EUVIMEDEUVIMED entdecken
Crossref

Accurate Camouflaged Object Segmentation via Dual-Structural Cues Collaborative Learning

Peng Ren, Cheng Jiang, Tian Bai

IEEE Transactions on Multimedia · 2026 · 28 · 11981-11991

DOI: 10.1109/tmm.2026.3714331 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Multimedia, 2026. Verfasst von Peng Ren und 2 weiteren Autor:innen. Die Arbeit nennt 53 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Science and Technology Project of Changbaishan Laboratory.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recently, some studies have made progress on the camouflaged object segmentation (COS) task via introducing structural cues. However, most existing methods rely on a single type of structural information ( e.g. edge, depth), struggling to fully learn the structural integrity and consistency of objects, which ultimately leads to segmentation maps with edge fractures and internal voids. To this end, we propose a Dual-Structural Cues Collaborative Learning Framework, named DC $^{2}$ LNet, which improves the consistent perception of internal regions by jointly modeling depth spatial and edge contour dual-structure cues. The DC $^{2}$ LNet primarily consists of two core components: a Hierarchical Structure Collaboration Unit (HSCU) and a Recurrent Decoder (RD). The HSCU first integrates adjacent-level features via an Adjacent Feature Aggregation Module (AFAM) to enhance local detail and semantic consistency, and then explicitly fuses the local constraints provided by edge cues and the structural priors provided by depth cues via a Structure Enhancement Module (SEM) to improve the structural integrity of camouflaged objects within the features. Further, we design an RD that refines coarse segmentation predictions and corrects erroneous regions by multiple iterative decoding steps, ultimately segmenting the camouflaged object with complete structure. Experiments on four benchmarks demonstrate that the DC $^{2}$ LNet outperforms 28 state-of-the-art algorithms.

Förderung

  • National Natural Science Foundation of China
  • Science and Technology Project of Changbaishan Laboratory

Zitieren

APA

Ren, P., Jiang, C., & Bai, T. (2026). Accurate Camouflaged Object Segmentation via Dual-Structural Cues Collaborative Learning. IEEE Transactions on Multimedia, 28, 11981-11991. https://doi.org/10.1109/tmm.2026.3714331

Vancouver

Ren P, Jiang C, Bai T. Accurate Camouflaged Object Segmentation via Dual-Structural Cues Collaborative Learning. IEEE Transactions on Multimedia. 2026;28:11981-11991. doi:10.1109/tmm.2026.3714331

BibTeX
@article{Ren2026,
  author = {Ren, Peng and Jiang, Cheng and Bai, Tian},
  title = {Accurate Camouflaged Object Segmentation via Dual-Structural Cues Collaborative Learning},
  journal = {IEEE Transactions on Multimedia},
  year = {2026},
  volume = {28},
  pages = {11981--11991},
  doi = {10.1109/tmm.2026.3714331}
}
Kostenlos starten →