Workshop on Mathematical Models for Plug-and-play Image Restoration

Abstract

Plug-and-Play (PnP) methods constitute a class of iterative algorithms for imaging problems where regularization is performed by an off-the-shelf denoiser. Specifically, given an image dataset, optimizing a function (e.g. a neural network) to remove Gaussian noise is equivalent to approximating the gradient or the proximal operator of the log prior of the training dataset. Therefore, any off-theshelf denoiser can be used as an implicit prior and inserted into an optimization scheme to restore images. After introducing the PnP and Regularization by Denoising (RED) frameworks, we will explore the different convergence analyses that have been proposed in the literature. From monotone operators theory to convex and non-convex analysis, we will propose various tools and deep denoisers that allow theoretical PnP convergence guarantees and state-of-the-art IR performance.

Samuel Hurault

Mathematics, Machine Learning, Computer Vision