一种新的求解无相关鉴别矢量集方法  

A New Method for Resolving the Uncorrelated Set of Discriminant Vectors

作　　者：陈绵书[1] 陈贺新[1] 刘伟[1] 

机构地区：[1]吉林大学通信工程学院,长春130025

出　　处：《计算机学报》2004年第7期913-917,共5页Chinese Journal of Computers

基　　金：国家自然科学基金 ( 60 172 0 46)资助

摘　　要：无相关鉴别矢量集方法是解决模式识别问题的有效方法 .通常情况下 ,无相关鉴别矢量集是通过递归方式获得的 ,计算时间较长 .该文提出了一种求解无相关鉴别矢量集的非递归方法 .首先根据总体散布矩阵构造无相关投影空间 .对于无相关投影空间中的任何正交矢量集 ,其在原空间中的特征统计无关 .然后在无相关投影空间求解基于Fisher线性判别准则的正交矢量集 ,从而得到原空间的无相关鉴别矢量集 .理论分析和实验结果表明 :该文方法和Jin等的方法所求解的无相关鉴别矢量集是一致的 .而应用本文方法求解无相关鉴别矢量集计算时间较短 ,在类别数为C的情况下 ,二者的时间比为 (C - 1)∶2 .The classification method using the uncorrelated set of discriminant vectors is a good method to resolve the problem of pattern recognition. Usually, the uncorrelated set of discriminant vectors is obtained recursively, so the computation is time-consuming. This paper proposes a non-recursive method for resolving the uncorrelated set of discriminant vectors. First, an uncorrelated projection space is constructed based on the total population scatter matrix. For any orthogonal vectors in the uncorrelated projection space, the obtained features in the original space are statistically uncorrelated. Then the uncorrelated set of discrimiant vectors is obtained by resolving the orthogonal vectors based on Fisher discrimiant criterion in the uncorrelated projection space. The theoretical analysis and the experiment results show that the uncorrelated set of discriminant vectors derived from proposed method is the same as that derived from the method proposed by Jin in 2001, while the computation time is shorter in our method. If the number of classification is C, the computation time ratio of these methods is C-1∶2.

关 键 词：总体散布矩阵 无相关空间 无相关鉴别矢量集 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]