多分类器结合的人脸识别

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。

Combining Classifiers inFace Recognition

This paper discussed the familiar combination methods like sum, product, median and vote rules according to the Bayesian theory and pointed out their weakpoints. 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 difference 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 better than those of old ones.