹配, 自适应权重, 图像分割, 测地距离

Abstract: Against the problem of the disparity map is not smoothly enough which produced by adaptive-weighted method, the paper puts forward a new color difference function that based on the geodesic distance between segmentation to weight the local support window, which combines with adaptive-weighted method and image segmentation priori. The dominant idea of the new scheme is to segment the image when matching, and regards the support window of the local method as a collection of different segmentations regions, then calculates the weight for segmentations of the window individually using geodesic distance, with that, it computes the matching cost, confirms the disparity by using the WTA method at last. Compared with the primary adaptive-
weighted and the primary segment-based method, the experimental result indicates that the new scheme has better performance.

Key words: locally stereo-matching, adaptive weights, image segmentation, geodesic distance