Image Processing for Talbot-Lau X-ray Imaging
X-ray imaging traditionally relies on measuring how much radiation is absorbed as it passes through an object. In recent years, a complementary approach has emerged: measuring how X-rays are refracted, using devices called Talbot-Lau interferometers. These consist of three finely structured gratings placed in the X-ray beam and can be retrofitted into many existing X-ray systems. Beyond refraction, they also produce a “dark-field” signal that reflects how uniform the imaged material is, but getting usable images out of this setup is considerably more complex than with standard X-ray imaging.
This thesis improves Talbot-Lau image quality through two lines of work. The first tackles artifacts caused by unwanted motion of the gratings, both during an acquisition and in the time between a calibration scan and the actual measurement. New correction methods were developed for each case, including approaches based on spatial and temporal correlation, one of which removes the need for a prior calibration altogether. These corrections make it possible to operate a Talbot-Lau interferometer reliably even under difficult mechanical conditions, and accompanying improvements to the preprocessing pipeline further reduce noise and improve accuracy of the dark-field signal at low X-ray doses.
The second line of work addresses a different problem: converting the raw refraction measurements into an actual refractive index requires integrating the data, which sharply amplifies noise. To manage this, the thesis introduces an efficient numerical optimization framework for combining and evaluating noise-suppression strategies. It also investigates how the orientation of the grating structures affects noise in tomographic reconstructions, and uses these findings to design and experimentally validate noise-suppressing filters.