Image matching using a local distribution based outlier detection technique

In this paper, an efficient image matching algorithm for finding the consistent correspondences between two sets of image feature points has been presented. Correct assignments are usually compatible with each other, and thus likely to form a strong cluster. The main idea of the proposed algorithm is to detect this cluster using a local distribution based outlier detection technique. Based on neighbor similarity (or affinity), we first define an inlier score for each assignment in candidate assignment set. Then, we iteratively detect the correct assignments from the candidate assignment set by exploiting the inlier score. Experimental results on several real-world image matching tasks demonstrate the effectiveness and robustness of the proposed algorithm.