Abstract: This paper proposes a novel face recognition algorithm of 2D Nonparametric Subspace Analysis(2DNSA) based on 2D Principal Componet Analysis(2DPCA) subspace. The original face matrices are performed to have feature dimension reduction, and the reduced feature matrices are used as a new training set, which can be conducted by 2D non-parametric subspace analysis. This method not only can reduce feature dimensions by 2DPCA, but also consider the impact of boundary samples for classification by taking full advantage of classification capacity of 2DNSA, which avoids the irrationality of using class centers to measure the distances of different classes. Experimental results on the two face databases(namely Yale and LARGE) show the improvements of the developed new algorithm over the traditional subspace methods such as (2D)2PCA, 2DPCA, (2D)2LDA, 2DLDA, 2DPCA+2DLDA, 2DNSA, etc.

Key words: face recognition, feature extraction, 2D Nonparametric Subspace Analysis(2DNSA), 2D Principal Component Analysis(2DPCA), 2D Linear Discriminant Analysis(2DLDA)

摘要： 提出一种结合二维PCA(2DPCA)的二维非参数子空间分析(2DNSA)人脸识别算法。利用2DPCA对原始图像矩阵进行特征降维，以降维后的特征为训练样本，进行二维非参数判别分析，并综合考虑类边界样本对分类的影响，采用2DNSA实现更合理的特征提取。基于Yale、LARGE人脸数据库的实验结果表明，与(2D)2PCA、2DPCA、(2D)2LDA、2DLDA、2DPCA+2DLDA、2DNSA算法相比，该算法性能更优。

关键词: 人脸识别, 特征提取, 二维非参数子空间分析, 二维主成分分析, 二维线性判别分析

CLC Number: 

• TP391.41