Deep learning-based distance estimation in incoherent digital holography

Incoherent digital holography (IDH) can generate holograms using spatially incoherent light; however, distance information cannot be directly extracted from holograms. Herein, we propose deep learning-based methods for fast distance estimation in IDH. To efficiently and accurately generate datasets, we construct an IDH simulator that reflects a real-world IDH setup. Moreover, we propose a preprocessing technique that compensates for the phase offset inherent in the IDH system and enhances hologram contrast to effectively extract features of holograms. We numerically and experimentally demonstrate the fundamental principles and effectiveness of the proposed methods. This technique facilitates fast and accurate acquisition of three-dimensional information in IDH.