摘要
随着人脸识别技术的不断发展,单样本人脸识别已成为当今的一个热点。针对单样本人脸识别问题,提出一种基于通用框架学习的人脸识别方法。以大量的通用样本与各个单样本按一定比例叠加的方式,增加每个类的训练样本总数,有效地运用FLDA方法进行特征抽取,将所有样本投影到特征子空间,再利用最近邻方法完成人脸识别,一定程度上减轻了人脸的表情、姿态、光照等因素对识别效果的影响,提高了识别率。该方法的有效性分别在ORL及Yale两大人脸库上得到了验证。
With the constant development of face recognition technology,single sample face recognition has become today's focus. In light of this issue,in the paper we present a face recognition method which is based on general frame learning. The method increases the total number of training samples of every class in the way of superimposing each single sample with a great deal of general samples in certain proportion,effectively utilises FLDA method to extract the features,and maps all the samples onto feature subspace,then makes use of the nearest neighbouring method to complete the face recognition,this mitigates the impacts of those factors including facial expression,attitude,illumination,etc. on recognition effect and raises recognition rate. The effectiveness of the proposed method has been verified on two major face libraries of ORL and Yale respectively.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第4期175-177,共3页
Computer Applications and Software
关键词
人脸识别
单训练样本
通用框架学习
FISHER线性判别分析
Face recognition Single training sample General frame learning Fisher linear discriminant analysis(FLDA)