A new region matching method for stereoscopic images

For region matching in stereoscopic images, most existing algorithms used the adjacency constraint (AC) between regions. However, in the case where there are occluded objects in the stereo images, this constraint may be violated. Therefore, there remain false matches and miss-correspondences in the final results. To overcome this problem, we propose here a new algorithm based on the relative position constraint (RPC) between regions. The relative positions between regions in one image are constrained to be the same as the relative positions between their corresponding ones in the other image. The matching process is achieved in two processing steps. The first one searches the candidate matches by using the epipolar constraint and the relative position constraint. The second step selects the correct matches among the possible ones, according to a proposed cost function. The matching process is propagated by using the adjacency relationship between regions. Starting with matched couples, the adjacent regions are examined first and matched if possible. The process is gradually extended to all the regions on the image. The proposed new matching algorithm is illustrated by a set of synthesized and real stereo images, with a comparison to the well-known algorithm based on the adjacency constraint. (C) 2000 Elsevier Science B.V. All rights reserved.