一种求解Fisher最佳鉴别矢量的新算法及人脸识别  

A Novel Algorithm Solving Fisher Optimal Discriminant Vector and Facial Recognition

作　　者：郭跃飞[1] 黄修武[1] 杨静宇 

机构地区：[1]南京理工大学计算机系

出　　处：《中国图象图形学报（A辑）》1999年第2期95-98,共4页Journal of Image and Graphics

基　　金：国家自然科学基金;博士点基金

摘　　要：Ｆｉｓｈｅｒ最佳鉴别矢量是高维模式分析中的有效方法。当训练样本数相对于特征空间的维数较小时，就成了小样本问题。为了求解小样本问题，人们提出了一系列方法并取得了良好的效果。但在类内距离为零的情况下已有的方法均得不到最佳解，该文从理论上说明了这一点，并给出了一种在任何情况下都能得到最佳解的新算法，试验结果证实了这样的推断。Fisher optimal discriminant vectors method is an effective method in pattern analysis of high dimension. When the number of the training samples is small compared with the dimensionality of the feature space, the problem becomes the case of a small number of samples. People have proposed many methods for solving the problem of a small number of samples, and made great progress. However, none of the previous methods can obtain the optimal solution when within-class distance equals zero. This point is illustrated in terms of theory, and a new solving method that can obtain the optimal solution in any situations is presented in this paper. The experimental result confirms our inference.

关 键 词：模式分析 人脸识别 最佳鉴别矢量 小样本 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]