Imaging through dynamic scattering media via deep unpaired data-driven single pixel detection

Abstract

Imaging through dynamic scattering medium is of great significance yet challenging topic in many fields. Computational single pixel imaging is capable of encoding object’s information into a sequence of intensity values by modulation and only a single-pixel detector is required to collect the relative distribution of intensity values. However, due to the perturbation of scattering media of dynamic scenes, the actual collected intensity values will deviate from the ideal intensity values, resulting in poor reconstructed quality. In this work, we present a new method for high clarity imaging through dynamic scattering scene based on single pixel detection and unpaired data-driven deep learning. Preliminary images are recovered firstly using measurement data. Then a cycle generative adversarial network trained on unpaired datasets is employed to improve the quality of the initial recovered images. Simulations and experimental results verify the effectiveness of the proposed method and shows good reconstructed quality at low sampling rate.