中文摘要：

传统的基于PCA（Principal component analysis）的人脸识别方法产生的人脸特征子空间通常是由人脸库中的所有训练样本产生的，此子空间包含的更多的是所有人脸样本的共性特征，而忽略了人脸的一些个性特征。本文提出了一种基于PCA图像重构的人脸识别方法，该方法以单个人的类内协方差矩阵为特征脸产生矩阵，获取个人的人脸特征子空间，然后将待识别图像对每个特征子空间进行映射提取人脸图像主成分，并以此主成分进行图像重构，采用最小重构误差作为判据实现人脸的识别，最后基于ORL及Yale人脸数据库，实验验证了该方法的有效性。

英文摘要：

Traditional PCA （Principal component analysis）-based face recognition methods can obtain a universal subspace by all training images, and the subspace represents the commonness of the human face and lose sight of the individuality of a person face. A new method for the image reconstruction is presented based on PCA for the facial recognition. Inner-class covariance matrix for the feature extraction is used, and then eigenvectors from each person are obtained. Furthermore, eigenvalues are obtained by mapping the test images to sub-eigenspace and the eigenvalues are utilized to reconstruct the images to realize the facial recognition based on the minimum reconstruction error. The simulation illustrates the effectivity of the method on the ORL face database and the Yale face database.