Multi-modal aerial image registration using spatial structure

Image registration is an important step prior to data fusion from multiple imaging sensors. However, it is challenging because the appearance of the same scene differs in multimodal images. We introduce a robust registration method that establishes landmark correspondences when many outliers exist due to different image appearances in multi-modal images. The proposed matching method leverages the spatial order of detected landmarks that is preserved despite the non-linear deformation in remote-sensing data. We demonstrate that our matching algorithm is robust to outliers in real multi-modal image pairs.