局部广义张量鉴别分析

摘要

在过去十年里,致力于研究把线性鉴别分析扩展到更高阶数据分类,即多线性鉴别分析,以得到更好的鉴别效果.广义张量鉴别分析(GTDA)方法是其中最具代表性的算法之一.文中提出了一种新的多线性鉴别分析方法,即局部广义线性鉴别分析(LGTDA)方法.其利用张量样本的局部近邻信息重新定义了鉴别分析中的类间散度矩阵和类内散度矩阵,使得提出的方法比其他方法在投影空间中更好地保留原始空间的局部结构信息.另外,用多种特征提取技术提取出原始样本图片的各种信息构成文中算法的张量样本,充分利用了张量数据的优势.在AR和CAS-PEAL人脸数据库上的实验结果验证了文中方法的有效性.%In the past decade,great efforts have been made to extend linear discriminant analysis for higher-order data classification,gen-erally referred to as Multilinear Discriminant Analysis (MDA) . General Tensor Discriminant Analysis (GTDA) is one of the most repre-sentative algorithms. In this paper,propose a new multilinear discriminant analysis method named Local General Tensor Discriminant A-nalysis (LGTDA) . It utilizes the local neighbor information of tensor sample to redefine the between-class scatter matrix and the within-class scatter matrix,which makes the proposed method can preserve the local structure information of the original space in the projective space. In addition,to make better use of tensor technique,the tensor sample consists of information extracted by several feature extraction techniques. The experimental results on AR and CAS-PEAL databases demonstrate the effectiveness of the proposed method.