性别分类中头发特征提取方法的研究  

Research on Hair Feature Extraction Method in Gender Classification

作　　者：谢金融[1] 卜佳俊[1] 

机构地区：[1]浙江大学计算机科学与技术学院,杭州310027

出　　处：《计算机工程》2010年第7期179-181,184,共4页Computer Engineering

摘　　要：针对人脸性别分类问题,提出基于头发特征的分类方法。对人脸重要外部特征之一的头发特征属性的提取与表示以及参数的选取进行分析与研究。在1680张AR人脸图片上,利用头发特征模型对性别进行分类,将实验结果与基于人脸内部特征的分类结果进行比较,结果表明,采用头发特征的性别分类,其准确度获得平均7.5%的提升,最高准确率达96%。This paper provides the method of human hair feature based gender classification. The details of algorithms applied to the hair attributes extraction and feature representation are mainly focused on, while the selection of parameters for the model is briefly discussed. Results on AR dataset of 1 680 subjects are reported. The comparison between identification using hair attributes and the eigenface-based recognition indicates that the hair attributes comparatively improve the average human gender classification rate by 7.5%. The best performance of 96% is achieved with the hair feature based extraction.

关 键 词：人脸外部特征 头发属性 性别分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]