多分类器结合的人脸识别  

Combining Classifiers in Face Recognition

作　　者：陈刚[1] 戚飞虎[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200030

出　　处：《上海交通大学学报》2001年第2期173-176,共4页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金!资助项目 (6 9772 0 0 2 )

摘　　要：用贝叶斯理论分析了常见的和、积、中值及投票多分类器结合方法 ,指出它们各自的缺陷 ,类比真实的选举情形 ,对原投票法进行了改进 :赋予不同分类器不同的“说话份量 ,被重视程度”,即不同权值 ,增加“第二候选人”备选 ,并考虑“第一与第二候选人”的可信度差给予“附加选票”.采用Olivetti和 Oracle研究室的人脸图像库 ,结合本征脸法、协同算法和自联想神经网络法分类器 ,对比了新方法和常见结合方法 .实际结果表明 ,改进的方法有较好的识别率 .This paper discussed the familiar combination methods like sum, product, median and vote rules according to the Bayesian theory and pointed out their weak points. Inspired by election in the real world it improved the majority vote rule: different classifiers have different 'loud voice' which means different weights; a 'second candidate' is added; the different reliability of the first and the second candidate is used to give 'bonus votes'. Finally, a face recognition experiment using ORL (Olivetti and Oracle Research Lab's) face image database was presented. Eigenface, synergetic algorithm and auto-associative neural network were combined. The result shows that the recognition rate of the new method is higher than those of old ones.

关 键 词：人脸识别 本征脸法 协同算法 自联想神经网络法 分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]