Abstract: For face recognition，most of traditional methods which reduce the high dimensional data are linear.Support Vector Machine can enhance the generation ability of study，and can overcome the disadvantage of overfitting.The paper proposes a method for face recognition based Supervised Neighborhood Preserving（SNPE） and Support Vector Machine（SVM）.The K of NPE is confirmed with supervised mode.SNPE aims at preserving the local manifold structure.Also，comparing to the recently proposed manifold learning algorithms such as Locally Linear Embedding，SNPE is defined everywhere，rather than only on the training data points，and it’s more applied.With SVM framework for multiple sequences，experiment results on ORL database demonstrate the algorithm is fast，steady and effective.

Key words: face recognition, Supervised Neighborhood Preserving Embedding（SNPE）, Support Vector Machine（SVM）

摘要： 在人脸识别方面，传统的特征提取方法大都是线性方法，不能很好保持样本的拓扑结构。分类方面，支持向量机能够尽量提高学习的泛化能力，防止过学习，是一种很好的分类器。提出了一种基于SNPE和SVM的人脸识别方法。采用有监督模式确定NPE算法中的K值。SNPE算法旨在保持数据的局部流型结构，而且相对于近期提出的LLE算法，它能够适用于训练样本和测试样本，具有更大的实用型。结合两分类支持向量机级联模型进行人脸识别，在ORL人脸数据库上实验表明，算法具有稳健性、快速性等优点，实验效果令人满意。

关键词: 人脸识别, 有监督近邻保持嵌入（SNPE）, 支持向量机（SVM）