人脸图像年龄估计  

Age estimation from facial images

作　　者：王俊艳[1] 苏光大[1] 林行刚[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《清华大学学报（自然科学版）》2007年第4期526-529,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家"十五"科技攻关项目(2001BA801B07)

摘　　要：为了解决人脸识别中识别率随年龄变化急剧下降的问题,提出一种自动的年龄估计方法。通过年龄函数建立人脸特征与年龄之间的对应关系,人脸特征由形状特征和纹理特征组成。首先,通过改进的活动形状模型自动定位人脸形状特征点,再通过形状拉伸获得单纯的纹理图像;然后用主分量分析分别得到形状特征向量和纹理特征向量,合成作为人脸特征;最后,按照最优化的准则得到年龄函数,结合人脸变老方式的分类,自动估计年龄。实验结果表明,这种方法对年龄估计十分有效,平均200人的多年龄人脸数据库上的估计误差为1.1 a。Automatic age estimation from facial images has applications in many areas. This paper describes an automatic age estimation method using front-view facial images. The age is estimated by an age function from the facial features. The algorithm first positions the facial shape points using an improved active shape model. Then the isolated facial texture image is extracted by spanning. The shape and texture feature vectors found using principal component analysis are combined into a single feature vector which is the input to the age function determined from a training set. Tests show that the method accurately estimates facial ages with a mean estimation error of 1.1 years.

关 键 词：年龄估计 特征提取 主元分析 人脸识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]