Robust phase correlation based feature matching for image co-registration and dem generation. The

This paper presents a robust phase correlation based sub-pixel feature matching technique and its application in motion flow estimation, pixel-to-pixel image-co-registration and DEM generation. We propose to use a phase fringe filter and a highly robust technique in the direct Fourier-based phase correlation algorithm for translational shift estimation in sub-pixel accuracy. Noting the problem that local phase correlation based feature matching may fail in areas either featureless or with significant spectral differences between an image pair, a direct frequency based motion estimation assessment technique and a novel motion flow refinement scheme are designed to improve the unreliable local motion estimates around these areas. With the robust phase correlation based local matching algorithm, we are able to derive accurate pixel-to-pixel image co-registration and disparity mapping for DEM generation in most synthetic and real images from different sensor platforms or different spectral bands