类间散度特征选择法及其应用

本文介绍了统计模式识别中特征选择的一种方法--类间散度法.它是以两类细胞样本在n维特征空间中Fisher映射点集的类间差与类内差之比作为特征选择的判据.用这种方法对五类白血球进行了特征选择,并对所选特征用训练集和测试集进行了识别效果检验,同时,得到了特征选择的判据D与识别率之间的关系曲线.

A Method of Between-Class Divergence Feature Selection and its Application

In this paper, a method of feature selection in statistical pattern recognition thebetween-class divergence algorithm is introduced. It takes as its criterion of featureselection the ratio of between-class difference to within-class difference of the point setobtained from. Fisher's mapping of two classes of cell smaple in n-dimentional featurespace. The method is applied to feature selection for five classes of leucocytes, and therecognition effectiveness of the selected features is examined with train set and testset. The relation between feature selection criterion D and The correct recognition taleis obtained.