Fully self-supervised physics-aware holographic depth estimation

Area of Science:

• Optics and Photonics
• Computational Imaging
• Computer Vision

Background:

• Autofocusing in holography lacks a universal method adaptable to varying experimental parameters like wavelength and pixel pitch.
• Existing approaches, including mathematical and learning-based methods, often require manual hyperparameter tuning and struggle with diverse holographic data.

Purpose of the Study:

• To introduce a universal holographic autofocusing methodology that eliminates manual hyperparameter tuning.
• To develop a robust and adaptable approach for autofocusing diverse input holograms, including synthetic and real-world data.

Main Methods:

• Autonomous extraction of optimal numerical reconstruction distances for holograms.
• Volumetric rendering of holograms to estimate underlying scene geometry.
• Iterative hologram regeneration process constrained by ground-truth depth values for accuracy.

Main Results:

• Demonstrated superior robustness and generalization capabilities on both synthetic computer-generated holograms and optically acquired phase-shifting holograms.
• Achieved precise autofocusing by autonomously estimating scene geometry and refining depth maps.
• Eliminated the need for manual hyperparameter tuning, offering adaptability to diverse experimental setups.

Conclusions:

• The proposed universal methodology represents a significant advancement toward reliable autofocusing in various holographic applications.
• The approach offers a robust solution for holographic autofocusing, overcoming limitations of existing methods.
• This work paves the way for more accessible and versatile holographic imaging systems.