Extendible ghost imaging with high reconstruction quality in strong scattering medium

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Ghost imaging (GI) faces challenges in scattering media where structured patterns degrade.
• Single-pixel imaging (SPI) frameworks struggle with scattering effects in both emission and reception paths.

Purpose of the Study:

• To present a deep learning (DL)-based ghost imaging method for enhanced performance in scattering environments.
• To numerically reproduce and address the degradation principles of scattering in optical imaging.

Main Methods:

• Developed a degradation-guided reconstruction (DR) approach using a convolutional neural network (CNN) trained on simulated datasets.
• Proposed a novel photon contribution model (PCM) with redundant parameters, implemented in a lightweight, two-branch CNN, to simulate scattering through volumetric media.

Main Results:

• The proposed DL-based GI scheme successfully recovers target semantics and suppresses noise in strong scattering conditions.
• Experimental verification demonstrated the effectiveness of the approach across various scattering coefficients and working distances.

Conclusions:

• Deep learning methods offer powerful solutions for computational imaging challenges, particularly in overcoming unanalyzable optical processes.
• Integrating optical principles with DL strategies enhances the robustness and applicability of ghost imaging in complex scattering scenarios.