Neural-network-based regularization methods for inverse problems in imaging

Andreas Habring,Martin Holler

2023-12-23

Abstract:This review provides an introduction to - and overview of - the current state of the art in neural-network based regularization methods for inverse problems in imaging. It aims to introduce readers with a solid knowledge in applied mathematics and a basic understanding of neural networks to different concepts of applying neural networks for regularizing inverse problems in imaging. Distinguishing features of this review are, among others, an easily accessible introduction to learned generators and learned priors, in particular diffusion models, for inverse problems, and a section focusing explicitly on existing results in function space analysis of neural-network-based approaches in this context.

Optimization and Control,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the method of applying neural networks (especially convolutional neural networks) for regularization in inverse problems in imaging. Specifically, the paper aims to provide an overview of the latest progress in regularization methods of current neural networks in inverse problems in imaging, focusing on explaining the concepts of the most successful methods so that readers can clearly understand these methods, rather than providing an exhaustive literature review. The paper mainly focuses on how to use neural networks to solve inverse problems in imaging, especially when these problems become difficult due to the non - invertibility or severe ill - conditioning of the forward operator. The paper explores this topic by introducing different types of neural network regularization methods, such as unsupervised models, supervised models and untrained models, as well as their applications in function space analysis and uncertainty quantification. These methods aim to overcome the limitations of traditional variational regularization methods and improve the quality of image reconstruction. For example, the paper compares the effects of zero - padding reconstruction, hand - crafted TV priors and CNN - based reconstruction methods in reconstructing magnetic resonance images from undersampled data, demonstrating the potential of neural - network - based methods in improving reconstruction quality. In summary, the core problem of this paper is to explore and explain how to use neural network techniques to improve regularization methods in inverse problems in imaging in order to achieve higher - quality image reconstruction.

Neural-network-based regularization methods for inverse problems in imaging

Regularization of Inverse Problems by Neural Networks

Big in Japan: Regularizing networks for solving inverse problems

Bayesian Inference and Deep Learning for Inverse Problems

A Graph-based Iterative Algorithm for Inverse Problems with Local and Non-local Regularizations

Neumann Networks for Inverse Problems in Imaging

Neural networks-based regularization for large-scale medical image reconstruction

A Neural-Network-Based Convex Regularizer for Inverse Problems

To be or not to be stable, that is the question: understanding neural networks for inverse problems

Untrained Neural Network Priors for Inverse Imaging Problems: A Survey

Learned convex regularizers for inverse problems

Data-proximal null-space networks for inverse problems

Invertible ResNets for Inverse Imaging Problems: Competitive Performance with Provable Regularization Properties

DRIP: Deep Regularizers for Inverse Problems

Guided Deep Generative Model-based Spatial Regularization for Multiband Imaging Inverse Problems

Robustness and Exploration of Variational and Machine Learning Approaches to Inverse Problems: An Overview

Weakly Convex Regularisers for Inverse Problems: Convergence of Critical Points and Primal-Dual Optimisation

Theoretical Perspectives on Deep Learning Methods in Inverse Problems

Invertible residual networks in the context of regularization theory for linear inverse problems

Convergent Regularization in Inverse Problems and Linear Plug-and-Play Denoisers

An accelerated inexact Newton regularization scheme with a learned feature-selection rule for non-linear inverse problems