多分类器融合的人脸识别与身份认证

人脸识别是生物特征识别技术中一个活跃的研究领域,取得了很多实践成果,但是单一分类器一般不能取得满意的识别率与身份认证效果。本文采用贝叶斯决策理论分析了常见的积、和、中值以及投票多分类器融合方法,并根据实际的选举情形,对投票法进行了2种改进。然后对协同人脸识别、特征脸法以及复合方法等人脸识别分类器进行决策层的融合,对ORL库中人脸识别仿真实验表明文中的多分类器融合的人脸识别方法具有较好的分类性能,对污损、低分辨率人脸图像具有可靠的识别率、鲁棒性强,而且应用于人脸身份认证中取得了较好的认证效果。

Fusion of Multiple Classifiers for Face Recognition and Person Authentication

Face recognition is an active subject in the area of biometrics. Lots of achievements have been obtained using face recognition techniques, but generally we cant get high recognition rate and satisfied person authentication effect using single classifier. This paper discusses the familiar fusion methods like product, sum, median and vote rules according to Bayesian theory. And two modified vote rules are presented comparing with election in the real world. Then, these methods are used for decision-fusion of multiple face recognition classifiers: synergetic face recognition classifier, eigenfaces classifier and composite-approach classifier. Experiments with face samples in ORL databases show that the proposed face recognition method has good classifying performance, reliable recognition rate, and strong robustness to defiled and low-resolution face images. More important is that the proposed method also has satisfied experimental results in face person authentication.