Fisher线性鉴别分析已成为特征抽取的最为有效的方法之一 .但是在高维、小样本情况下如何抽取Fisher最优鉴别特征仍是一个困难的、至今没有彻底解决的问题 .文中引入压缩映射和同构映射的思想 ,从理论上巧妙地解决了高维、奇异情况下最优鉴别矢量集的求解问题 ,而且该方法求解最优鉴别矢量集的全过程只需要在一个低维的变换空间内进行 ,这与传统方法相比极大地降低了计算量 .在此理论基础上 ,进一步为高维、小样本情况下的最优鉴别分析方法建立了一个通用的算法框架 ,即先作K L变换 ,再用Fisher鉴别变换作二次特征抽取 .基于该算法框架 ,提出了组合线性鉴别法 ,该方法综合利用了F S鉴别和J Y鉴别的优点 ,同时消除了二者的弱点 .在ORL标准人脸库上的试验表明 ,组合鉴别法所抽取的特征在普通的最小距离分类器和最近邻分类器下均达到 97 的正确识别率 ,而且识别结果十分稳定 .该结果大大优于经典的特征脸和Fisherfaces方法的识别结果-Fisher linear discrimination analysis has become one of the most effective way to feature extraction, but in the case of high dimension and small sample how to extract Fisher optimal identification features is still a difficult, still hasn t completely solve the problem. In this paper, introducing the idea of compression mapping and isomorphism, ingeniously solved the high-dimensional theoretically, singular case to solve the problem of optimal identification vector set, and the whole process of the method to solve the optimal identification of vector set just in a low dimensional transformation space, compared with the traditional method greatly reduces the amount of calculation. Based on this theory, further to high dimension and small sample situation of optimal discrimination analysis method to establish the framework, a generic algorithm is first as K L transform, reoccupy Fisher identification transformations as a secondary feature extraction. Based on this algorithm framework, a