改进的双向2DPCA的人脸识别方法  

Face Recognition Method Based on Improved Two-Directional 2DPCA

作　　者：叶永凯[1] 封玲娟[1] 刘敏丽[1] 王玉德[1] 

机构地区：[1]曲阜师范大学

出　　处：《电子技术（上海）》2012年第11期4-7,共4页Electronic Technology

摘　　要：为了提高人脸识别的准确率,且考虑到训练样本的平均值不一定是训练样本分布中心,提出了改进的双向2DPCA人脸识别方法。首先,应用样本中间值代替样本的平均值来重建图像的总体散布矩阵,求解图像总体散布矩阵得到行列两个方向的最优投影向量,然后把人脸图像向这两个方向变换得到人脸识别特征矩阵,最后应用支持向量机进行分类识别。在ORL人脸库和Yale人脸库上对该算法进行实验研究,表明此方法在识别性能上优于普通的二维主成分分析和普通的双向二维主成分分析算法。In order to improve the accuracy of face recognition, and take into account of fact that the average of the training samples is not always the scatter center of the samples, two directional two dimensional principal component analysis method（2DPCA） based on the sample median is proposed in this paper. Firstly, the median of training samples are is used instead of the average to rebuild the covariance matrix, and find the solution of the covariance matrix to obtain the optimum projection vector of two directions in line and raw. Then the face images are transformed to these two directions and the face recognition characteristic matrixes are gotten. Finally, SVM is used to carry out classification recognition. The experimental results based on ORL face database and Yale face database indicate that the proposed method is better than traditional 2DPCA and two directional 2DPCA.

关 键 词：人脸识别 双向二维主成分分析方法 样本中间值 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]