Ps-4net: a deep learning solution for in-line holography

Abstract

In-line holography has been widely used in various fields because of its advantages such as simple optical path, low requirement of light source coherence and low utilization of the camera spatial bandwidth product, but it is difficult for in-line holography to restore object information from a single in-line hologram. The traditional phase-shifting algorithm requires at least three phase-shifting holograms, moreover the error caused by the intermediate phase-shifting may be accumulated and amplified. In recent years, deep learning has been widely used in the optical field due to its advantages in data analysis. Deep learning has provided new solutions for in-line holographic reconstruction, which can solve the problems that are difficult to avoid in traditional methods. In this paper, a method for generating phase-shifting holograms based on a modified Y-4net is proposed to reduce the experimental workload in data collection and the noise in phase-shifting image, which is referred to as Ps-4net. The proposed Ps-4net can generate four virtual phase-shifting fringe patterns from a single frame hologram, and calculate the phase from virtual phase-shifting holograms. Simulation and experimental results show that the Ps-4net can effectively reduce the workload of data collection and the phaseshifting hologram is generated and the noise is removed at the same time.