Abstract: Nonnegative Matrix Factorization（NMF） is a new matrix decomposition technique.A modified NMF algorithm called Neighborhood Preserving Discriminant Nonnegative Matrix Factorization（NPDNMF） is proposed for enhancing the classification accuracy of the NMF algorithm.Neighborhood Preserving Discriminant Analysis（NPDA） is a method constructed by combining the ideas of both Linear Discriminant Analysis（LDA） and Local Preserving Projection（LPP），which can hold the strong power of LDA and preserve the intrinsic geometry of the data samples.The proposal incorporates the discriminant constraints of NPDA inside the NMF decomposition，which yields a part based decomposition with enhanced discriminant power.The experiment on ORL face database for face recognition shows that the proposal obtains a better performance.

Key words: Linear Discriminant Analysis（LDA）, Neighborhood Preserving Discriminant Analysis（NPDA）, Local Preserving Projection（LPP）, Nonnegative Matrix Factorization（NMF）

摘要： 非负矩阵分解（NMF）是一种新的矩阵分解技术，为了提高NMF算法的识别率，提出了一种新的方法——邻域保持判别非负矩阵分解（NPDNMF），该方法通过将邻域保持判别分析（NPDA）与NMF相结合来实现。邻域保持判别分析是一个将线性判别分析（LDA）与局部保持投影（LPP）综合考虑的判别分析方法，该算法既保持了LDA的判别能力，同时又可以保持原始数据的几何结构。通过将NPDA与NMF相结合，提取得到局部化同时又有很强判别能力的基图像。在ORL人脸数据库上进行人脸识别实验，结果表明该方法得到较好的识别效果。

关键词: 线性判别分析, 邻域保持判别分析, 局部保持投影, 非负矩阵分解