摘要： 针对高维数据的特征提取问题，将广义低秩矩阵近似(GLRAM)与对角主成分分析(DialPCA)相结合，提出一种新的特征提取方法GLRAM Plus DialPCA用于进行图像识别。通过广义低秩矩阵对原始图像进行近似，再做对角化变化，采用二维主成分分析(2DPCA)提取数据行列之间的相关性特征，并利用最近邻分类器计算图像识别率。基于FERET和ORL人脸数据库的实验结果表明，与单一的GLRAM或2DPCA相比，GLRAM Plus DialPCA在姿态、光照和表情变化的情况下识别率更高，特征提取速度更快。

关键词: 广义低秩矩阵近似, 对角主成分分析, 二维主成分分析, 人脸识别, 特征抽取, 最近邻分类器

Abstract: Aiming at the feature extraction problem for high dimensional data, this paper proposes a feature extraction method combining Generalized Low Rank Approximations of Matrices(GLRAM) and Diagonal Principal Component Analysis(DialPCA) to realize image recognition. It uses generalized low rank matrix to approximate original image, does diagonalization changing, and uses 2D Principal Component Analysis(2DPCA) to extract the correlation between the row and column of data matrix. Nearest neighbor classifier is adopted for recognition rates calculation. Experiments on FERET and ORL face databases show that under the pose illumination and expression changing circumstances, recognition rate of GLRAM Plus DialPCA is higher than single GLRAM or 2DPCA.

Key words: Generalized Low Rank Approximations of Matrices(GLRAM), Diagonal Principal Component Analysis(DialPCA), 2D Principal Component Analysis(2DPCA), face recognition, feature extraction, nearest neighbor classifier

中图分类号: 

• TN911.73