Ghost imaging through multiple scattering media at low sampling rates via an enhanced UNet

Area of Science:

• Optics
• Image Processing
• Computational Imaging

Background:

• Imaging through scattering media is a significant challenge in various scientific fields.
• Traditional methods struggle with noise and detail loss when imaging in scattering environments.
• Differential ghost imaging (DGI) offers a potential solution but requires improvements for practical applications.

Purpose of the Study:

• To introduce GI_UNet, an advanced framework for imaging through scattering media.
• To enhance the performance of DGI by integrating Monte Carlo photon-transport modeling and an improved UNet architecture.
• To demonstrate the robustness and effectiveness of GI_UNet under challenging imaging conditions.

Main Methods:

• Developed a DGI framework (GI_UNet) combining Monte Carlo photon-transport modeling with an enhanced UNet.
• Utilized high-fidelity point spread functions (PSFs) to predict measurement matrices of trainable speckles propagating through scattering media.
• Employed an enhanced UNet with optimized components to suppress scattering noise and restore image details.

Main Results:

• GI_UNet demonstrated superior performance compared to existing mainstream methods in both simulations and experiments.
• The framework showed robustness in imaging through scattering media.
• Effective image restoration was achieved even at low sampling rates.

Conclusions:

• GI_UNet provides an effective solution for scattering-limited imaging scenarios.
• The integration of DGI, Monte Carlo modeling, and enhanced UNet significantly improves imaging quality.
• This framework offers a promising approach for applications requiring imaging through optically dense media.