Robust multi-scale image matching for deriving ice surface velocity field from sequential satellite images

The cross-correlation-based image matching method has been widely used to derive ice surface motion information from sequential satellite images through tracking spatial displacements of surface features over time. However, this conventional method is not adequate for handling areas with a high velocity variation, in which case a large search window has to be specified in order to find the correct match points for fast-moving features. The computation involved in the cross-correlation matching with a large search window is often so intensive that only a sparse set of velocity measurements can be attempted. Furthermore, with a greater search window the cross-correlation method is prone to more spurious matches. This article presents a robust multi-scale image matching method to address the deficiencies of the conventional cross-correlation technique. The main idea is to use approximate match points obtained at a coarse resolution as a guide for searching for more precise matches at a higher resolution. A robust local statistical operator is embedded at each scale in the multi-scale matching process to eliminate match outliers. The strategy of progressively refining the match precision from coarse-resolution images to the full-resolution image allows for a small search range in pixels. Our robust multi-scale matching method significantly speeds up the computation and also reduces the occurrences of bad and spurious match points. We have implemented our method as a software tool with a graphical user interface and successfully applied it to process sequential Radarsat synthetic aperture radar images for extracting high-resolution velocity fields for Antarctic glaciers and ice shelves. This software tool will be freely available to the public through the Internet.