摘要
基于内容的敏感图像过滤是近年来兴起的一个新的研究领域。本文重点研究了基于内容敏感图像过滤中的若干关键技术,分析比较了三种常用的肤色检测算法,利用统计阈值法评价了各个特征分类图像的能力强弱,并依此实现了一个决策树敏感图像过滤器。为有效降低肖像类正常图像的误判率,又在过滤器中使用了人脸检测机制。在1 207幅实验图像库上(624幅敏感图像和583幅正常图像),本系统的正确检测率为84.01%,其中敏感图像为89.74%,正常图像为75.13%。
Research on content-based erotic image filtering is a new technology emerged in recent years. This paper studies some key technologies of content-based erotic image filtering, analyses and compares three algorithms of detecting skin-color pixels commonly used, estimates the effectiveness of the feature-classified images using statistical threshold method. In the end, a Decision Tree Classifier is constructed for filtering erotic image. In order to reduce the error rate of classification for portrait images, human face detector is used in the filter. The filter was tested on our image set with 1 207 images （624 erotic images and 583 benign images）, the correct classification rate is 84.01% , within which the correct rate for erotic image recognition is 89.74% , and the one for benign images is 75.13%.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2007年第11期2059-2066,共8页
Chinese Journal of Scientific Instrument
基金
2004年度珠海市科技项目(PC20041101)资助
关键词
敏感图像
肤色检测
人脸检测
决策树分类器
erotic image
detection of skin-color
detection of human face
decision tree classifier.
