EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2585. Fractional graph Laplacian for image reconstruction

Invited abstract in session MA-34: Optimization and learning for data science and imaging (Part I), stream Advances in large scale nonlinear optimization.

Monday, 8:30-10:00
Room: 43 (building: 303A)

Authors (first author is the speaker)

1. Marco Donatelli
Università degli studi dell'Insubria
2. Stefano Aleotti
Scienza ed alta tecnologia, Università degli Studi dell'Insubria
3. Alessandro Buccini
Università degli Studi di Cagliari

Abstract

A popular approach for regularization involves replacing the original problem with an optimization problem that minimizes the sum of two terms: an l^2 term and an l^q term with q smaller than or equal to one. The first penalizes the distance between the measured and reconstructed data, while the second imposes sparsity on certain coefficients of the computed solution.

In [1], we propose to use the fractional Laplacian of a properly constructed graph in the l^q term to compute extremely accurate reconstructions of the desired images. Furthermore, we propose automatic approaches to determine the involved parameters so that the proposed method is completely plug-and-play. We show that the algorithm, under some reasonable assumptions, is a regularization method. Some selected numerical examples show the performances of our proposal.

[1] S. Aleotti, A. Buccini, M. Donatelli, Fractional graph Laplacian for image reconstruction, Applied Numerical Mathematics, 2023.

Keywords

Status: accepted


Back to the list of papers