研究了核理论和二维矩阵表示的非线性特征提取方法。在对向量和矩阵表示最大间距准则进行分析基础上,提出了一种核二维最大间距准则的非线性判别方法。该算法是对二维最大间距准则的核化推广,不但有效利用了图像的空间结构信息,而且分别在两个特征空间提取判别特征。在ORL和Extended Yale-B人脸数据库上的实验表明了该算法的有效性。
This paper studied the kernel theory and two dimensional matrix representation for nonlinear feature extraction me-thod.Based on the analysis of maxmum marginal criterion of vector and matrix representation,it proposed a nonlinear discriminant analysis method called kernel two dimensional maximum margin criterion(KTDMMC) to extract nonlinear features.It was an extension to kernelization of two dimensional maxmum marginal criterion(TDMMC),which not only effectively utilized the underlying spatial structure of images,but also extracted the discriminating information in two kernel subspaces respectively.Experiments on ORL and Extended Yale-B face recognition databases demonstrate the effectiveness of the proposed method.