基于相关向量机的含噪声人脸图像识别  

Noisy face images recognition based on relevance vector machine

作　　者：柳长源[1,2] 毕晓君[1] 韦琦[2] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,哈尔滨150001 [2]哈尔滨理工大学电气与电子工程学院,哈尔滨150080

出　　处：《吉林大学学报（工学版）》2013年第4期1121-1126,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(61175126);中央高校基本科研业务费专项项目(HEUCFZ1209);高等学校博士学科点专项科研基金项目(20112304110009)

摘　　要：提出一种采用相关向量机的人脸识别方法,利用机器学习对小波分解和PCA变换后的人脸数据库样本进行训练,得到的相关向量构成'超平面'作为差异样本的分类面,并利用'一对一'方法实现多类别模式识别。对加噪声的识别对象进行了大量的仿真实验,结果表明:与传统方法相比,本文方法对图像噪声不敏感,具有更高的识别率和很强的鲁棒性。A new method of face recognition based on relevance vector machine was developed. After the wavelet decomposition and Principal Component Analysis （PCA） transformation, the relevance vectors from sample training constitute a ＂hyperplane＂ as the differences in the classification of the samples by machine learning algorithm. The ＂one against one＂ method is used to achieve multi-class pattern recognition. Compared with former methods, a large number of simulation results of recognition of noisy objects show that the new method is not sensitive to image noise, with high recognition accuracy and strong robustness.

关 键 词：信息处理技术 人脸识别 相关向量机 图像噪声 

分 类 号：TN911[电子电信—通信与信息系统]