人脸的性别分类  

Face Gender Classification

作　　者：武勃[1] 艾海舟[1] 肖习攀[1] 徐光祐[1] 

机构地区：[1]清华大学计算机科学与技术系智能技术与系统国家重点实验室,北京100084

出　　处：《计算机研究与发展》2003年第11期1546-1553,共8页Journal of Computer Research and Development

摘　　要：人脸的性别分类是指根据人脸的图像判别其性别的模式识别问题.系统地研究了不同的特征提取方法和分类方法在性别分类问题上的性能,其中包括主分量分析(PCA)、Fisher线性鉴别分析(FLD)、最佳特征提取、Adaboost算法、支持向量机(SVM).给出了在9姿态人脸库、FERET人脸库和一个网络图片人脸库上的对比实验结果.实验表明人脸中的性别信息集中存在于某个子空间中,因此,在分类前对样本进行适当的压缩降维不但不会明显降低分类器的性能,而且可以大大减少分类的时间开销.最后介绍了将性别分类器与自动人脸检测和特征提取平台集成起来的基于人脸图像的性别判别系统.Gender classification is a pattern recognition problem of deciding the gender according to the face image. The performance of several feature extraction and classification methods on the gender classification problem is investigated systematically. These methods include principal component analysis (PCA), Fisher linear discriminant analysis (FLD), statistically independent linear discriminant analysis, Adaboost algorithm, and support vector machine (SVM) . Comparative experimental results on the 9-pose face database, the FERET face database, and a Web picture face database are given. The results show that the gender information of human face clusters within certain sub-space, and therefore proper compression before gender classification can greatly reduce the time consumed, without sacrificing the good performance. Finally, the face-image-based gender classifier system is introduced, that is integrated with an automatic face detection and feature extraction platform.

关 键 词：性别分类 主分量分析 FISHER线性鉴别分析 ADABOOST 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]