NL-Ridge: a novel statistical patch-based approach for image denoising
Sébastien Herbreteau  1@  , Charles Kervrann  2  
1 : SERPICO Project-Team Inria Centre Rennes - Bretagne Atlantique. UMR144 CNRS Institut Curie, PSL Research University, Sorbonne Université.
Institut National de Recherche en Informatique et en Automatique, CNRS : UMR144
2 : SERPICO Project-Team Inria Centre Rennes - Bretagne Atlantique. UMR144 CNRS Institut Curie, PSL Research University, Sorbonne Université.
L'Institut National de Recherche en Informatique et e n Automatique (INRIA), CNRS : UMR144

We present a novel statistical patch-based approach for image denoising. The state-of-the-art unsupervised methods that only use a single noisy image are two-step algorithms. Leveraging the Stein's unbiased risk estimate (SURE) for the first step and the "internal adaptation", a concept borrowed from deep learning theory, for the second one, we show that our NL-Ridge approach enables to reconcile several previous patch-based methods for image denoising. In the second step, our closed-form aggregation weights are computed through multivariate Ridge regressions. Experiments on artificially noisy images demonstrate that NL-Ridge may outperform state-of-the-art unsupervised denoisers such as BM3D and NL-Bayes, and recent unsupervised deep learning methods such as Noise2Self, Self2Self, and Deep Image Prior as well as supervised techniques such as DnCNN, while being much more simple conceptually.


Personnes connectées : 12 Vie privée
Chargement...