Zero-order term suppression in off-axis holography based on deep learning method

Abstract

Zero-order image is the main source of noise in the hologram acquisition process. In this paper, a preprocessing method based on deep learning are proposed. Generative adversarial network is applied to suppress zero-order image of off-axis digital holograms. To simplify data preparation process, we propose a digital hologram corresponding database generation method. The simulated data is used to replace the experimental data in the training network. After testing, the experimental data can be processed into a hologram without zero-order image successfully. And the different types of images are reconstructed with high quality. This method shows high efficiency and excellent robustness.