最佳鉴别矢量集在无监督模式下的扩展  

Extending the optimal set of discriminant vectors for an unsupervised pattern

作　　者：曹苏群[1,2] 王士同[1] 陈晓峰[1] 邓赵红[1] 

机构地区：[1]江南大学信息学院,江苏无锡214122 [2]淮阴工学院机械系,江苏淮安223001

出　　处：《智能系统学报》2008年第6期511-522,共12页CAAI Transactions on Intelligent Systems

基　　金：教育部优秀人才支持计划资助项目(NCET-04-0496);教育部重点科学研究资助项目(105087)

摘　　要：基于Fisher准则函数的最佳鉴别矢量集是一种重要的有监督特征提取方法,在模式识别领域有着重要的影响.提出一种将最佳鉴别矢量集扩展到无监督模式下的方法,其基本思想是通过定义的模糊Fisher准则函数将Fisher线性判别扩展成一种半模糊聚类算法,通过该算法求得最佳鉴别矢量和模糊散布矩阵,进而构造出最佳鉴别矢量集.实验表明,在聚类有效性、分类准确率均优于无监督模式下常用的主成分分析特征提取算法.The optimal set of discriminant vectors, based on the Fisher criterion function, is an important supervised feature extraction method and has great influence in the area of pattern recognition. In this paper, an extension of the optimal set of discriminant vectors in unsupervised patterns is presented. The basic idea is to extend Fisher linear discriminants to a novel semi-fuzzy clustering algorithm through a predefined fuzzy Fisher criterion function. With the proposed algorithm, an optimal discriminant vector and fuzzy scatter matrixes can be figured out and then an unsupervised optimal set of discriminant vectors can be obtained. Experimental results for real datasets, testing clustering validity and correct classification recognition rates, demonstrated that this method is superior to the principal component analysis feature extraction algorithm in unsupervised patterns.

关 键 词：最佳鉴别矢量集 无监督模式 FISHER准则 半模糊聚类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]