Multi-scale Cross Aggregation Network for Camouflage Object Detection Based on Boundary Guidance

Camouflaged object detection (COD) is challenging due to objects blending seamlessly with their surroundings. This task is crucial in medical imaging, surface defect detection, and surveillance. To address these challenges, we propose MABNet, a framework that aggregates multi-scale cross features and uses edge-guided localization. MABNet integrates multi-level contextual features and mines information from boundary pixels. Key components include the Multi-scale Cross Adaptive Aggregation Block (MCAB), the Edge-guidance Feature Block (EGB) for boundary exploration, and the Multi-scale Residual Block (MRB) for focusing on object structures and details. Extensive experiments on five benchmark datasets show that MABNet outperforms seven state-of-the-art methods, validating its effectiveness in accurately detecting camouflaged objects in various challenging scenarios.