Diffraction-Net: a robust single-shot holography for multi-distance lensless imaging.

Area of Science:

• Optics and Photonics
• Digital Imaging
• Machine Learning Applications

Background:

• Lensless digital holography is crucial for microscopy and micro-scale measurements.
• Traditional methods require multiple images for complex-amplitude retrieval, limiting speed and introducing errors.
• Existing single-shot deep learning methods lack generalization ability and stability.

Purpose of the Study:

• To develop a robust single-shot digital holography method using deep learning.
• To enhance the generalization ability and stability of lensless imaging techniques.
• To enable accurate complex-amplitude recovery from a single diffraction image.

Main Methods:

• A diffraction network (Diff-Net) was designed to connect diffraction images at varying distances.
• A hybrid approach combining a physical model with deep learning was employed.
• An iterative complex-amplitude retrieval algorithm utilized Diff-Net generated images and light transfer functions.

Main Results:

• The Diff-Net based single-shot holography demonstrated robustness by eliminating practical errors between multiple images.
• The hybrid iterative approach successfully recovered complex-amplitude information.
• Experimental results confirmed the Diff-Net's qualified generalization ability across diverse sample morphologies.

Conclusions:

• The proposed Diff-Net offers a stable and robust single-shot solution for digital holography.
• This hybrid deep learning and physical model approach overcomes limitations of previous lensless imaging techniques.
• The method shows significant potential for advanced microscopy and micro-scale metrology.