Diffusion-powered plug-and-play reconstruction for full-color ghost imaging through dynamic scattering media

Abstract

Color imaging through dynamic scattering media is challenging. Although ghost imaging (GI) has been studied, the distorted single-pixel measurements hinder high-fidelity reconstruction. Here, we report a full-color GI framework that eliminates wave distortions caused by dynamic scattering media to achieve high-quality reconstruction. A convolution-based polynomial estimation method is designed to estimate and correct dynamic scaling factors. Then, a diffusion-powered plug-and-play algorithm is designed and applied, using a pre-trained diffusion prior to regularizing chromatic structures with a GI-based physical model to maintain data consistency. Experiments are conducted to demonstrate that the developed full-color GI framework exhibits high robustness and superior performance, always achieving high-fidelity reconstruction in dynamic scattering environments. Our method provides a promising solution for full-color GI in harsh environments, showing excellent stability and application potentials.