Fisher鉴别特征的最近邻凸包分类

基于Fisher准则的特征提取方法是模式识别技术的重要分支，其中，Foley-Sammon变换和具有统计不相关性的最佳鉴别变换是这一技术典型代表，本文将它们与一种新型分类器一最近邻凸包分类器相结合，从而实现Fisher鉴别特征的有效分类。最近邻凸包分类器是一类以测试样本点到各类训练集生成类别凸包的距离为分类判别依据的模式分类新方法，具有非线性性，无参性，多类别适用性等特点。实验证实了本文方法的有效性。

Nearest Neighbor Convex Hull Classification of Fisher Discriminant Features

Feature extraction based on Fisher criteria is a branch of pattern recognition.Foley-Sammon algorithm and the Uncorrelated Fisher Linear Discriminant Analysis(ULDA)are the classic two of those correlative methods.As preprocessors,they usually cooperate with other classification algorithm such as the minimum distance classifier,the nearest neighbor classifier or support vector machines(SVMs).In this paper,the results of them are used as inputs to a new classification method named the nearest neighbor convex hull(NNCH)classifier,which takes the convex hull of one class training data as a new unit class.The test sample will belong to the class of the nearest convex hull in the feature space.Nonlinearity,no parameters and multi-class applicability are the characters of NNCH.The experiments compared with the other cooperators mentioned above,indicate the good performance of the proposed methods.