基于核偏最小二乘回归的人脸识别

Face Recognition Using Kernel Partial Least Squares Regression

在采用主成分分析进行人脸重构和识别时,仅从样本自身提取特征向量会导致识别误差。因此,在参考主成分分析的基础上,采用偏最小二乘回归进行人脸图像的训练和识别,并对偏最小二乘回归引入核函数。在ORL人脸数据库上的实验结果表明,偏最小二乘回归明显优于主成分分析,同时核偏最小二乘回归也显著提高了识别正确率。

When representing and recognizing face using principal component analysis（PCA） method,the eigenvectors extracted from training samples may lead to some errors.Based on PCA method,face training and recognition method using partial least squares regression（PLSR） is presented in this paper.Furthermore,a kernel function is introduced to PLSR.Experimental results on ORL face database show that PLSR method is superior to PCA method and the accuracy of recognition is improved obviously using kernel PLSR.