Image Registration Based On a Novel Enhanced Scale Invariant Geometrical Feature
Ehsanhosein Kalatehjari, Mohammad Mehdi Hosseini, Abdorreza Alavi Gharahbagh · International Journal of Machine Learning and Computing · 2012
Image registration can find similarities between one image and another one from the same scene taken by different angles.Registration is widely used in robot localization, remote sensing, medical imaging and etc.In this paper, a new method based on the image geometrical features has introduced by using two important characteristics of image scale and image rotation.Feature points of two images by SIFT algorithm has extracted.Then, an initial matching is estimated based on descriptor matrix of SIFT features with nearest neighbor (NN) criterion.The novel geometrical method is used for discarding incorrect matched points.Finally, the correct image is recognized by the number of matched points.