Visual surface segmentation from stereo

This paper presents an algorithm to extract and segment surface description from stereo images. Since the feature-based stereo matching algorithm only provides sparse and scattered depth data, the traditional range image segmentation technique cannot be applied to our problem. The proposed algorithm integrates the boundary- and region-based approaches to achieve more satisfactory segmentation results. Our segmentation algorithm yields a rich description of a scene in terms of global surface patches and closed surface boundaries. The surface boundary is determined by the local and global measures of surface smoothness. This results in a consistent interpretation of the scene.