A segment-based two-step stereo correspondence algorithm is proposed. In the first step

Dynamic Programming and Left-Right Consistency are used to detect ground control points

and the original image is segmented into small areas

where each of them is assumed as a single disparity

mean while the detected GCP are used to compute the disparity value of the part of the image; In the second step

the left unmatched areas are coded as a single chromosome

and are optimized with genetic algorithm to generate the dense stereo disparity map. The algorithm is tested on the standard image pairs

and the experimental results show its good performance.