摘要
为克服成人图像基于低层视觉特征识别产生的大量误检情况,在检测中引入了图像的局部形态特征.将图像中突出的局部形态量化为视觉单词,并消除了视觉单词中的多义和同义现象.然后对测试图像建立了视觉单词敏感度分布缩图,从中获知与色情信息高度相关的区域.通过分析这些区域的色情相关性,判断图像是否为成人图像.在与传统类型方法的对比实验中,该策略显著提高了检测性能,特别是减少了很多对人物类图像的误检.实验结果表明,基于分析局部视觉单词分布的策略,能有效提高成人图像识别系统的性能.
Most pornographic image detection technologies proposed in recent years are based on low level visual features and often lead to many false detections.Try here to overcome the problem by introducing image local features into the detection.The appearance descriptors of the interested points are quantized as visual words in a refined visual vocabulary.The system builds miniature images describing the distribution of pornographic visual words.From the miniature,regions involving pornography remarkably are extracted,and are analyzed to judge whether the whole image is pornographic.Compared with traditional type of detection system,the new approach performs better and makes fewer mistakes on the test set including many benign portraits.Experimental results showed that the method based on the distribution of visual words can provide a promising direction to detect pornographic images.
出处
《北京理工大学学报》
EI
CAS
CSCD
北大核心
2008年第5期410-413,共4页
Transactions of Beijing Institute of Technology
基金
国家“八六三”计划项目(2003AA142140)
关键词
敏感图像
图像识别
视觉单词
语义分析
pornographic image
image recognition
visual word
semantic analysis
