Gray correlation technique is adopted to extract gray correlation peaks as a coarse matching set, which is called multi-peak set. Disparity gradient limited constraint is utilized to optimize the multi-peak set. The unique match can be obtained by calculating cross correlations between the matrices consisting of reference differences and disparities from the multi-peak set. Dynamic limited-search is introduced to reduce the computational complexity. A pair of obvious mark points in the left and right images, respectively, is set as corresponding reference points to determine search direction and scope at first. After the unique match being gotten, we take this match as a new reference point till all feature points in the left (or right) image being matched with points in the right (or left) image. It is proved that the proposed algorithm in the paper is feasible and accurate by 3-dimensional reconstruction on a pair of actual natural images and an object with given 3-D coordinates.