Asymmetric 2+1 pass stereo matching algorithm for real images

Abstract

An asymmetric 2+1 pass stereo matching algorithm is proposed for generating a depth map from the input of two real stereo images. In the first pass, the Canny edge detector is used to acquire edge images from the inputs. Component images are then generated from the edge images. The edges of each component in the left component image are cut to obtain asymmetric left and right component images. The concept of a disparity-space image (DSI) by components is introduced and dynamic programming (DP) techniques are applied to match the asymmetric left and right component images. In the second pass, the previous DSI is modified, applying the DP methods to acquire the matched pixel pairs of the matched component pairs, and to generate ground control pairs (GCPs). In the final pass, these GCPs are incorporated into the DP algorithms to determine the optimal path and acquire the depth image. The results show that the proposed approach produces reliable and efficient matching and preserves the shapes of objects effectively for real stereo images. The proposed algorithm makes contours clearer than other conventional methods for real stereo images. (C) 2006 Society of Photo-Optical Instrumentation Engineers.