Abstract: Locality Preserving Projection (LPP) is a manifold learning method, while the face recognition application of LPP is known to suffer from singular value problem, so a solution scheme using Singular Value Decomposition (SVD) was proposed for recognition application. In this model, the sample data were projected on a non-singular orthogonal matrix to solve the problem of singular value. Then the data of the low dimensional sample space projection subspace were obtained according to the LPP method. The training samples and testing samples were projected onto low-dimensional subspace respectively. Finally the nearest neighbor classifier was used for classification. A series of experiments to compare the proposed algorithm with the traditional local projection algorithm and Principal Component Analysis (PCA) were given on ORL face database. The experimental results demonstrate the efficacy of the improved LPP approach for face recognition.

Key words: manifold learning, Locality Preserving Projection (LPP), Singular Value Decomposition (SVD), face recognition, pattern recognition

摘要： 局部保持投影算法是基于流形的学习方法，在人脸识别过程中容易遇到奇异值问题，为此提出一种利用奇异值分解的方法。在模型中，样本数据被投影到一个非奇异正交矩阵中，解决了奇异值问题；然后再根据局部保持投影算法求出新样本空间的低维投影子空间。将训练样本和测试样本分别投影到低维子空间中，再利用最近邻分类器进行分类识别。在ORL人脸数据库中，采用了一系列的实验来对比该算法与传统局部保持投影算法和主成分分析算法的识别效果。实验结果验证了改进的局部保持投影算法在人脸识别的有效性。

关键词: 流形学习, 局部保持投影, 奇异值分解, 人脸识别, 模式识别