Single-shot deep-learning based 3D imaging of Fresnel incoherent correlation holography

Abstract

Fresnel incoherent correlation holography (FINCH) is a non-scanning and non-coherent light 3D imaging technology that has the potential of axial super-resolution imaging. However, the large depth-of-field artifacts of out-of-focus information greatly limits the axial resolution of FINCH. Here, we propose a single-shot deep-learning based 3D imaging method of FINCH. First, a hologram is collected and back propagated to obtain holograms at different back propagation distances. Subsequently, the designed network is used to identify and remove zero-order, conjugate, and defocused image from the hologram, thereby obtaining 3D information of the sample without phase-shifting operation. The experimental results demonstrate that the proposed method can effectively remove the interference of zero-order and conjugate image in back propagation hologram, and the imaging quality is comparable to that of multi-step phase-shifting based FINCH technology. In addition, clear focused images at different distances on the z-axis can be obtained without interference from defocused images at different back propagation distances, indicating that the proposed method can greatly improve temporal and axial resolution of 3D imaging for FINCH.