Photon-Limited Deblurring Using Algorithm Unrolling

Abstract

Image deblurring in a photon-limited condition is ubiquitous in a variety of low-light applications such as photography, microscopy and astronomy. However, presence of photon shot noise due to low-illumination and/or short exposure time makes the deblurring task substantially more challenging . This paper presents an algorithm unrolling approach for the photon-limited deblurring problem that unrolls a Plug-and-Play algorithm using a fixed-iteration network. By modifying the typical two-variable splitting to a three-variable splitting, our unrolled network is differentiable and can be trained end-to-end. We demonstrate the usage of our algorithm on real photon-limited image data.