基于白化 PCA图像重构的特征补偿人脸识别新方法

摘要

针对基于主成分分析（principal component analysis，PCA）方法在特征提取过程中丢弃高阶统计信息的缺陷，提出了一种基于图像重构的特征补偿人脸识别算法。首先利用白化PCA方法提取原始图像特征，对图像进行重构并计算残差图像；然后对残差图像进行白化PCA特征提取，并将其作为第一次提取特征的有效补偿以得到新的特征；最后用最近邻分类器进行识别分类。在ORL、YALE、XM2VTS和AR人脸数据库上的实验结果验证了算法的有效性。%According to the defect of PCA method which discards high-order statistical information in the process of feature ex-traction,this paper proposed a new feature compensation method for face recognition based on image reconstruction.Firstly,it extracted features from the original images using whitening PCA method,and it reconstructed the images and calculated the re-sidual images.Secondly,it extracted features from the residual images using whitening PCA method,these features were effec-tive compensation for previously obtained features to get the new features.Finally,it used nearest neighbor classifier for classifi-cation.Experiments on ORL,YALE,XM2VTS and AR face databases demonstrate the effectiveness of the proposed algorithm.