Microsoft Word - WG2-4.DOC

In this paper, we present a new stereo matching algorithm using global and local color segmentation. After we perform a quad-tree decomposition to generate global and local color segments, we determine meaningful depth information between global and local disparities based on the disparity space distribution. Unlike the previous segment-based stereo matching methods, we consider globality and locality at the same time to handle textureless regions. Expereimental results demonstrate that the proposed scheme generates high-quality disparity maps.