Ghost imaging through complex scattering media with random light disturbance

Abstract

Imaging in a complex environment is recognized to be challenging in various applications. Imaging with single-pixel detection, e.g., ghost imaging (GI), emerges as a solution in recent years. Here, we report a unified GI framework based on untrained neural networks (UNNs) to eliminate the effect of complex environments and realize high-resolution object reconstruction. Two UNNs are designed to respectively estimate the corrected realizations and a series of dynamic scaling factors from the collected realizations. A GI-formation-based physical model is incorporated into the network to ensure the validity of the corrected realizations and enable object reconstruction. Experimental results demonstrate that the proposed method is effective and robust for high-resolution and high-contrast object reconstruction in complex environments, i.e., dynamic scattering media with high-randomness light disturbance. In addition, the proposed method is validated at low sampling ratios to alleviate data acquisition burden. With the advantages in the integration, adaptability, and efficiency, the proposed method provides a promising solution for GI in complex environments.