AbstractThere have been proposed a large number of methods to extract the three‐dimensional information in image understanding. Among those, stereo‐imaging is considered interesting as a passive method that does not affect the object. The problem of stereo imaging can be reduced to the corresponding point search problem. An attempt is made to use the following method to arrive at high‐density distance information. The energy functional is defined for the disparity considering the smoothness constraint. Then, pointwise stereo matching is realized by minimizing the functional over the whole image.The active net is one such approach. This technique often is combined with the iterative solution method based on the coarse‐to‐fine strategy using the multiple resolution, but it is not easy to achieve the matching for a wide range of disparity.This paper proposes a new stereo‐matching method, where the far‐search power of the genetic algorithm and the local optimization power of the active net are combined. It is a method in which a large number of partially corresponding patterns are prepared and calculated using the active net. Then, the genetic manipulation is applied to the set of patterns, and the optimal correspondence is sought in a stable way by iterating the change of generation. The procedure has the feature that the inadequate parameters are sorted out, and the optimal parameters are select