基于多分类器组合的人脸识别

根据人类认知的规律，文中提出了一种基于整体和局部特征组合的人脸识别方法。奇异值特征是一种比较有效的代数特征，文中提取了整个人脸、双眼以及嘴部的奇异值特征。在组合过程中，提出了一种改进的将距离转换为后验概率估计值的方法，该方法既缩减了单一分类器的可能的模式类别，又对各分类器的输出进行了加权。实验结构表明文中方法是有效的。

Face Recognition Based on the Fusion of Multiple Classifiers

A face recognition system is presented based on the fus ion of global and local features in the aspect of cognitive psychology. Singular value features (SVFs) are utilized in the system. Besides the SVF of the w hole face the SVFs of the eyes and the mouth are also calculated. For the four k inds of features,Bayes classifier is adopted. When combining the results of differe nt classifiers,we put forward an improved version of the approach to the estima tion of the maximum a posterior probability (MAP),which can reduce the output c lasses of the single classifier and weight its results at the same time. Given a set of training samples,the intra-class distances of each kind of feature are first obtained. Then they are used as thresholds to control the output classes of each single classifier before the fusion of multiple classifiers. Experimenta l results validate the method.