摘要
本文提出了一种适用于对互联网敏感图像进行监控和检测的解决方案。该方案以实例图像的匹配为基本识别策略,以测试图像的K近邻作为分类依据,创新性地提出了按多模式特征,分组组织训练实例的方法,并在匹配中融入了局部视觉元素的描述。系统在各种测试图像(尤其是人物类图像)中取得了出色的性能。实验结果证明,本系统有效地兼顾了对敏感图像多样性的适应能力和识别效率,相比传统类型策略,能使检测性能(尤其是误检率)得到明显改善。
A novel filtering system is presented to detect pornographic images. The system recognizes pornographic images based on the image classification using K nearest neighbors. Multi-mode features are adopted to group labeled image examples, and local visual components are integrated into the image classification. Our method performs well on a wide range of test data, in particular on human-related images. Experimental results demonstrate that the system outperforms traditional skin-region and human-body-model based methods.
出处
《电信科学》
北大核心
2008年第12期11-15,共5页
Telecommunications Science
基金
国家"863"计划基金资助项目(No.2003AA142140)
国家自然科学基金资助项目(No.60702035)
关键词
敏感图像监控
分组K近邻
视觉元素
pornographic image detection, grouped K nearest neighbor, visual component
