基于BP神经网络的人脸识别后续分类算法  

Face recognition fellow-up classification algorithm based on BP neural network

作　　者：温洲[1] 邵晓巍[1] 龚德仁[1] 

机构地区：[1]上海交通大学航空航天学院,上海200240

出　　处：《计算机应用》2011年第A02期133-136,共4页journal of Computer Applications

摘　　要：采用BP神经网络(BPNN)对人脸识别进行分类。为了准确地将BP网络输出特征量进行特征归类划分,采用3种不同的后续分类方法将得到的待辨识数据进行特征归类和人脸识别:第一种方法是常用的根据输出向量的最大分量值辨别法;第二种是基于各分量值的门限阈值归类法;第三种是基于N维向量空间的中心区域分类法。实验表明,后两种方法在全局环境人脸识别中可行且有效,并在ATR人脸库仿真实验中,错误辨识率可低至2.2%,拒绝准确率可达到93.21%。Back-Propagation Neural Network（BPNN） was used in the face recognition in the paper.How to accurately classify and recognize these outputs which are obtained under the BPNN model,three different fellow-up classification approaches were used to classify obtained outputs.The first method,which is widely used,is maximum-component value recognition based on output vector;the next is threshold classification based on all components of output vector;the third approach,based on N-dimensional vector space,is central area classification.Experimental results through the face data of Auto Targets Recognition（ATR） show that the last two algorithms are feasible and effective for face recognition,and the error recognizing-accuracy is down to 2.2%,rejecting-accuracy is up to 93.21%.

关 键 词：BP神经网络 人脸识别 后续分类 阈值 向量空间 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]