The Maximum Entropy on the Mean Method for Image Deblurring

Published in Inverse Problems, 2020

This paper extends our earlier work on QR code deblurring to the case of general images. We develop a deblurring algorithm that achieves strong results on images that include calibration patterns and in cases where the blur kernel is known in advance.

Authors

Gabriel Rioux, Rustum Choksi, Tim Hoheisel, Pierre Maréchal, Christopher Scarvelis

Technical summary

As in our earlier paper “Blind Deblurring of Barcodes via Kullback-Leibler Divergence”, we use Fenchel-Rockafellar duality to transform an infinite-dimensional optimization problem over a space of probability measures into a smooth, convex, and finite-dimensional dual problem. We provide a novel stability analysis for our method and examine the role of the prior measure in our problem.

Download paper here