A contour-based stereo matching algorithm using disparity continuity

摘要

A contour-based stereo matching algorithm is presented using the disparity continuity property along image contours. The basic idea is to first match specific high-information points along a contour, then use the matches found to guide other matches on the same contour by sequentially propagating the disparities along the length of the contour. The matching algorithm consists of three stages: (1) extraction of high-curvature edge points, terminations and junctions (collectively refered to as extremal points), (2) matching of the extremal points, and (3) matching other edge points by propagating the measured disparities at the extremal points along contours. We present a number of examples illustrating the performance of the algorithm in natural stereo images ranging from aerial images to laboratory images.