Automatic high resolution satellite image registration by combination of curvature properties and random sample consensus

Image registration is the process of overlaying images (two or more) of the same scene taken at different times, from different viewpoints, and/or by different sensors. The registration geometrically aligns two images (the reference and sensed images). We propose in this paper an automatic feature-based approach to high resolution satellite image registration. The algorithm consists of two phases. The first one is the automated generation of control points. For the feature detection we use an curvature-based corner detector that detects both fine and coarse features accurately at low computational cost. The feature matching phase is carried out using a bidirectional correlation approach leading to a robust set of control points in the multi temporal remote sensing images. The second one is the robust estimation of mapping functions from control points. We used the random sample consensus (RANSAC) algorithm for this step. We argue that it is the second step which gives the robustness of any automated registration algorithms. RANSAC was applied to the control points. All outliers were correctly identified and mapping functions estimated without outliers. Finally, a global linear spatial transformation is applied and the remotely sensed image is efficiently registered. The results support that our algorithm can be used for robust automated registration.