UTIRnet

UTIRnet is a supervised universal convolutional neural network for twin-image effect removal in digital in-line holographic microscopy (DIHM)

Contents and codes

• main_example.m - main code with examples how to: (1) generate network training data, (2) train UTIRnet and (3) reconstruct holograms with UTIRnet
  
• AS_propagate_p.m - function for optical field propagation with angular spectrum (AS) method
  
• GenerateDataset.m - function for generating a whole dataset for UTIRnet training
  
• GenerateHologram.m - function for generating a single hologram and a pair of input-target images for network training
  
• NetworkArchitecture.m - function for creating CNN architecture
  
• UTIRnetReconstruction.m - function for reconstructing holograms with the UTIRnet
  
• ./Holograms - directory where experimental holograms may be stored (see Experimental data section)
  
• ./Networks - directory with two trained UTIRnets that we employed in our article (see Cite as section)
  

How does it work

Follow the steps in main_example.m code to generate synthetic training data and then to train UTIRnet network for a specified system parameters (wavelength, pixel size, magnification, sample-focus plane distance (or sample-camera distance in lensless DIHM system)). Then, generated network (composed from CNN_A and CNN_P networks) along with AS propagation may be used to reconstruct holograms without twin-image effect.

Experimental data

Our experimental data (holograms and reference reconstructions) may be found at:
M. Rogalski, P. Arcab, L. Stanaszek, V. Micó, C. Zuo, and M. Trusiak, “Physics-driven universal twin-image removal network for digital in-line holographic microscopy - dataset,” Jun. 2023, doi: 10.5281/ZENODO.8059636.
https://zenodo.org/record/8059636

Cite as

M. Rogalski, P. Arcab, L. Stanaszek, V. Micó, C. Zuo, and M. Trusiak, “Physics-driven universal twin-image removal network for digital in-line holographic microscopy,” Opt. Express, vol. 32, no. 1, p. 742, Jan. 2024, doi: 10.1364/OE.505440.

Created by

Mikołaj Rogalski,
[email protected]
Institute of Micromechanics and Photonics,
Warsaw University of Technology, 02-525 Warsaw, Poland