Non local approaches for image denoising
The subject of this practical session is patch-based image denoising. It covers PCA image denoising, inspired by the paper C.-A. Deledalle and J. Salmon and A. Dalalyan, Image denoising with patch based PCA: local versus global, proceedings of BMVC 2011. and DCT image denoising based on the algorithm described on Guoshen Yu, and Guillermo Sapiro, DCT image denoising: a simple and effective image denoising algorithm, Image Processing On Line, 1 (2011).
Image: PCA denoising (Notre Dame of Órleans)
Non local means for denoising
This practical session explains how to implement the Non local means algorithm for denoising images, introduced in 2005 by Buades, Coll and Morel in Buades, A., Coll, B., & Morel, J.-M. (2005). A review of image denoising algorithms, with a new one. Multiscale modeling & simulation, 4(2), 490-530.. The session describes various implementations of the algorithm: from the naive version, through an implementation using integral images, and finally the pytorch version.
Image: Pytorch NL means denoising (Saint Emilion)
Networks for image denoising
This session works with convolutional neural networks (CNN) to denoise images. Special emphasis is placed on FFDnet as well as on generalization problems and solutions such as Bias free networks, studied in Mohan et al., Robust and interpretable blind image denoising via bias-free convolutional neural networks, 2020.
Image: FFDNet denoising (Paris)
Variational approaches for image restoration
In this practical session we address different restoration problems such as inpainting, denoising and deblurring. Several variational approaches are used depending on the conditions of the problem: Tikhonov regularization, Total variation L1 and L2, and Wiener filtering.
Image: inpainting and denoising (Simpson)
Plug and Play optimization
Plug-and-play (PnP) is a non-convex framework that integrates modern denoising priors into proximal algorithms. This practical session explains and shows inpainting end deblurring experiments using the following PnP algorithms: PnP ADMM, PnP FBS, PnP BBS, PnP BBS2 and PnP BBS3.
Image: deblurring using BBS3 (Camila & Paula)
Total Variation sampling with the Metropolis algorithm
This practical session explains how to draw samples from different distributions. It uses the Metropolis algorithm, inspired by what is used in Louchet, C., & Moisan, L. (2008, August). Total variation denoising using posterior expectation, EUSIPCO 2008.
Image: total variation sampling with mask operator (Martina)
Total Variation - Iterated Conditional Expectation (TV-ICE)
This session focuses on image denoising using the classical TVL2 model using the algorithm Iterated Conditional Expectation (ICE). It is inspired by the article Louchet, C., Moisan, L., Total variation denoising using iterated conditional expectation. In 2014 22nd European Signal Processing Conference (EUSIPCO) (pp. 1592-1596). 2014.
Image: TV-ICE (Martina)
Unadjusted Langevin Algorithm for inverse problems in imaging
This practical session uses ULA (Unadjusted Langevin Algorithm) to sample from posterior distributions in simple inverse problems. In particular, this practical session shows how to use the algorithm to remove blur and noise from an image.
Image: Total variation ULA sampling for deblurring (Simpson)