Object image reconstruction: method for reconstructing images from digital off-axis holograms using a generative adversarial network

The article considers the use of digital holography in reconstructing object images from different 3D scene cross-sections. This reconstruction enables the study of different materials, characterization of microparticles in a medium, and analysis of microplastic content in water bodies. A method is proposed for reconstructing object images from digital off-axis holograms using a generative adversarial network (GAN). The generative adversarial network is used to reconstruct 3D scene cross-sections in which off-axis objects are present. The application of neural networks is shown to improve the speed and quality of reconstruction, as well as to reduce image noise. The proposed method was tested on numerically synthesized and optically measured digital holograms. By means of this method, eight 3D scene cross-sections were reconstructed using a single synthesized hologram. An average structural similarity index measure of at least 0.73 was obtained. In the study, the authors experimentally recorded sets of digital off-axis holograms of phase objects displayed on spatial light modulators to form the cross-sections of the 3D scene. In the reconstruction of object images using optically registered holograms, the average structural similarity index measure for the cross-sections of the scene amounted to 0.83. The proposed method enables a high-quality reconstruction of object images and will be useful in the analysis of micro- and macro-objects, including in biomedical applications, metrology, as well as characterization of materials, surfaces, and volume media.