Abstract: In this paper, we integrate feature-based image matching method with feature-based one. Zero-crossing edge points are first extracted from image pairs with the Gauss-Laplace operator caculating. They are matched according some constraint conditions. Then, being looked as symmetries, image pairs' other point to point correspondences are searched from two directions by pixel neighbor area correlation values. Applying this approach, we have designed Visual C++ programs for rectified image pairs and extracted their condense disparity map not only fast but also efficiently.