Abstract

Subject of study. Image reconstruction from digital holograms using neural networks and quality enhancement of the obtained reconstructed images were considered. Aim of study. The study aimed to develop a method for 2D and 3D holographic image reconstruction. Method. The method is based on a selective extraction of amplitude and phase information from a digital hologram. The extracted amplitude and phase can subsequently be used to reconstruct images of objects or cross-sections of 3D scenes that are free from parasitic diffraction orders. Digital holograms are used as input information for the neural network. These holograms are then transformed into amplitude and phase components of the object wave at the network output. Main results. The developed method was used to reconstruct images from holograms of 2D and 3D scenes. The training set was composed of holograms, each of which contained one or several 2D scenes. In addition to the reconstruction of images, the method ensured suppression of parasitic diffraction orders. Based on all metrics, the proposed method demonstrated a higher reconstruction quality than that of the standard method based on the light propagation calculation. Practical significance. The proposed method can be applied for reconstruction of images from digital holograms. When an image from an in-line hologram is reconstructed, it can be entirely obscured by parasitic diffraction orders. While methods for suppression of these orders can be complex to use and/or require recording of several holograms, the developed method can directly extract the object information from a single hologram without additional filtering.

© 2022 Optica Publishing Group

Digital holographic particle volume reconstruction using a deep neural network

Tomoyoshi Shimobaba, Takayuki Takahashi, Yota Yamamoto, Yutaka Endo, Atsushi Shiraki, Takashi Nishitsuji, Naoto Hoshikawa, Takashi Kakue, and Tomoyosh Ito
Appl. Opt. 58(8) 1900-1906 (2019)

Holographic reconstruction enhancement via unpaired image-to-image translation

Romane Scherrer, Thomas Quiniou, Thierry Jauffrais, Hugues Lemonnier, Sophie Bonnet, and Nazha Selmaoui-Folcher
Appl. Opt. 61(33) 9807-9816 (2022)

Continuous-wave terahertz in-line holographic diffraction tomography with the scattering fields reconstructed by a physics-enhanced deep neural network

Xiaoyu Jin, Jie Zhao, Dayong Wang, John J. Healy, Lu Rong, Yunxin Wang, and Shufeng Lin
Photon. Res. 11(12) 2149-2158 (2023)

Automated single cardiomyocyte characterization by nucleus extraction from dynamic holographic images using a fully convolutional neural network

Ezat Ahmadzadeh, Keyvan Jaferzadeh, Seokjoo Shin, and Inkyu Moon
Biomed. Opt. Express 11(3) 1501-1516 (2020)

Automated red blood cells extraction from holographic images using fully convolutional neural networks

Faliu Yi, Inkyu Moon, and Bahram Javidi
Biomed. Opt. Express 8(10) 4466-4479 (2017)