分块PCA鉴别特征抽取能力的分析研究

The Study of Extracting Ability of Discriminant Features for Modular PCA

基于主成分分析(Principal Component Analysis,PCA),本文提出了分块 PCA 人脸识别方法。分块 PCA 从模式的原始数字图像出发,先对图像进行分块,对分块得到的子图像矩阵采用 PCA 方法进行特征抽取,从而实现模式的分类。新方法的特点是能有效地抽取图像的局部特征,正是这些特征使此类模式区别于彼类。在 Yale 人脸数据库上测试了该方法的鉴别能力。实验的结果表明,分块 PCA 在识别性能上优于通常的 PCA 方法,也优于基于 Fisher 鉴别准则的鉴别分析方法:Fisherfaces 方法、F-S 方法、组合鉴别方法,识别率可以达到100%。

Based on Principal Component Analysis（PCA）, a new technique called Modular PCA is developed for human face recognition in this paper. First, in proposed approach, the original images are divided into smaller modular images, which are also called sub-images. Then, the well-known PCA method can be directly used to the sub-images obtained from the previous step for feature extraction, so the pattern classification can be implemented. The advantage of the represented way, when compared with conventional PCA algorithm on original images, is that the local discriminant features of the original patterns can be efficiently extracted, which are available to differentiate one class from another. To test Modular PCA and to evaluate its performance, a series of experiments were performed on Yale human face image databases. The experimental results indicate that the performance of the new method in terms of recognition rate is obviously superior to that of ordinary PCA algorithm on original images, and is superior to that of some discriminant analysis based on the Fisher discriminant criterion such as Fisherfaces, F-S and combination method.