基于SOFM的人脸图像聚类方法比较  

Comparison of face cluster methods based on SOFM

作　　者：刘燕子[1] 曾晓勤[1] 

机构地区：[1]河海大学计算机与信息学院智能科学与技术研究所,南京211100

出　　处：《信息技术》2013年第4期162-165,共4页Information Technology

摘　　要：通过实验探讨了几种人脸图像聚类方法的效果,并给出了一些定性的结论。首先是试图找出数据量的大小对聚类效果的影响,实验结果表明,聚类数据量的增加可以使聚类结果更好,并且使用PCA方法提取人脸特征时,人脸轮廓信息越多,聚类结果越好;其次是将人脸图像按五官分割成不同的部分,然后分别使用PCA和ICA方法提取特征进行聚类,实验结果表明使用ICA方法比使用PCA方法提取的特征的聚类效果好。This paper discusses on the effects of several face cluster methods through the experiments, and draws some qualitative conclusions. Firstly, the effect of image amount on the cluster is considered. The experimental results show that the increase of image amount can make a better clustering, and if facial features are extracted with PCA method, the more contour information, the better. Secondly, the segmentation of a face according to its components is considered and the features are extracted separately with PCA and ICA methods. The experimental results show that the clustering result of using ICA is better than the one using PCA.

关 键 词：人脸聚类 SOFM PCA ICA 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]