Randomness assisted in-line holography with deep learning.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Conventional holography often suffers from low image quality and twin-image artifacts.
• In-line holography simplifies setup but is prone to twin-image noise, hindering quantitative analysis.

Purpose of the Study:

• To develop and demonstrate an advanced holographic imaging scheme for high-quality, quantitative image reconstruction.
• To resolve the twin-image issue inherent in in-line holography using a novel approach.

Main Methods:

• Hologram recording using random illuminations and second-order intensity correlation.
• Numerical reconstruction of holograms.
• Unsupervised deep learning (auto-encoder) for blind, single-shot twin-image removal and reconstruction.

Main Results:

• The proposed method achieves high-quality quantitative image reconstruction, surpassing conventional in-line holography.
• The auto-encoder based twin-image removal effectively resolves artifacts without requiring ground truth training data.
• Experimental validation on two objects demonstrates the technique's efficacy and superior reconstruction quality.

Conclusions:

• The developed holographic imaging scheme offers a robust solution for high-fidelity 3D imaging.
• Unsupervised deep learning provides an efficient and data-independent method for twin-image removal in holography.
• This technique advances quantitative phase imaging and holographic microscopy applications.