Abstract: FREAK algorithm has the defects of not having scale invariance and single feature point matching strategy, and being prone to unsatisfactory results. Based on SIFT and RANSAC algorithm, an improved FREAK algorithm is proposed: SFREAK (SIFT and FREAK). First of all, in the generation of Gauss differential Pyramid image, the feature points are detected with scale invariance; then the feature points are described with FREAK descriptor, obtaining binary descriptor; finally, in the process of feature points matching using Hamming distance matching for coarse matching, the matching points are purified with RANSAC algorithm, and the feature points of two images are matched. The experimental results show that the proposed algorithm can effectively solve the FREAK not having scale invariance in image scaling, and SFREAK algorithm for feature point matching accuracy rate reached 95.7%, increased by 61.9% compared with FREAK. Therefore, compared with the traditional SIFT algorithm and FREAK algorithm, the improved algorithm shows better robustness.

Key words: improved FREAK, SIFT, RANSAC, feature points matching