基于统计模式的人脸识别方法  

Face Recognition Approaches Based on Statistical Pattern

作　　者：陈旭[1] 苏开娜[1] 段娟[1] 

机构地区：[1]北京工业大学计算机学院,北京100022

出　　处：《计算机工程与应用》2005年第31期55-57,81,共4页Computer Engineering and Applications

基　　金：北京市自然科学基金项目(编号:40031004);北京市教委科技发展计划项目(编号:km200310005006)

摘　　要：人脸识别的研究具有广泛的应用价值。论文对基于统计模式的人脸识别方法进行了深入的研究,提出了使用D-LDA与KDDA算法对特征进行提取。D-LDA算法克服了传统方法计算复杂的缺点;KDDA算法利用了基于核的分析方法,避免了利用线性方法处理问题产生的缺陷。文中重点研究了RBF网络分类器,对于参数的选择以及网络的学习过程采用了改进的算法,克服了常用方法中存在的问题。实验表明采用改进后的算法,识别率得到了提高。The research of face recognition has found a wide range of applications.In this article,more detailed discussions have been concerned based on the statistical pattern of face recognition.D-LDA and KDDA algorithms have been adopted to extract features.D-LDA method solves computational problems associated with traditional algorithms.By using kernel-based analysis method,KDDA algorithm avoids the problem existed in linear methods.The article pays more attention to the research of RBF neural classifier.Improved approaches have been proposed to the selection of parameters and the learning procedure of network,which solves the problems of conventional methods.The experiments show that the recognition accuracy has been enhanced after adopting improved methods.

关 键 词：人脸识别 D-LDA KDDA RBF网络分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]