Refining scan positions in Ptychography through error minimisation and potential application of Machine Learning

Ptychography is an advanced technique whose latest developments have allowed for exceptionally high resolution in X-ray microscopy, as well as new setups that can even be lensless; the spatial resolution has already reached < 5 nm in synchrotron beamlines. Still there are certain demanding requirements such as beam coherence and strong involvement of computational methods that make it a rather difficult technique to implement. In this paper we introduce a computational work-flow aiming at refining the individual probe positions of a ptychography scan. Indeed, the precision of those positions is of high importance as it impacts the reconstruction. Special setups using precise sample stages with nanopositioning and advanced interferometers are important; however, positioning errors still remain a problem for many operating laboratories. This paper examines the Structural Similarity Index as a suitable metric for evaluating the registration in alignment of individually reconstructed probes. Eventually it suggests the use of Machine Learning techniques as future direction in probe alignment. It also presents a new software tool for precise manual alignment and visualization. All the software developed during this research project is provided to the scientific community as open source.