Local Stereo Matching Using Adaptive Local Segmentation

Figure 5

Disparity results for the stereo pairs (1st row: Tsukuba, 2nd row: Venus, 3rd row: Teddy, and 4th row: Cones) from the Middlebury testbed database. The columns show, from left to the right, the left image, ground truth, result computed by our stereo algorithm, and disparity error map larger than 1 pixel. The nonoccluded regions errors with ranking are, respectively, Tsukuba 1.33% (37), Venus 0.32% (39), Teddy 5.32% (17), and Cones 2.73%  (14).