Scholarlink

<p>Camouflaged object detection (COD) remains inherently challenging due to the strong appearance and structural similarity between foreground objects and their surrounding environments. In collaborative scenarios, where multiple images contain objects sharing similar camouflage patterns, modeling inter-image consensus becomes essential for reliable localization. However, existing collaborative camouflaged object detection (CoCOD) methods often struggle with scale inconsistency across views, structural attenuation during hierarchical feature fusion, and boundary ambiguity caused by background interference. To address these challenges, we propose WCAF-Net, a wavelet-enhanced consensus adaptive fusion network, which integrates scale-aware feature interaction and frequency-disentangled structural modeling in a unified architecture. Specifically, a multi-scale adaptive feature fusion (MAFF) module reconciles global consensus cues with fine-grained local variations through dynamic cross-scale weighting. A wavelet-enhanced high-frequency reconstruction (WHR) module performs wavelet-based frequency disentanglement to selectively reinforce structurally informative high-frequency components, alleviating detail dilution caused by hierarchical feature fusion. Furthermore, an edge-aware feature refinement (EFR) module leverages local structural contrast to improve contour completeness while avoiding over-enhancement artifacts. Extensive experiments on the CoCOD8K dataset demonstrate that the proposed method consistently outperforms representative COD, co-salient object detection (CoSOD), and CoCOD approaches across multiple evaluation metrics. The code will be available at <a href="https://github.com/lishiyuan11/WCAF-Net">https://github.com/lishiyuan11/WCAF-Net</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

WCAF-net: wavelet-enhanced consensus adaptive fusion network for collaborative camouflaged object detection

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Shiyuan Li,
  • Hongbo Bi,
  • Yulin Zeng,
  • Cong Zhang,
  • Yue Li

摘要

Camouflaged object detection (COD) remains inherently challenging due to the strong appearance and structural similarity between foreground objects and their surrounding environments. In collaborative scenarios, where multiple images contain objects sharing similar camouflage patterns, modeling inter-image consensus becomes essential for reliable localization. However, existing collaborative camouflaged object detection (CoCOD) methods often struggle with scale inconsistency across views, structural attenuation during hierarchical feature fusion, and boundary ambiguity caused by background interference. To address these challenges, we propose WCAF-Net, a wavelet-enhanced consensus adaptive fusion network, which integrates scale-aware feature interaction and frequency-disentangled structural modeling in a unified architecture. Specifically, a multi-scale adaptive feature fusion (MAFF) module reconciles global consensus cues with fine-grained local variations through dynamic cross-scale weighting. A wavelet-enhanced high-frequency reconstruction (WHR) module performs wavelet-based frequency disentanglement to selectively reinforce structurally informative high-frequency components, alleviating detail dilution caused by hierarchical feature fusion. Furthermore, an edge-aware feature refinement (EFR) module leverages local structural contrast to improve contour completeness while avoiding over-enhancement artifacts. Extensive experiments on the CoCOD8K dataset demonstrate that the proposed method consistently outperforms representative COD, co-salient object detection (CoSOD), and CoCOD approaches across multiple evaluation metrics. The code will be available at https://github.com/lishiyuan11/WCAF-Net.