Linear inverse problems with nonnegativity constraints: singularity of optimisers

DOI10.3934/IPI.2023026MaRDI QIDQ6066518zbMATH OpenOpenAlexFDO

Authors Camille Pouchol, Olivier Verdier

Publication date 13 December 2023

Published in Inverse Problems and Imaging (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2006.15845

zbMATH Keywords

inverse problemsnonnegativitypositron emission tomographyRadon measuressparsity

Mathematics Subject Classification ID

Optimality conditions (49K99) Numerical solutions of ill-posed problems in abstract spaces; regularization (65J20)

Abstract: We look at continuum solutions in optimisation problems associated to linear inverse problems $y=Ax$ with non-negativity constraint $xgeq0$. We focus on the case where the noise model leads to maximum likelihood estimation through general divergences, which covers a wide range of common noise statistics such as Gaussian and Poisson. Considering $x$ as a Radon measure over the domain on which the reconstruction is taking place, we show a general singularity result. In the high noise regime corresponding to $yotinAxmidxgeq0$ and under a key assumption on the divergence as well as on the operator $A$, any optimiser has a singular part with respect to the Lebesgue measure. We hence provide an explanation as to why any possible algorithm successfully solving the optimisation problem will lead to undesirably spiky-looking images when the image resolution gets finer, a phenomenon well documented in the literature. We illustrate these results with several numerical examples inspired by medical imaging.

Review text

In the paper under review, the authors study linear inverse problems of the type \(Ax = y\), where \(A\) is a linear operator, and \(y\) lies in a suitable linear space. In many applications, for example medical imaging and Positron Emission Tomography, \(x\) is a nonnegative number. In this paper, the authors focus on the case where the noise model leads to maximum likelihood estimation through general divergences, which cover a wide range of common noise statistics such as Gaussian and Poisson. Thinking of the unknown \(x\) as a function (say \(f\)) in some functional space \(X\) of functions over a compact set \(K\in \mathbb R^p\) and where the operator \(A\) is a linear mapping from \(X\) to \(\mathbb R^m\), leads to the optimisation problem \(\min_{f\geq 0}D(y,A\mu)\). This non-negativity constraint has been shown to cause sparsity in various contexts in optimal control and optimization. The authors prove several key results and provide several numerical examples. The paper is well written with a good set of references.

Recommendations

• Multiplicative Algorithms for Maximum Penalized Likelihood Inversion with Non Negative Constraints and Generalized Error Distributions
• Positive solutions to linear inverse problems
• MCMC Algorithms for Computational UQ of Nonnegativity Constrained Linear Inverse Problems
• Inverse problems with nonnegative and sparse solutions: algorithms and application to the phase retrieval problem
• The ML–EM algorithm in continuum: sparse measure solutions

Cites work

• A first-order primal-dual algorithm for convex problems with applications to imaging
• A Statistical Model for Positron Emission Tomography
• Algorithms for nonnegative matrix factorization with the -divergence
• Applied functional analysis. Functional analysis, Sobolev spaces and elliptic differential equations
• Extension of Fenchel's duality theorem for convex functions
• Families of alpha-, beta- and gamma-divergences: flexible and robust measures of similarities
• Fast Positive Deconvolution of Hyperspectral Images
• scientific article; zbMATH DE number 1817650 (Why is no real title available?)
• Minimal controllability time for the heat equation under unilateral state or control constraints
• On representer theorems and convex regularization
• On the uniqueness of solutions for the basis pursuit in the continuum
• Optimal control of elliptic equations with positive measures
• Positron emission tomography, Borel measures and weak convergence
• Solution of the general moment problem via a one-parameter imbedding
• Sparsity of solutions for variational inverse problems with finite-dimensional data
• Statistical model and ML-EM algorithm for emission tomography with known movement
• Support recovery for sparse super-resolution of positive measures
• The expectation-maximization algorithm for ill-posed integral equations: a convergence analysis
• The ML–EM algorithm in continuum: sparse measure solutions

Cited in

(4)

• Optimal map-making with singularities
• Off-the-grid regularisation for Poisson inverse problems
• Hermite solution for a new fractional inverse differential problem
• A fractional-time PDE-constrained parameter identification for inverse image noise removal problem

This page was built for publication: Linear inverse problems with nonnegativity constraints: singularity of optimisers

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6066518)