Paperlane

no. 14 · Friday 13 November 2026

A Weighted Second-Order Nonlocal Total Variation Regularizer for Image Restoration

Francesc Alcover, Joan Duran and Catalina Sbert

math.OC

Abstract

Despite the remarkable progress of deep learning in image processing, variational methods remain a fundamental and actively developing framework, offering interpretability, mathematical rigor, and well-founded guarantees. In this work, we introduce a weighted second-order nonlocal total variation regularizer for image restoration, which is based on measuring gradient differences across the image domain and allows for weights that are neither smooth nor symmetric and may vanish outside a search window. We first provide a rigorous definition of the regularizer and prove that it enjoys several desirable properties under realistic assumptions on the underlying weight function. We then consider a nonlocal variant of the variational image restoration model of Chambolle and Lions, and prove the existence and uniqueness of solutions to the associated minimization problem. On the computational side, we develop a discrete formulation of the model and derive the corresponding Chambolle-Pock primal-dual algorithm. Finally, we conduct an extensive experimental evaluation, comparing the proposed regularizer with a range of local and nonlocal, first- and second-order regularizers across several image restoration tasks. The results demonstrate its strong noise-suppression and texture-preservation capabilities, as well as its robustness to nonlocal weights computed under non-ideal conditions.

fields
Optimization and Control
pdf ↗ page arXiv:2611.16618 ↗