An Image Registration Method Based On SIFT And SUSAN Features

An automatic image registration method base on mix SIFT and SUSAN features is proposed for improving the efficiency of corner detection using SIFT method. The local invariant feature of SIFT and SUSAN which are of complementary are apply to image registration.The method firstly use the SIFT method to detect the extreme space feature points and the SUSAN method to detect the corner points, then apply the features generation method of SIFT to generate the local invariant feature vector through the combination of the positions of two characteristic points. The initial matching points are determine by the ratio of the nearest neighbor feature points to the second nearest neighbor feature points; finally achieve the goal of image registration.Through the use of Ransac, LM and improve projection model parameter estimation algorithm to strike a projection matrix, to verify the validity of the algorithm. Through experimental results, come to the conclusion that the algorithm is effective and can improve the accuracy of Image Registration.