Photon-limited imaging through scattering medium based on deep learning

Lei Sun,Jianhong Shi,Xiaoyan Wu,Yiwei Sun,Guihua Zeng
DOI: https://doi.org/10.1364/OE.27.033120
IF: 3.8
2019-01-01
Optics Express
Abstract:Imaging under ultra-weak light conditions is affected by Poisson noise heavily. The problem becomes worse if a scattering media is present in the optical path. Speckle patterns detected under ultra-weak light condition carry very little information which makes it difficult to reconstruct the image. Off-the-shelf methods are no longer available in this condition. In this paper, we experimentally demonstrate the use of a deep learning network to reconstruct images through scattering media under ultra-weak light illumination. The weak light limitation of this method is analyzed. The random Poisson detection under weak light condition obtains partial information of the object. Based on this property, we demonstrated better performance of our method by enlarging the training dataset with multiple detections of the speckle patterns. Our results demonstrate that our approach can reconstruct images through scattering media from close to 1 detected signal photon per pixel (PPP) per image. (C) 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
What problem does this paper attempt to address?