Phase-shifting interferometry from single frame in-line interferogram using deep learning phase-shifting technology

Abstract

Phase-shifting interferometry is a highly accurate phase measurement technology. Its application in dynamic phase measurement is limited by the recording process of multiple phase-shifting interferograms. In this paper, a one-to-multiple deep learning (DL) framework is developed to generate an equivalent to multiple image phase shifting interferometry, using only a single in-line interferogram. The network is trained to map the single frame interferogram to three phase-shifting interferograms. And then, all these four phase-shifting interferograms are used for the phase-shifting phase recovery. Simulation and experimental results prove that the proposed method can realize phase measurement in in-line interference system by recording single interferogram. And it has similar precision with traditional phase-shifting holography by recording four frames phase-shifting interferograms. Further, compared with the strategy of generating the phase distribution from single frame interferogram by trained DL framework directly, the DL phase-shifting strategy can reduce the network error in the non-linear mapping partly owing to the introduction of phase-shifting algorithm and the network only needs to deal with the mapping relationship between the interferograms with different phase shifts, showing higher accuracy and better generalization ability. The proposed method can provide a new alternative approach for the high-accuracy dynamic phase measurement.