一种改进的Fisher判别准则

A Modified Fisher's Diseriminant Criterion

摘要: 在Fisher 判别准则下求得的最佳鉴别矢量是一种有效的模式分析技术。可是在小样本集条件下,由于Fisher 准则中的类内散布矩阵Sw 是奇奇矩阵,使得通常在大样本下建立的形如S_bX=λSwX 的广义本征方程无解,因而Fisher 准则不能用于求解小样本下的最佳鉴别矢量。本文提出了一种改进的Fisher 判别准则,它能在大,小样本条件下求出精确的最佳鉴别矢量。实验结果表明,改进后的准则比原Fisher 准则更通用和有效。

Abstract: In this paper a modified Fisher's discriminant criterion is presented,which can find accurate optimal discriminant vectors for a large number of samples and a small number of samples. Our experimental results show that the modified criterion is more effective and general than Fisher's discriminant criterion.