Phase Retrieval Method for Phase-shifting Interferometry with Machine Learning

Phase contrast imaging attracts attention in the biomedical field thanks to higher contrast than absorption contrast. However, the phase retrieval for phase-shifting interferometry (PSI) often involves the problems, e.g., the noise, stepping error and phase wrapping. In the conventional way, each issue had to be addressed individually. In this paper, we propose the machine learning based method, which uses the neural network and learns features in an end-to-end manner. The proposed method can resolve the noise, stepping error artifacts and phase wrapping, simultaneously. The results are shown by numerical simulation.