Modeling and removal of correlated noise using nonlocal patch-based collaborative filters, with applications to direct and inverse imaging

Thursday, October 17, 2019 - 11:20am - 12:05pm
Keller 3-180
Alessandro Foi (Tampere University of Technology)
Noise in imaging systems rarely conforms to the simple IID additive white Gaussian noise model. This talk starts with a brief overview of alternative noise models that can be adopted in practical applications. We emphasize two noise models: signal-dependent variance models and stationary spatially correlated models. We particularly focus on the latter models, and thoroughly explore how to deal with them through nonlocal patch-based collaborative filters such as BM3D. We then discuss the role of these noise models and filters as a versatile regularization prior for solving inverse imaging problems under the plug-and-play framework.