An automatic global-to-local image registration based on SIFT and thin-plate spline (TPS)

For remote sensing applications, automatic image registration is an essential part for further image processing such as image fusion, change detection and so on. In this paper, we propose an effective and fast automatic image registration based on both global and local method. In the first stage, we extract SIFT features and make the affine transformation with RANSAC, a robust outlier removal method. After the image globally registered, uniform spacing control points are selected in the common area. Through template matching, control points are refined and more local to the image. At last thin-plate spline, a nonrigid image registration transformation function is applied to achieve local matching. The experiment results indicate that this method not only distinctive and robust in overall registration but also significantly improve matching performance in local area1.