基于RBF神经网络的人脸识别算法  

Face Recognition Algorithm Based on RBF Neural Networks

作　　者：肖南峰[1] 姚永刚[1] 

机构地区：[1]华南理工大学计算机科学与工程学院,广州510006

出　　处：《重庆理工大学学报（自然科学）》2010年第10期56-67,共12页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金与中国民用航空总局联合资助项目(60776816);广东省自然科学基金重点资助项目(251064101000005)

摘　　要：针对人脸识别技术中存在的高维问题、小样本问题和非线性问题这3个难点,围绕人脸特征提取和人脸识别2个方面展开研究。在特征提取中,采用基于主成分分析和Fisher线性鉴别来克服在人脸识别中的小样本问题,同时也将人脸图像从高维空间映射到低维空间,从而解决了高维问题;在分类识别方面,采用具有很强非线性映射功能的RBF神经网络进行模式分类,解决人脸识别中的非线性问题。利用Matlab分析了RBF网络的聚类性能和分类性能。在ORL人脸数据库上的仿真实验中,人脸识别率达到97.5%,取得比较满意的结果。结合OpenCV中的Haar特征和AdaBoost算法进行人脸检测,在VC平台下开发出基于VC++和OpenCV的人脸识别系统软件,系统界面友好,操作简便,扩展性良好。In this paper, feature extraction and recognition of facial images is studied in order to resolve the high-dimension problem, small size samples problem and no-linear separable problem that exist in face recognition technology. The proposed feature extract method based on Principal Component Analysis （PCA） and Fisher＇ s Linear Discriminate （FLD） can solve the small size samples problem and the high-dimension problem by mapping the samples from a high-dimension space to a low-dimension Eigen space. In the recognition stage, RBF neural network, which represents a brilliant performance on small training set, non-linear separable and high-dimension pattern recognition problems, is used for pattern classification. In the experiment, the performances of both clustering and classification are evaluated by Matlab. Simulation results on the ORL face database indicate that the proposed method for face recognition yields a good recognition rate nearly 97.5 %. At last, after detecting face with Haar features and AdaBoost algorithm, a face recognition system software is implemented based on VC ＋＋ and OpenCV, which has a good interface and expansibility.

关 键 词：主成分分析 FISHER线性鉴别 RBF神经网络 ADABOOST 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]