Stereo correspondence by surface segmentation

The authors present a new algorithm for stereo matching which makes use of simultaneous feature matching, surface reconstruction and segmentation. Each of these three phases is described. The algorithm integrates these three related subprocesses in the process of extracting surfaces from stereo data. The integration is synergetic: the whole is greater than the sum of the parts. After the segmentation phase, each segmented surface is bounded by a closed boundary. The algorithm yields a rich description of a scene in terms of global surface patches and closed surface boundaries. Experimental results are included for a pair of stereo images of an aerial view of the Pentagon building