摘要
在人脸识别中,高维、小样本是一个问题。对此,提出了一种基于Gabor小波与径向基函数(RBF)神经网络的人脸识别方法。首先对人脸进行Gabor滤波,选取有效的Gabor组合。进行小波分解,获取低频图像,构造特征矢量,采用主分量分析降低特征维数。接着,提出了一种聚类方法用于确定RBF神经网络的结构和初值,采用混合学习法训练RBF神经网络。用ORL人脸库进行试验,结果表明本文提出的方法具有优秀的学习效率和识别效果。
A method based on Gabor wavelets and radial basis function (RBF) neural networks to cope with small training sets of high dimension, which is a problem encountered in human face recognition, is presented. Human face images are processed firstly by the Gabor filter, and to select effective Gabor combinations. The results are decomposed by the wavelet to get low frequency images. Human face feature vectors is formed, which are further processed by the principal component analysis to reduce feature dimensions. Then, a classification method is proposed to determine the structure and initial parameters of the RBF neural networks. A hybrid algorithm is used to train the RBF neural networks. Finally, experiment results conducted on the ORL database show that the system achieves excellent performance both in recognition rates and learning efficiency.
出处
《电路与系统学报》
CSCD
北大核心
2008年第1期73-78,共6页
Journal of Circuits and Systems