Deep-learning-based hologram generation using a generative model.

Area of Science:

• Optics and Photonics
• Computer Science
• Artificial Intelligence

Background:

• Digital holography enables 3D reconstruction of objects.
• Traditional hologram generation methods can be computationally intensive.
• Deep learning offers novel approaches for complex pattern generation.

Purpose of the Study:

• To propose a novel learning and inferring model for generating digital holograms using deep neural networks (DNNs).
• To demonstrate the capability of the proposed model to achieve generality across different spatial recording parameters.
• To validate the optical reconstructability of the generated holograms.

Main Methods:

• A generative adversarial network (GAN) was employed to infer complex 2D fringe patterns from single object points.
• Inferred intensity and fringe patterns were multiplied and accumulated for hologram generation.
• The model was trained and tested for hologram generation in 16 Space and 32 Space.

Main Results:

• The DNN model successfully generated digital holograms with high fidelity.
• Reconstruction results closely matched numerical computer-generated holograms, achieving 44.56 dB and 35.11 dB for the tested spaces.
• Optical reconstruction of the generated holograms was successfully demonstrated.

Conclusions:

• The proposed DNN-based method offers an efficient and effective approach for digital hologram generation.
• The model exhibits generality and can produce optically reconstructable holograms.
• This work highlights the potential of deep learning in advancing digital holography techniques.