融合2DUDT与2DLDA的人脸识别研究  

Fusing 2DUDT and 2DLDA for face recognition research

作　　者：贾红雯[1] 曹丹华 武小红[2] 武斌[1] JIA Hongwen;CAO Danhua;WU Xiaohong;WU Bin(Department of Information Engineering, Chuzhou Vocational Technology College, Chuzhou 239000, China;School of Electrical and Information Engineering, Jiangsu University,Zhenjiang 212013 ,China)

机构地区：[1]滁州职业技术学院信息工程系,安徽滁州239000 [2]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《南昌工程学院学报》2018年第1期45-48,53,共5页Journal of Nanchang Institute of Technology

基　　金：滁州职业技术学院科研项目(YJY-2015-04);安徽省质量工程项目(2016ckjh137);安徽省职业与成人教育学会教育科研规划课题(azjxh1635);安徽省高等教育振兴计划人才项目"高级优秀青年人才支持计划"(皖教秘[2014]181号)

摘　　要：提出了一种融合二维非相关判别转换和二维线性鉴别分析分别提取人脸图像矩阵行方向的非相关鉴别信息和列方向的线性鉴别信息的人脸识别方法。该融合方法将线性鉴别向量提取方法与非相关鉴别向量提取方法相结合。首先计算进行人脸图像矩阵列压缩时的类间散射矩阵和类内散射矩阵,用二维线性鉴别分析计算特征值和所对应的特征向量,用特征向量集对人脸图像矩阵进行列压缩。其次,计算人脸图像矩阵行压缩时的类间散射矩阵、类内散射矩阵和总体散射矩阵,用二维非相关判别转换求出最优投影矩阵并用最优投影矩阵的转置矩阵对人脸图像矩阵行压缩。最后用最近邻分类器对压缩的ORL人脸图像测试样本进行分类处理,可实现人脸图像的准确识别。A fusion of two-dimensional uncorrelated discriminant analysis（ 2 DUDT） and two-dimensional linear discriminant analysis（ 2 DLDA） is proposed for extracting uncorrelated discriminant information and linear discriminant information in the row and column of face image matrixes,respectively. The proposed method is the integration of 2 DUDT and 2 DLDA. On the one hand,after the within-class scatter matrix and the between-class scatter matrix are computed for face image matrix column compression,eigenvalues and eigenvectors are calculated with 2 DLDA,and then the image matrixes are compressed by the eigenvectors. On the other hand,after the within-class scatter matrix,the between-class scatter matrix and the total scatter matrix are computed for face image matrix row compression,the image matrixes are compressed by the optimal projection matrix calculated with 2 DUDT. Finally,the nearest neighbor classifier is applied to classify the ORL face images correctly.

关 键 词：二维非相关判别转换 人脸识别 二维线性鉴别分析 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]