Multistage image denoising model based on cascaded dual subnets
The Computer Journal · 21.08.2026
DOI: 10.1093/comjnl/bxag077 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Computer Journal, online erschienen am 21.08.2026. Verfasst von Xiao Wang und 4 weiteren Autor:innen. Die Arbeit nennt 62 Literaturangaben. Gefördert von: Tianjin Education Commission Scientific Research Program.
Abstract (Originalsprache)
To flexibly extract and balance local spatial details and nonlocal contextual information, we propose a multistage model with cascaded dual subnets (CDS-MSM) for image denoising. CDS-MSM divides the denoising process into three stages, each equipped with a spatial detail subnet and a cross-scale fusion subnet (CSFNet). The spatial detail subnet leverages stacked dense residual channel attention blocks to capture local spatial details, while an edge enhancement branch network (EEBNet) is employed to extract high-frequency edge information. CSFNet fuses nonlocal contextual information via parallel dilated convolutions, attention modules, and fusion layers. For EEBNet training, we adopt an improved Laplacian algorithm and introduce an edge-weighted loss function to better capture contextual information. Experiments demonstrate that CDS-MSM outperforms state-of-the-art models on multiple synthetic and real noise denoising datasets, achieving both qualitative and quantitative improvements.
Förderung
- Tianjin Education Commission Scientific Research Program
Zitieren
APA
Wang, X., Li, N., Zhao, J., Shen, Y., & Li, Y. (2026). Multistage image denoising model based on cascaded dual subnets. The Computer Journal. https://doi.org/10.1093/comjnl/bxag077
Vancouver
Wang X, Li N, Zhao J, Shen Y, Li Y. Multistage image denoising model based on cascaded dual subnets. The Computer Journal. 2026. doi:10.1093/comjnl/bxag077
BibTeX
@article{Wang2026,
author = {Wang, Xiao and Li, Na and Zhao, Jun and Shen, Yuanzhe and Li, Yaoze},
title = {Multistage image denoising model based on cascaded dual subnets},
journal = {The Computer Journal},
year = {2026},
doi = {10.1093/comjnl/bxag077}
}
EUVIMED