Fast robust perspective transform estimation for automatic image registration in disaster response applications

Jim Thomas, Ahsan Kareem, Kevin W. Bowyer

Abstract: While automatic image registration has been extensively studied in other areas of image processing, it is still a complex problem in the framework of remote sensing for disaster response. This problem is difficult because there can be substantial change in the image content between the two images, and the time of day and lighting typically are different between the two images. In this work, we propose a two-step approach to achieve fast and robust registration of before- after-disaster aerial image pairs. First, the images are coarsely registered using a phase-correlation based algorithm. In the second step, transformed images are finely registered by matching features across grids and estimating the perspective transform. Our proposed algorithm is evaluated for robustness, accuracy and speed. It is found to achieve 100% registration success on 23 image pairs which proved challenging to either of the component approaches.