Super-resolution ghost imaging through complex scattering in dynamic media

Abstract

Super-resolution imaging has attracted much attention in various fields due to its capability to reveal fine structures beyond the diffraction limit. In this paper, super-resolution ghost imaging (GI) through complex scattering media is reported using neural networks with a physical model and the priors of a diffusion model. Dual deep image priors (DIPs) incorporated with a GI formation model are adopted to overcome the challenge posed by complex scattering media. With the designed dual DIPs, effective object information can be retrieved using the realizations and speckle patterns without any datasets or labels. A super-resolution model, fine-tuned from a large pre-trained stable diffusion model, is further designed to recover a high-resolution object image beyond the diffraction limit. Experimental results demonstrate that the developed GI can be applied to address complex scattering in dynamic media and achieve a ∼2.4-fold resolution enhancement beyond the diffraction limit. It is also illustrated that the proposed method pushes the boundaries of optical imaging in complex scenarios.