RELD: Regularization by Latent Denoising
IEEE Signal Processing Letters · 2026 · 33 · 2605-2609
DOI: 10.1109/lsp.2026.3705374 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Pasquale Cascarano und 5 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: INdAM-GNCS, Cariplo Project Dasacaf.
Abstract (Originalsprache)
Diffusion Models have recently emerged as state-of-the-art generative models. Inspired by the Regularization by Denoising principle, we propose Regularization by Latent Denoising (RELD), a variational framework that integrates a denoising-trained Latent Diffusion Model via Half-Quadratic Splitting. This approach, under appropriate conditions that can be easily met in various imaging applications, allows for reduced computational cost while achieving high-quality results. The proposed strategy is tested on a dataset of natural images, for image denoising, deblurring, and super-resolution tasks, demonstrating competitive performance against other state-of-the-art techniques, with particularly strong results in terms of perceptual quality metrics.
Förderung
- INdAM-GNCS
- Cariplo Project Dasacaf
Zitieren
APA
Cascarano, P., Stacchio, L., Sebastiani, A., Benfenati, A., Kamilov, U. S., & Marfia, G. (2026). RELD: Regularization by Latent Denoising. IEEE Signal Processing Letters, 33, 2605-2609. https://doi.org/10.1109/lsp.2026.3705374
Vancouver
Cascarano P, Stacchio L, Sebastiani A, Benfenati A, Kamilov US, Marfia G. RELD: Regularization by Latent Denoising. IEEE Signal Processing Letters. 2026;33:2605-2609. doi:10.1109/lsp.2026.3705374
BibTeX
@article{Cascarano2026,
author = {Cascarano, Pasquale and Stacchio, Lorenzo and Sebastiani, Andrea and Benfenati, Alessandro and Kamilov, Ulugbek S. and Marfia, Gustavo},
title = {RELD: Regularization by Latent Denoising},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {2605--2609},
doi = {10.1109/lsp.2026.3705374}
}
EUVIMED