一种基于正则化方法的非负矩阵分解算法研究与应用

Research and Application of a Nonnegative Matrix Factorization Algorithm Based on Regularization Method

信息化技术的快速发展为非负矩阵分解提供了更加广阔的应用空间,这种全新的矩阵分解及特征提取方法,在数据聚类方面具有良好的应用前景。但是,非负矩阵的特征决定了这种方法无法同时考虑样本信息特征和几何结构信息特征。本文提出基于正则化方法的非负矩阵分解算法,通过硬约束维持样本分类信息,并有效保持样本的原有几何结构,实验表明该方法在图像分类方面有良好的应用效果。

Non-negativxe matrix factorization is a new matrix decomposition and feature extraction method,which has a prominent role in linearly separating data to achieve data clustering.However,non-negative matrix factorization cannot consider the characteristics of geometric structure information and sample class information at the same time.This paper proposes a non-negative matrix factorization algorithm based on regularization method,which maintains sample classification information through hard constraints and effectively maintains the original geometry of the sample.