一种基于DiaPCA和概率神经网络的人脸识别方法  

Face Detection Method Based on DiaPCA and Probabilistic Neural Network

作　　者：赵磊[1] 袁春婉 

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164 [2]湖南中科电气股份有限公司物联中心,湖南岳阳414100

出　　处：《军事通信技术》2010年第3期16-19,44,共5页Journal of Military Communications Technology

摘　　要：文章提出一种结合对角主成份分析DiaPCA(Diagonal Principal Component Analysis)的概率神经网络人脸检测方法。该方法先对样本数据进行降维处理,并采用了DiaPCA方法来避免降维时的信息丢失问题。它缩短了网络的训练时间,并大大的提高了判别精度。使用MITEx人脸数据库的1000个人脸与非人脸图像进行实验分析,并与结合LLE(Locally Lnear Embedding)方法降维后结合BP神经网络的方法比较。结果表明其较其他的方法具有更高的识别准确性。A face detection method based on Diagonal Principal Component Analysis(DiaPCA) and Probabilistic Neural Network(PNN) was proposed.Firstly,the dimension reduction was made to the sample data and the DiaPCA method was used to avoid the loss of information.Then,the classics BP neural network was used to do the face detection.It not only shortens the net training time,but also improves the accuracy of the recognition.It used 1000 face and non-face pictures of the MITEx face database for test and analysis,and was compared with the method that combined LLE and neural network.The simulation demonstrates that its effect of detection is better.

关 键 词：对角线性判别 神经网络 人脸检测 构造法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]