Abstract: There are large numbers of face recognition algorithms and face databases，and how to select classified algorithms and training sets has become a sixty-four-dollar question of face recognition at present.In allusion to the face recognition algorithms of wave transform and principal component analysis，this paper demonstrates that the 3-neighbor classification is propitious to Euclid classification，and that selecting preceding 8 images of ORL as training sets is propitious to other instances through large numbers of simulative experimentation on the platform of Matlab6.5.The experiment educes eigen-face databases of best instances，and the linearity expression of the test images using average face images and eigen-face images.It has high applicable worthiness at the domain of face recognition.

Key words: wave transform, principal component analysis, 3-neighbor classification, face recognition, simulation

摘要： 目前，存在大量的人脸识别算法和人脸库，如何选择分类算法，如何选择训练集，成为人脸识别的一个关键问题。针对小波变换和主元分析的人脸识别算法，在Matlab6.5平台上，通过大量的仿真实验，根据计算速度和识别率得出，三阶近邻分类优于欧几里德分类，选择ORL人脸库的前8幅图像作为训练集优于其他情况。实验得出了最优情况下的特征脸库，及测试图像用平均脸图像和特征脸图像的线性表示。在人脸识别领域具有很强的应用价值。

关键词: 小波变换, 主元分析, 三阶近邻分类, 人脸识别, 仿真