Two-step differential phase-shifting phase retrieval using generative adversarial network

Abstract

Phase-shifting interferometry (PSI) is a high-accuracy optical measurement technique which requires accurate phase-shift estimation or sufficient phase-shifting interferograms to ensure phase-retrieval accuracy. Therefore, obtaining the highest phase retrieval accuracy from fewer interferograms remains a challenge. Based on this, we propose a two-step differential phase-shifting phase retrieval method using generative adversarial network, which can achieve higher phase-retrieval accuracy from two arbitrary phase-shifting interferograms, compared to representative two-step arbitrary phase-shifting phase retrieval methods. A phase retrieval generative adversarial network (PR-GAN) which can be trained with artificial training data for phase retrieval from experimental interferograms is employed. Experimental results demonstrate that accurate phase results can be obtained, even when the phase shift is approximately π or the fringe number is relatively small.