文献类型：期刊文献

中文题名：基于类内均值的双投影2DPCA人脸识别方法

英文题名：A Method of Face Recognition Based on Double-projective 2DPCA and Within-Class Average Value

作者：贝宗钧[1];朱煜[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2009

卷号：26

期号：10

起止页码：248

中文期刊名：计算机仿真

外文期刊名：Computer Simulation

收录：CSTPCD;;北大核心:【北大核心2008】；CSCD:【CSCD_E2011_2012】；

语种：中文

中文关键词：人脸识别;特征提取;双投影;类内均值

外文关键词：Face recognition ; Feature extraction ; Double - projective ; Within - class average value

摘要：在人脸识别问题中,如何提取具有鲁棒性的人脸特征和降低特征维数是两个关键。根据二维主成分分析方法直接利用二维图像来构建方差矩阵的优点,引入了类内均值的思想,首先计算每类训练样本的类内平均脸,并用它对各类样本进行规范化处理,有效扩大了类间样本的差别,缩小了类内样本的差别,进行协方差矩阵的计算并提取最优投影特征向量进行人脸图像的特征提取,通过在水平和垂直两个方向上顺序执行两次提取和投影的操作,极大压缩了特征的维数,克服了传统二维主成分算法的不足。在人脸库ORL和Yale的试验对比结果表明,方法对光照与表情变化有较好的鲁棒性,能实现较高的识别率,因此在实际中具有一定的应用意义。
It is a key problem to obtain robust and appropriately low - dimensioned face features in the face recognition task. In this paper, a method based on two directional and two dimensional principal component analysis and combined with the within - class average value is presented for face recognition. First, the training samples are nor- malized by using the corresponding within - class average value, in which the classification distance of between - class samples is enlarged, while that of the within - class is reduced. Then the image covariance matrix is calculated to obtain a family of the optimal feature vectors. All the processes are performed in both horizontal and vertical direc- tions sequentially to compress the coefficients of the feature vectors . The experimental results on ORL and Yale data- bases show that the method proposed in this paper has better robustness to the influences of greatly illuminative and expressive changes . At the same time, it also can achieve a higher recognition rate than the traditional 2DPCA and is of some practical significance.