摘要
本文在ZHong等人使用的奇异值分解 (SVD)基础上 ,将人脸图像矩阵的奇异值作为识别特征 ,解决了奇异值处理、神经网络训练策略和竞争选择问题 ;运用BP网络进行识别 ,提出了一种基于奇异值特征的神经网络人脸识别新方法 .基于ORL人脸数据库的多次反复实验结果表明 ,在大样本情况下 ,识别方法具有实现简单、识别速度快、识别率高的特点 ,为人脸的实时识别提供了一种新途径 .
In this paper,singular value processing,training strategies of neural networks and competition selection are solved on the basis of singular value decomposition(SVD) applied by Z Hong et al.A new approach for face recognition on singular value features and neural networks is presented,in which singular values of face image matrix are used as features,and Back-Propagation(BP) network as recognition.Repeated experimental results on ORL(Olivetti Research Laboratory) database demonstrate that under the conditions of large samples recognition method has the characteristics of simple realization,rapid recognition speed and high recognition rate,and also profits real-time realization.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2004年第1期170-173,共4页
Acta Electronica Sinica
基金
广东省自然科学基金 (No .0 32 356)
关键词
人脸识别
奇异值特征
神经网络
模式识别
face recognition
singular value features
neural networks
pattern recognition
