ARPNet: Adaptive Reliable Points Selection Network for Camouflaged Object Detection

Camouflaged object detection (COD) holds significant application value in fields such as military reconnaissance, wildlife monitoring, and autonomous driving. Existing methods face challenges in addressing core issues such as blurred edges and feature confusion caused by the high similarity between targets and backgrounds. To overcome these limitations, we propose a novel Adaptive Reliable Points Selection Network (ARPNet) that locates camouflaged objects by selecting reliable and high-confidence camouflaged points. First, a Feature Correlation Fusion module (FCF) deeply integrates camouflaged features extracted by the backbone network with frequency-texture features to construct a comprehensive multi-level representation with rich detail cues. Next, we design an Adaptive Reliable Points Selection (ARPS) module that incorporates learnable camouflage prototypes and a gated selection strategy to achieve high-purity reliable feature extraction while effectively suppressing background interference. Finally, an end-to-end camouflage decoder progressively refines the integrated features to generate the final camouflage object mask. Experimental results on benchmark datasets (CHAMELEON, COD10 K, CAMO, and NC4K) demonstrate that the proposed method significantly outperforms existing state-of-the-art approaches in both quantitative and qualitative evaluations.