Image registration combining accelerated segmentation feature algorithm and fast retina keypoint (FREAK) algorithm is proposed. Firstly, the scale space is constructed for the image, and the image feature points are detected by the accelerated segmentation feature optimization algorithm. Keypoints are filtered by Harris algorithm and some strong corners retained are reserved for image registration. Secondly, the strong corners are described by FREAK and eigenvectors are calculated. Keypoints are matched by Hamming distance instead of traditional Euclidean distance. Matches are filtered with random sample consensus algorithm to avoid mismatch due to noise and moving objects. From the two aspects of registration accuracy and registration time, the comparative experiments between scale-invariant feature transform, binary robust independent elementary features, original FREAK and the proposed algorithm are carried out. The experimental results show that the proposed algorithm has the characteristics of fast registration speed, high accuracy and well-stability.