一种具有统计不相关性的最佳鉴别平面  

AN OPTIMAL DISCRIMINANT PLANE WITH UNCORRELATED FEATURES

作　　者：金忠[1] 娄震[1] 杨静宇[1] 

机构地区：[1]南京理工大学计算机科学系,南京210014

出　　处：《模式识别与人工智能》1999年第3期334-339,共6页Pattern Recognition and Artificial Intelligence

摘　　要：在模式识别领域,基于Fisher鉴别准则函数的Sammon最佳鉴别平面技术有着重大的影响。特征抽取的一般原则是最好抽取模式的不相关的特征,而Sammon最佳鉴别平面的两个鉴别特征是统计相关的。文中提出了一种具有统计不相关性的最佳鉴别平面。对IRIS数据作了实验,实验结果表明,文中所提出的新的最佳鉴别平面优于Sammon最佳鉴别平面。Based on Fisher's discriminant function, Sammon's optimal discriminant plane has great influence in the area of pattern recognition. A general rule for feature extraction is to extract features as uncorrelated as possible, but the two features in Sammon's optimal discriminant plane are correlated. This paper presents an optimal discriminant plane with uncorrelated features. Experiments on IRIS data have been performed. Experimental results show that new optimal discriminant plane is superior to Sammon's optimal discriminant plane.

关 键 词：模式识别 特征抽取 特征选择 最佳鉴别平面 

分 类 号：O235[理学—运筹学与控制论] TP391.4[理学—数学]