PnP-ReG: Learned Regularizing Gradient for Plug-and-Play Gradient Descent

DOI10.1137/22M1490843zbMATH Open1517.62071arXiv2204.13940OpenAlexW4321747284MaRDI QIDQ6173535FDOQ6173535

Authors: Mikaël le Pendu, Christine Guillemot 

Publication date: 21 July 2023

Published in: SIAM Journal on Imaging Sciences (Search for Journal in Brave)

Abstract: The Plug-and-Play (PnP) framework makes it possible to integrate advanced image denoising priors into optimization algorithms, to efficiently solve a variety of image restoration tasks generally formulated as Maximum A Posteriori (MAP) estimation problems. The Plug-and-Play alternating direction method of multipliers (ADMM) and the Regularization by Denoising (RED) algorithms are two examples of such methods that made a breakthrough in image restoration. However, while the former method only applies to proximal algorithms, it has recently been shown that there exists no regularization that explains the RED algorithm when the denoisers lack Jacobian symmetry, which happen to be the case of most practical denoisers. To the best of our knowledge, there exists no method for training a network that directly represents the gradient of a regularizer, which can be directly used in Plug-and-Play gradient-based algorithms. We show that it is possible to train a network directly modeling the gradient of a MAP regularizer while jointly training the corresponding MAP denoiser. We use this network in gradient-based optimization methods and obtain better results comparing to other generic Plug-and-Play approaches. We also show that the regularizer can be used as a pre-trained network for unrolled gradient descent. Lastly, we show that the resulting denoiser allows for a better convergence of the Plug-and-Play ADMM.

Full work available at URL: https://arxiv.org/abs/2204.13940

Recommendations

• The Little Engine that Could: Regularization by Denoising (RED)
• Regularization by denoising via fixed-point projection (RED-PRO)
• Learning maximally monotone operators for image recovery
• Plug-and-play gradient-based denoisers applied to CT image enhancement
• Plug-and-Play Unplugged: Optimization-Free Reconstruction Using Consensus Equilibrium

zbMATH Keywords

regularizationinverse problemsplug-and-play priorgradient descent

Mathematics Subject Classification ID

Image analysis in multivariate analysis (62H35) Image processing (compression, reconstruction, etc.) in information and communication theory (94A08) Computing methodologies for image processing (68U10) Artificial intelligence (68T99)

Cites Work

• Nonlinear total variation based noise removal algorithms
• Distributed optimization and statistical learning via the alternating direction method of multipliers
• The Little Engine that Could: Regularization by Denoising (RED)
• A Review of Image Denoising Algorithms, with a New One
• An introduction to total variation for image analysis
• An EM algorithm for wavelet-based image restoration
• On single image scale-up using sparse-representations
• Hessian Schatten-Norm Regularization for Linear Inverse Problems
• Weighted Schatten <inline-formula> <tex-math notation="LaTeX">$p$ </tex-math> </inline-formula>-Norm Minimization for Image Denoising and Background Subtraction
• Weighted nuclear norm minimization and its applications to low level vision
• Solving ill-posed inverse problems using iterative deep neural networks
• Should Penalized Least Squares Regression be Interpreted as Maximum A Posteriori Estimation?
• Waterloo Exploration Database: New Challenges for Image Quality Assessment Models
• Bayesian Imaging Using Plug & Play Priors: When Langevin Meets Tweedie
• Preconditioned Plug-and-Play ADMM with Locally Adjustable Denoiser for Image Restoration

Cited In (6)

• Plug-and-play gradient-based denoisers applied to CT image enhancement
• Regularization by denoising via fixed-point projection (RED-PRO)
• Learning maximally monotone operators for image recovery
• Averaged deep denoisers for image regularization
• Preconditioned Plug-and-Play ADMM with Locally Adjustable Denoiser for Image Restoration
• Provably Convergent Plug-and-Play Quasi-Newton Methods

This page was built for publication: PnP-ReG: Learned Regularizing Gradient for Plug-and-Play Gradient Descent

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6173535)