一种小波树和非参数鉴别分析的人脸识别

Face Recognition Based on Wavelet Tree and Non-parametric Discriminant Algorithm

结合小波树和子空间分析中的非参数鉴别分析,提出了基于小波树和非参数鉴别分析的人脸识别算法(WANDA).该算法先利用小波变换公式,在人脸图像上计算出一个小波近似分量和三个细节分量;然后对小波近似分量进行二次小波分解,相应地计算出各小波近似分量;其后对三层分解的小波近似系数进行重新组合,得到新的样本集;最后在此样本集上使用非参数鉴别方法进行人脸识别.ORL和CAS-PEAL-R1人脸库的实验结果表明,与基于线性非参数鉴别分析的人脸识别(NDA)和鉴别分析的人脸识别(LDA)方法相比较,WANDA方法的人脸识别率为97.5%,对光照条件、脸部表情变化有良好的鲁棒性.

Face recognition is proposed based on wavelet tree and non-parametric discriminant analysis(WANDA),which combines with the wavelet tree and non-parametric discriminant analysis.The algorithm calculates a wavelet approximation coefficient and three detail components in face images with the help of the wavelet transform formula.Then the wavelet approximation coefficient decomposes two wavelets.Each wavelet approximation coefficients is calculated and three level of wavelet approximation coefficient are recombined.In the end,a new sample set is obtained and face recognition is completed by using non-parametric discriminant analysis.The simulation results of experiments on the ORL and CAS-PEAL-R1 face database show that compared face recognition based on linear nonparametric discriminant analysis(NDA)with face recognition based on linear discriminant analysis(LDA),the new algorithm has a high recognition rate of 97.5% and better robustness to illumination condition and face expression change.