Сопоставление стереоизображений как задач о назначении

An algorithm for calculating dense disparity map for stereo images is proposed. Stereo correspondence problem is formulated in terms of an assignment problem in this algorithm. The algorithm processes each image scan-line individually similar to a dynamic programming approach to the dense stereo correspondence problem, however unlike dynamic programming, it does not rely on ordering constraint. Ordering constraint elimination is useful for the scenes with narrow objects in front of the camera. The proposed algorithm supports occlusions detection and guaranties uniquness of found matches.