Learning-based complex field recovery from digital hologram with various depth objects.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Digital holography enables 3D reconstruction but often struggles with noise and object variations.
• Existing learning-based methods typically reconstruct in the object plane, limiting robustness to depth and amplitude variations.

Purpose of the Study:

• To develop a learning-based complex field recovery technique that directly extracts object information from the hologram plane.
• To enhance the robustness of holographic reconstruction for 3D objects with significant amplitude variations.

Main Methods:

• A deep learning network was designed to directly process hologram data.
• The network learns to extract the object's complex field in the hologram plane, bypassing object plane propagation.
• The method was validated using numerical simulations and optical experiments.

Main Results:

• The proposed technique directly recovers the complex field from digital holograms.
• It demonstrates robustness to object depth variations and is suitable for 3D objects.
• The method successfully handles objects with large amplitude variations, unlike previous approaches focused on transparent samples.

Conclusions:

• The developed learning-based technique offers a robust and versatile approach for complex field recovery from digital holograms.
• This method expands the applicability of digital holography to more complex objects and scenarios.
• The feasibility is confirmed through simulations and experimental validation.