Auto localization for coastal satellite imagery based on curve matching

In this paper, a novel framework for automatic localization of coastal satellite imagery is proposed. To deal with multimodal and multitemporal situations, we use the salient information in coastal images, and thus coastline contour is extracted as the feature to be used for localization. First, Fast Marching Method (FMM) is adopted to get coastline curve. Then, we represent each curve by a scale invariant signature based on the curvature integral, and normalized cross-correlation is used to find the point correspondences between the matched parts along the two curves. Finally, transformation is estimated accordingly. The proposed method has been successfully applied to localize multimodal and multitemporal Landsat and SPOT images from coastal area. Experimental results of different test datasets demonstrate the effectiveness, efficiency and robustness of the proposed method.