Multi-Scale Image Registration Algorithm based on Improved SIFT

This paper proposes an image registration algorithm based on improved SIFT feature. In the feature detection, evenly distributed feature points are detected using non-maximum suppression and reduces detection times by dynamic adjustment of detecting step. While employs feature matching, the threshold of Euclidean distance ratio is adaptively determined according to the properties of images. Lastly, the error matching points are removed by geometric constraint consistency between the feature points through RANSAC algorithm. Experimental results show the effectiveness of the proposed algorithm.