不相关最佳鉴别矢量集的有效算法

Efficient Algorithm to Optimal Set of Uncorrelated Discriminates Vectors

线性鉴别分析中处理小样本问题的方法有两类:①在模式识别之前,通过降低模式样本特征向量的维数达到消除奇异性的目的;②发展算法获得低维鉴别特征。将这两种方法结合起来,解决了高维小样本情况下基于广义Fisher线性鉴别准则的不相关最优鉴别矢量集的求解问题,给出了抽取最优鉴别矢量的有效算法。

Nowadays there are two kinds of methods for dealing with the problems of small sample size in linear discriminant analysis. One is that the aim of avoiding singularity is arrived by dimension reduction of feature vector of pattern samples before pattern recognition. The other is to develop an algorithm to gain the lower discriminant features. By combining the above two kinds of methods, the problem has been solved that how to gain the optimal set of uncorrelated discriminant vectors for small sample size problem based on the generalized Fisher＇ s linear discriminant criterion. An efficient algorithm has been presented in this paper.