Abstract: Image stabilization is the image registration applied to one video image sequence from a single camera, which has been identified as the key first step for the task of multi-spectral image fusion for aerial surveillance applications. The stabilization of infra-red (IR) aerial image sequences is challenging owing to the low contrast and signal-to-noise ratio of the images, and to potentially large viewpoint changes that result in images with large rotation and scale change and perspective distortion. We demonstrate a new feature-based method for IR aerial image sequence stabilization. We use the greylevel differential invariant (GDI) matching due to Schmid and Mohr which is invariant to rotation and scaling. Extensions to the basic GDI method are introduced that improve the performance of the method. We use M-estimation for the image registration parameters. The method is robust to outliers returned by the GDI method. We verify the point correspondence under orthographic projection using the epipolar constraint. Experimental results are reported for real-world and synthesized IR image sequences.