光学
计算机视觉
目标检测
计算机科学
人工智能
边界(拓扑)
图像处理
对象(语法)
反射(计算机编程)
边缘检测
物理
噪音(视频)
点扩散函数
信号处理
背景噪声
探测理论
探测器
伪装
稳健性(进化)
过程(计算)
图像质量
机器视觉
空间频率
作者
Liangchao Chen,Guanying Huo,Weifeng Kong,Ziheng Cao,Jiaying Chen
出处
期刊:Applied optics-OT
[Optica Publishing Group]
日期:2026-08-18
卷期号:65 (26): 9102-9102
摘要
Camouflaged object detection (COD) remains challenging due to the weak appearance contrast between targets and their surroundings, which limits the discriminative capability of RGB images. To address this issue, we propose a polarization-driven boundary prompt segment anything model (PBP-SAM), a multimodal SAM-based framework for COD. A boundary prior refinement module (BPRM) is designed to enhance discriminative polarization cues and refine boundary priors for suppressing background interference, while further exploiting the complementary characteristics between polarization and RGB features to generate contour prompts for SAM. In addition, a lightweight tuning scheme based on an adapter and low-rank adaptation (LoRA) is introduced to facilitate COD feature learning in SAM, enabling SAM to better handle the challenging characteristics of COD with limited additional computational cost. Extensive experiments on benchmark datasets demonstrate that PBP-SAM achieves superior performance compared with existing state-of-the-art methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI