Computational ghost imaging with PSF-guiding deep learning through various unknown turbid scattering media

Abstract

Achieving high signal-to-noise ratio (SNR) imaging through scattering media is challenging. Computational ghost imaging with deep learning (CGIDL) has unique advantages for solving this challenge. However, image reconstruction cannot be guaranteed due to low correlation between real signal and training dataset, when the CGIDL is applied in different unknown scattering media. Point spread function (PSF) determines the quality of CGIDL reconstruction, linking the mathematical features of the scene and the quality of reconstruction. In this study, an innovative CGIDL technology based on the measured PSF method is proposed to improve the correlation between real signal and training dataset. When five new turbid scattering media with unknown scattering strength are introduced, classification of PSF enables high SNR imaging through various turbid scattering media.