Suppressing defocus noise with U-net in optical scanning holography

Optical scanning holography (OSH) records both the amplitude and phase information of a 3D object by a 2D scan. To reconstruct a 3D volumetric image from an OSH hologram is difficult, as it suffers from the defocus noise from the other sections. The use of a random phase pupil can convert defocus noise into speckle-like noise, which may require further processing in sectional image reconstruction. In this paper, we propose a U-shaped neural network to reduce this speckle haze. Simulation results show that the proposed method works effectively and efficiently both in simple and complex graphics.