基于最佳鉴别向量集的多级分类方法

Multistage Classification Method Based on Optimal Set of Discriminant Vectors

根据人的识别经验，采用多级分类方法，改进了最佳鉴别变换在模式类别数较多时的识别效果。首先从熵的大小和最佳鉴别向量集的分离能力两个角度来分析，当进行多级分类后，前者减小，即分类结果的可分性增加，后者得到增强。然后对人脸图像做奇异值分解和离散傅立叶变换，并分别提取最佳鉴别变换特征，用最近邻方法进行分类。在实验中，采用两级分类方法。

According to recogntion experiences, this paper uses a multistage classification method which improves the results of optimal discriminant transformation under more classes. From theoretical aspect, the method can make the entropy value of sort results decrease, and the divisibility of optimal set of discriminant vectors strengthen. The optimal discriminant features of face are extracted using singular value decomposition and fourier transform, and then they are classified by the nearest neighbor method. In the experiment, two stage classification method is used, thus the computation is raised very limitedly.