EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Research on Co‐Salient Object Detection Methods via Deep Correlation Learning

Feng Wang, Bin Wang, Jia Zhao, Huiling Wang, Yingzhou Chen

IET Image Processing · 2026-01 · 20 (1)

DOI: 10.1049/ipr2.70427 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 06.07.2026. Verfasst von Feng Wang und 4 weiteren Autor:innen. Die Arbeit nennt 53 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Anhui Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Co‐salient object detection (CoSOD) aims to extract common salient objects from semantically related images and has wide applications in video analysis, image retrieval and multi‐view understanding. Existing methods often suffer from inadequate cross‐image consistency modelling because conventional normalization and activation mechanisms lack adaptive representation capability, making them less effective under complex backgrounds and significant appearance variations. To address these issues, this paper proposes an improved deep correlation learning framework for CoSOD. First, a dynamic tanh (DTH) module is introduced in the decoder stage, where the proposed DTH module replaces the traditional LayerNorm with a learnable normalization‐activation mechanism, reducing computational complexity while improving adaptive feature distribution modelling and nonlinear representation capability. Second, a multi‐scale self‐attention enhancement module is designed in the global feature enhancement stage, combining global priors with self‐attention mechanisms to achieve long‐range dependency modelling and multi‐scale feature fusion. Extensive experiments on benchmark datasets, including CoCA, CoSal2015 and CoSOD3k, demonstrate that the proposed method consistently outperforms mainstream state‐of‐the‐art approaches. Specifically, compared with the best existing method, the proposed framework improves the S‐measure by 0.92% and reduces the MAE by 0.86% on the CoCA dataset, while also achieving superior performance in E‐measure and F‐measure across multiple benchmarks. These results verify the effectiveness, robustness and generalization capability of the proposed framework. The code is available at: https://github.com/cyzroot‐jpg/CoSOD‐DCL.git .

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Anhui Province

Zitieren

APA

Wang, F., Wang, B., Zhao, J., Wang, H., & Chen, Y. (2026). Research on Co‐Salient Object Detection Methods via Deep Correlation Learning. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70427

Vancouver

Wang F, Wang B, Zhao J, Wang H, Chen Y. Research on Co‐Salient Object Detection Methods via Deep Correlation Learning. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70427

BibTeX
@article{Wang2026,
  author = {Wang, Feng and Wang, Bin and Zhao, Jia and Wang, Huiling and Chen, Yingzhou},
  title = {Research on Co‐Salient Object Detection Methods via Deep Correlation Learning},
  journal = {IET Image Processing},
  year = {2026},
  volume = {20},
  number = {1},
  doi = {10.1049/ipr2.70427}
}
Kostenlos starten →