For aerial images, the SURF algorithm is applied to image feature extraction and matching. In this paper, a novel compensation matching algorithm is proposed to improve the matching accuracy. According to the transformation matrix estimated by feature matching results from SURF algorithm, RANSAC algorithm is used to compute transformation matrix iteratively. Then under the condition of epipolar constraint, it makes use of transformation matrix for feature points compensation matching. Finally using different algorithm for transformation matrix, combined with image matching point pairs, registration experiment can be conducted with a set of actual aerial images. Experiment shows that, the algorithm in this paper can greatly improve registration accuracy in a relatively short time, having accuracy more than 90%, which provides a new train of thought for the aerial image matching technology.