Sing Bing Kang, Richard Szeliski, and Jinxiang Chai
While stereo matching was originally formulated as the recovery of 3D shape from a pair of images, it is now generally recognized that using more than two images can dramatically improve the quality of the reconstruction. Unfortunately, as more images are added, the prevalence of semi-occluded regions (pixels visible in some but not all images) also increases. In this paper, we propose some novel techniques to deal with this problem. Our first idea is to use a combination of shiftable windows and a dynamically selected subset of the neighboring images to do the matches. Our second idea is to explicitly label occluded pixels within a global energy minimization framework, and to reason about visibility within this framework so that only truly visible pixels are matched. Experimental results show a dramatic improvement using the first idea over conventional multibaseline stereo, especially when used in conjunction with a global energy minimization technique. These results also show that explicit occlusion labeling and visibility reasoning do help, but not significantly, if the spatial and temporal selection is applied first.