Abstract: In order to further improve the robustness, speed and adaptability of the image registration algorithm, a fast and robust image registration algorithm based on contrast Harris was proposed. Firstly, the contrast of the block image was calculated according to the difference of the gray value between the central pixel and its neighborhood pixels, and the threshold of the corner detection was determined adaptively. The pseudo corner points were removed by the gray similarity. Then the corner points were detected in the constructed scale space. As a result, the problems were solved, such as scale sensitive of corner points extracted with Harris algorithm, manual setting of threshold, existence of pseudo corner points, uneven distribution. Finally, the partial mismatched pairs of matched points were removed by slope and distance constraint and the matched pairs were finely aligned by random sample consensus. The experimental results show that, compared with four registration algorithms, the proposed registration algorithm possesses better robustness to JPEG compression, blurred images, visual angle, illumination, rotation and scale changes. The accuracy of registration is higher, the degree of adaptation is stronger and the time of registration can be reduced significantly.