基于LLE和BP神经网络的人脸识别_吴俊强.pdfVIP

56 2006 275 LASER JOURNAL(Vol.27.No.5.2006) 71 LLE BP 1 1,2 1 1 吴俊强, 周激流 , 何 坤, 郎方年 1. 四川大学电子信息学院,成都, 610064; 2. 四川大学计算机学院,成都, 610064 : LLE , BP , LLE , , , ORL , : :LLE; ; BP ; :TN248. 1 :A :0253- 2743(2006)02- 0071- 03 Face recognition on LLE algorith and BP- based neural networks 1 1, 2 1 1 WU Jun- qiang ZHOU Ji- liu ,HE Kun ,LAN Fang- nian 1. ElectronicInformation school, Sichuan University, Chengdu, 610064 China; 2. Computer School, Sichuan University, Chengdu, 610064 China Abstract:The locally linear embedding ( LLE) algorithm is presented to extract thefacefeature in thepaper. And then thefeaturecoefficients of each sample are trained in theBP neural networks for facerecognition. Thelocally linear embedding algorithm can reveal theintrinsicdistribution of data, which cannot be pro- vided by classical linear dimensionalityreductionmethods. The nonlinear structure in high dimensional data spacewas exploitedwith thelocal symmetries of linear reconstructions. The data points in high dimensional space were mapped into corresponding data points in lower dimensional space under preserving distance be- tween data points. In this paper, theLLE algorithm reduces the dimensionality of image data and reveals the intrinsic distribution of data. And this method also has invariability in translation and rotation. Experiment results on ORL database demonstrate that themethod is effective. Key words:locally linear embedding; LLE; non- linear dimensionality reduction; BP neural networks;face recognition , 1 引言 , ( )