基于非负矩阵分解的特征向量抽取方法特点研究

计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (17): 154-156.

基于非负矩阵分解的特征向量抽取方法特点研究

郭勇^1,2，鲍丽春³

1.国防科技大学信息系统与管理学院，长沙 410073
2.北京系统工程研究所，北京 100101
3.北京航天飞控中心，北京 100094

收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-06-11 发布日期:2007-06-11
通讯作者: 郭勇

Characteristics of non-negative matrix factorization for feature extraction

GUO Yong^1,2，BAO Li-chun³

1.Information System and Management College，National University of Defense Technology，Changsha 410073，China
2.Beijing Institute of System Engineering，Beijing 100101，China
3.Beijing Aerospace Control Centre，Beijing 100094，China

Received:1900-01-01 Revised:1900-01-01 Online:2007-06-11 Published:2007-06-11
Contact: GUO Yong

摘要/Abstract

摘要：

非负矩阵分解算法可以作为一种新型的特征抽取方法。将非负矩阵分解算法和现有的其它三种现有的特征抽取算法进行详细比较：奇异值分解方法和非负矩阵分解方法本质上是不同的两种特征抽取方法，非负特性使得由非负矩阵分解比奇异值分解方法更接近人们的认知习惯。基于聚类的特征提取方法是一种简化了的非负矩阵分解算法;基于概率的特征提取方法等价于非负矩阵分解在特定约束条件下的变体。通过比较充分体现了非负矩阵分解算法的非负性和局部性特点。

Abstract: Non-negative Matrix Factorization（NMF） is a new algorithm for feature extraction.This paper compares NMF with three other existing feature extraction method：Singular Value Decomposition（SVD） is fundamentally different from NMF in feature extraction，but the non-negative constraints make the decomposition procedure of NMF much more like the process of human cognition than SVD;the clustering-based feature extraction method can be considered as a simplified NMF algorithm;and the probabilistic-based feature extraction method is proved to be one type of NMF algorithm with special constraints.Through these comparisons，we catch the non-negative and local characteristics of NMF.

郭勇^1,2，鲍丽春³. 基于非负矩阵分解的特征向量抽取方法特点研究[J]. 计算机工程与应用, 2007, 43(17): 154-156.

GUO Yong^1,2，BAO Li-chun³. Characteristics of non-negative matrix factorization for feature extraction[J]. Computer Engineering and Applications, 2007, 43(17): 154-156.