基于形状无关纹理和Boosting学习的人口统计学分类

Demographical Classification by Shape Free Texture and Boosting Learning

基于形状无关纹理和boosting学习,该文提出了对性别和年龄分类的方法,其中年龄被划分为儿童、青年、中年和老年4类。检测到人脸后,利用人脸配准的结果规范化人脸图像获得形状无关纹理。在此基础上提取Haar型特征、LBP直方图和Gabor Jet3种特征,通过boosting学习分别训练分类器。实验表明,LBP直方图特征能够鲁棒地区分儿童和老人,Haar型特征用作区分青年和中年人则更为有效,而Gabor Jet特征更适于性别分类。

In this paper, a gender and age classification method, in which age is classified into four classes： child, youth, midlife and agedness, based on shape free texture and boosting learning is introduced. After a face is detected, face alignment extracts 88 facial landmarks by which the face image is normalized to a shape free texture. Further more, three kinds of local feature, Haar like feature, LBP histogram and Gabor jet are extracted from the shape free texture; and boosting learning method is used for training classifiers. The experimental results show that LBP histogram can be used for robust recognition of children and old people, Haar like feature is more efficient for discriminating young and middle aged people, and Gabor Jet fits for gender classification best.