Learning to Mean-Shift in O(1) for Bayesian Image RestorationMS52

Finding strong oracle priors is an important topic in image restoration. In this talk, I will show how denoising autoencoders (DAEs) learn to mean-shift in O(1), and how we leverage this to employ DAEs as generic priors for image restoration. I will also discuss the case of Gaussian DAEs in a Bayesian framework, where the degradation noise and/or blur kernel are unknown. Experimental results demonstrate state of the art performance of the proposed DAE priors.

This presentation is part of Minisymposium “MS52 - A Denoiser Can Do Much More Than Just... Denoising (2 parts)
organized by: Yaniv Romano (Technion - Israel Institute of Technology) , Peyman Milanfar (Google Research) , Michael Elad (The Technion - Israel Institute of Technology) .

Siavash Arjomand Bigdeli (EPFL)
Meiguang Jin (University of Bern)
Paolo Favaro (University of Bern)
Matthias Zwicker (University of Maryland)
deep learning, image reconstruction, image representation