基于多子空间线性判别分析的特征提取与分类

Feature extraction and classification based on multisubspace linear discriminant analysis

为了解决高维小样本数据的分类中Fisherface思想判别分析方法的不足,在最大散度差准则的基础上,提出了利用多线性子空间技术对每类样本进行单独描述的方法,该方法能更准确地反映样本在类内类间的分布关系。在分类中不是以距离作为判别依据,而是按照贝叶斯决策规则得到的隶属置信度作为衡量标准。实验结果表明了该方法的有效性,和同类方法相比,有更高的识别率。

In classification of high-dimensional statistical data underlying small sample size problem, in order to solve Fisherface thought lack of discriminant analysis method, sample of each class is described using multisubspace technique and maximum scat- ter di{ference criterion to reflect sample between-class and within-class distribution more accurate. Criterion of classit＇ication is confidence from Bayes rule ranther than distance. Results demonstrated that the performance of the proposed method is superior to that of traditional approaches. As far as recognition rate is concerned, a marked improvement is obtained.