Random two-frame interferometry based on deep learning

Abstract

A two-frame phase-shifting interferometric wavefront reconstruction method based on deep learning is proposed. By learning from a large number of simulation data based on a physical model, the wrapped phase can be calculated accurately from two interferograms with an unknown phase step. The phase step can be any value excluding the integral multiples of and the size of interferograms can be flexible. This method does not need a pre-filtering to subtract the direct-current term, but only needs a simple normalization. Comparing with other two-frame methods in both simulations and experiments, the proposed method can achieve better performance. (C) 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement