An automated registration of RS images based on SURF and piecewise linear transformation

Image registration is an important procedure for multi-source image applications, such as image fusion, object detection, change detection, etc. This paper presented an automatic registration approach based on Speed Up Robust Features (SURF) and piecewise linear (PL) transformation. SURF is based on the scale-space representation of an image, and is invariant to image scale, rotation, illumination, and affine transformation. By using PL transformation, images will be divided by triangulated irregular networks (TIN). Different affine transform will be made in every triangle so that the local distortion can be processed. The parallel implementation of SURF and PL was also presented in this paper for accelerating the speed of registration procedure. The results of experiments show that the presented approach can achieve high accuracy and satisfy the real-time demand.