Ghost imaging in highly complex scattering environments

It is well recognized that optical imaging still lacks an investigation of dynamic scattering media with time-varying perturbations on both sides of an object, i.e., the simultaneously disturbed illumination and detection paths. In this paper, we report a physics-enhanced untrained neural network (UNN) framework enabling high-quality ghost imaging (GI) through complex media. The experimental system employs rotating diffusers on both sides of the object to generate dynamic scattering environments, combined with a tunable-turbidity liquid chamber under mechanical agitation during the imaging process. A joint-optimized UNN framework with a GI physical model is designed to compensate for dynamic scattering distortions without any training datasets. Optical experiments demonstrate the effectiveness of the proposed method to obtain high-quality reconstruction under dual-path scattering conditions. It is believed that the proposed method can provide an insight into optical imaging through scattering media where illumination and detection paths are severely disturbed.