Abstract

White light phase shifting is an important technique in interferometry to extract the high-resolution quantitative phase images with high spatial phase sensitivity i.e., of the order of sub nanometer. The dynamical information about a biological sample is limited in white light phase-shifting interferometry (WL-PSI) due to multiple frame requirement. The multiple frame requirement with controlled phase-shift is the key limitation for the high-resolution phase information extraction about the sample. A high-cost piezoelectric transducer (PZT) is required to introduce equal phase-shifts between the frames in WL-PSI. Here, we introduce a deep learning (DL)-based phase-shifter instead of PZT to introduce equal phase-shift in WL-PSI. We use deep neural network to introduce the equal phase-shift between the data frames. The idea is to train the network with multiple equal phase-shifted frames for training and learn the basic application of a PZT. After sufficient training of the network, it will generate multiple phase-shifted frames. Our study is validated by simulating step-like object with equal phase-shifts for training and testing of the network. The network is trained for a total 4 phase-shifted frame generation from a single interferogram. Further, the line profile of DL-based phase-shifter generated data frames are compared with the line profile of simulated data frames. Finally, generated phase-shifted data frames are used for the final phase reconstruction of the sample and compared with the phase reconstruction from simulated data frames.

© 2023 SPIE