Abstract: Abstract: Many objects have clear contour and texture in the high spatial resolution satellite stereo pair. Due to the elevation differences in many objects, and the existence of building shades, and the similar objects, and so on, the extraction of corresponding feature points from the high spatial resolution satellite stereo pair is difficult, which leads to a rough disparity map. Aiming at the problem, the graph cuts algorithm, which has a successful application in the computer vision field, was introduced and improved for the stereo matching. The core problem of stereo matching is to compute the optimal disparity value. Based on this rule, the graph cuts constructs the global energy function by using the disparity value of all the pixels, and transforms the problem of stereo matching to the problem of minimization of the global energy function. However, there are two problems existing in the process of stereo matching by using the traditional graph cuts for the high-resolution satellite stereo pair. The first one is that the time complexity is high; the other one is that the disparity map has a lower precision. Aiming to the aforementioned two problems, the graph cuts algorithm was improved. The improved graph cuts constructed the network and energy function based on the epipolar images, which not only lowered the time complexity, but also improved the matching precision. Based on the improved graph cuts algorithm, the global energy function was minimized, and then the minimum cut was solved and the accurate disparity map was obtained. The EROS-B satellite stereo pair was used for the experiment. Based on the equal distribution rule, 25 control points were selected for the precision check. The true disparity map was computed and compared with the disparity map created by the improved graph cuts stereo matching algorithm. By comparison and analysis on the results, two conclusions were obtained. First, compared with the traditional graph cuts algorithm, the improved graph cuts algorithm had a higher precision and an obviously lower time complexity. Second, compared with the traditional stereo matching algorithm based on the correlation coefficient, the time complexity of the improved graph cuts algorithm was a little higher, but the precision of the improved graph cuts algorithm was evidently high. This research can provide a basis for the construction of the precise digital elevation model based on the satellite stereo pair.