GLCONet: Learning Multisource Perception Representation for Camouflaged Object Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Camouflaged Object Detection (COD) heavily relies on local spatial information from convolutional operations.
• Existing methods often overlook long-range dependencies crucial for global structure understanding in COD.
• Accurate image representation for precise camouflaged object detection remains a challenge.

Purpose of the Study:

• To propose a novel network, GLCONet, for improved camouflaged object detection.
• To address the limitation of neglecting long-range dependencies in existing COD methods.
• To enhance feature representation by integrating local details and global context.

Main Methods:

• Developed a Global-Local Collaborative Optimization Network (GLCONet).
• Introduced a Collaborative Optimization Strategy (COS) for multisource perception, modeling local details and global relationships.
• Designed an Adjacent Reverse Decoder (ARD) with cross-layer aggregation and reverse optimization for high-quality representations.

Main Results:

• GLCONet effectively activates significant pixels for camouflaged object detection.
• The proposed method achieves superior performance compared to 20 state-of-the-art methods.
• Experiments conducted on three public COD datasets validate the effectiveness of GLCONet.

Conclusions:

• GLCONet offers a powerful approach for camouflaged object detection by optimizing global-local features.
• The integration of local details and long-range dependencies significantly boosts detection accuracy.
• GLCONet provides a robust and effective solution for challenging COD tasks.