基于集成BP网络的人脸识别研究

Face recognition based on integrated back-propagation neural network

在对人脸图像使用小波变换进行数据压缩的基础上,使用PCA进行特征提取,再将特征输入集成BP神经网络实现对人脸图像的识别。集成BP网络将多分类问题转换为多个相互独立的二分类问题,在提高网络泛化能力的同时缩短了网络的训练时间。另外,集成网络通过增添子网络或者重新训练子网络的方法解决了网络"失忆"问题,使其具有增量式学习的能力。通过在ORL人脸库上仿真的实验,证明了集成网络的人脸识别以及增量学习都具有良好的性能。

After compressing of image in wavelet transform,obtaining features by PCA,then this paper inputted the features to integrated BP neural network to realize face recognition.It divided the multi-class classification into many independent 2-class classifications by the integrated BP neural network,then it could get better generalization ability and reduced training time.Additionally,the integrated BP neural network solved net＇s ＂forgetfulness＂ by adding new subnets into the integrated net or retraining subnets,so integrated network had the ability of incremental learning.The experiments in ORL show that the integrated BP neural network has a good performance on face recognition and the incremental learning.