Efficient and robust image registration for two-dimensional micro-X-ray fluorescence measurements

X-ray fluorescence spectrometry (XRF) is a technique that allows determining non-destructively the composition of elements within a sample. Focussing the excitation X-ray beam to a small spot that is moved in the x-y-direction relative to the sample adds lateral information. Such a two-dimensional micro-X-ray fluorescence (2D mu-XRF) spectrometer for desktop use is commercially available providing a resolution down to approximately 10 mu m. With a mu-XRF spectrometer, it is inexpensive to take many scans of the same sample. With super-resolution methods, these can potentially be combined into a higher-resolution image. As a prerequisite, the misalignments of multiple scans (shifts and rotation) in the subpixel range have to be detected. We present a method for image registration of multiple images based on expander graphs that provides adjustable tradeoffs between registration quality and running time. We evaluate the algorithms on artificial and real mu-XRF data and we argue that our findings show that subpixel information is present in real mu-XRF data. This is a necessary condition for the applicability of multi-image super-resolution techniques to mu-XRF data in future work.