一种无约束的最优判别向量集  

Unconstrained optimal set of discriminant vectors

作　　者：曾宪贵[1] 黎绍发[1] 左文明[1] 

机构地区：[1]华南理工大学计算机学院

出　　处：《计算机工程与设计》2006年第3期492-493,共2页Computer Engineering and Design

摘　　要：在正交约束条件下,求使Fisher准则判别函数式取极大值的向量,这样的最优判别向量就是F-S最优判别向量集。基于Fisher判别准则函数式,提出了一种无约束的最优判别矢量集,并给出了求解算法。另外,当训练样本矢量数小于样本矢量维数(即小样本问题),类内散布矩阵奇异,为了使它非奇异,采取对样本进行降维的措施,那维数至少要降到多少维才能确保它非奇异,给出了计算公式。实验结果表明鉴别矢量集有良好的分类能力。Under condition oforthogonal constraints, the vectors that make the fisher disciminant criterion function attain the maximum values are F-S optimal set of discriminant vectors. An optimal set of discriminant vectors which need not fill any constraint condition was presented, and the way of how to get the set was presented too. In addition, when the number of training samples was smaller than the dimensions of training samples （i.e. small number of training samples problem）, the within-class scatter matrix was singular. In order to let it nonsingular, the measure of making the dimension reduction was used, but it should be reduced to how much at least to ensure it was nonsingular, the calculation formula was given out. Experiment results indicate the ability of the novel optimal set ofdiscriminant vectors.

关 键 词：人脸识别 最佳鉴别矢量集 特征提取 模式识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]