An adaptive image registration method based on SIFT features and RANSAC transform

Scale Invariant Feature Transform (SIFT) is one of the most applicable algorithms used in the image registration problem for extracting and matching features. One of the efficient methods in reducing mismatches in this algorithm is the RANdom Sample Consensus (RANSAC) method. Besides the applicability of RANSAC, its threshold value is fixed, and it is empirically chosen. In this paper, a mean-based adaptive RANSAC is proposed at first. In this method, the threshold value of RANSAC is chosen based on the mean of distances between each point and it's model-transformed one. To increase the capability of the method, the second adaptive RANSAC method is proposed, which exploits the variance of the distances in addition to the mean value. Simulation results confirm the superiority of the proposed methods in comparison with classic ones in terms of True Positive rate, mismatches ratio, total number of matching, and two newly proposed evaluation criteria.