分块非参数特征分析在人脸识别中的应用  

Modular nonparametric feature analysis for face recognition

作　　者：程强[1] 陈伟琦[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《计算机工程与科学》2013年第5期112-117,共6页Computer Engineering & Science

基　　金：江苏省科研创新计划项目(CXLX11_04910);中央高校基本科研业务费专项资金资助(JUSRT211A70)

摘　　要：基于非参数特征分析NFA方法,提出了分块NFA算法,并将其应用到人脸识别上。分块NFA算法首先对图像矩阵进行分块,对分块得到的子图像矩阵再应用NFA进行图像特征提取。这样做有两个优点:(1)能有效地抽取图像的局部特征,对人脸表情和光照条件变化较大的图像表现尤为突出;(2)与NFA相比,由于使用子图像矩阵,分块NFA可以避免使用奇异值分解理论,过程简便,并且克服了小样本问题。此外,NFA是分块NFA算法的特殊情况。在ORL和YALE数据库上进行的实验验证了本文所提方法要优于NFA算法。Based on the nonparametric feature analysis （NFA） method, the paper proposed a modular NFA algorithm and applied it on face recognition. Firstly, the modular NFA algorithm divides the original images into modular sub-images. Secondly, the NFA method is employed to extract the features of the modular sub-images. There are two advantages： 1） local feature of the images can be extracted efficiently, and it has outstanding effect for the images that have large variations in facial expression and illumination; （2） Singular value decomposition of matrix may be avoided in the process of feature extraction, which is simpler than that of other technologies such as NFA. Meanwhile, it overcomes the small sample size problem. Moreover, NFA is a special case of modular NFA. Experimental results on ORL and YALE face databases show that the performance of modular NFA is superior to NFA and modular NSA.

关 键 词：人脸识别 非参数特征分析 分块非参数特征分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]