Aiming at the problem that the robustness and registration accuracy are difficult to be compatible in the existing image registration methods, an image registration algorithm using SURF feature and local cross-correlation information was proposed. Firstly, the SURF feature extraction method was used for preliminary rough registration to improve the robustness of registration. Then, the homography matrix was calculated by using the correlation coefficient of local key areas in the image. Finally, the homography matrix was applied to the rough registered image results for rotation transformation, so as to realize high precision and robust image registration. The experimental results show that the proposed registration method is compared with the image registration method based on SIFT, ORB, SURF and cross-correlation information on multiple groups of data, which not only shows higher registration accuracy and efficiency, but also shows better robustness.