小样本条件下基于随机森林和Haar特征的多姿态人脸识别  

MULTI-POSE FACE RECOGNITION IN SMALL SAMPLE ENVIRONMENT BASED ON RANDOM FOREST CLASSIFIER AND HAAR FEATURE

作　　者：周致富[1] 吴怀宇[1] 张志芬[1] 陈洋[2] 

机构地区：[1]武汉科技大学信息科学与工程学院,湖北武汉430081 [2]冶金工业过程系统科学湖北省重点实验室(武汉科技大学),湖北武汉430081

出　　处：《计算机应用与软件》2015年第12期108-113,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61075087;61203331)

摘　　要：针对小样本环境下存在人脸姿态、表情变化等干扰时的人脸识别问题,提出利用基于Haar特征的随机森林分类器完成对注册样本和待识别人脸图像的关键点自适应定位,再以SURF(Speed-Up Robust Features)特征的欧氏距离决策得出初匹配和再匹配关键点,完成人脸识别,解决在小样本环境下识别多姿态人脸图像的问题。实验结果证明,该方法在表情、姿态变化等干扰情况下能有效提高小样本人脸识别的识别率。In view of the face recognition problems in small sample environment that there are the interferences of human face pose,expression change,etc.,in this paper we first propose to use Haar feature-based random forest classifier to complete adaptive localisation of key points of registration samples and the face image to be recognised,then make the decision on eliciting the key points of early matching and re-matching based on Euclidean distance of SURF feature to complete the face recognition,and to solve the problem of recognising multi-pose face image in small sample environment. Experimental results show that the method can effectively improve the recognition rate of faces with small sample under the interference condition of expression and pose face changes.

关 键 词：人脸识别 小样本 随机森林 HAAR 多姿态 SURF算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]