Feature based image registration

In this study, a curve based feature extraction method is proposed in order to image matching and registration. For this purpose, geometric distance between curves are computed by using Hausdorff distance method. Later, Harris points on curves are determined and matched as a second matching. Once the best matched curves obtained, the geometric transformation between the reference and target images is estimated by using RANSAC method and images are registered. The widely known database which is proposed by Mikolajczyk and Schmid is carried to obtain experimental results. The comparison of results are done based on precision and recall criteria. The results of this study are compared with results of SIFT, SURF, MSER, BRISK, FREAK, ORB, FAST and BRIEF feature and descriptor extraction methods. By looking the results, while a decline is observed on performance of other methods, the proposed method is able to survive its performance in case of rotation and zoom, blurring, viewpoint, illumination, JPEG compression changes and high level noise conditions.