基于Fisher判别分析的增量式非负矩阵分解算法

摘要

增量式非负矩阵分解算法是基于子空间降维技术的无监督增量学习方法.文中将Fisher判别分析思想引入增量式非负矩阵分解中,提出基于Fisher判别分析的增量式非负矩阵分解算法.首先,利用初始样本训练的先验信息,通过索引矩阵对新增系数矩阵进行初始化赋值.然后,将增量式非负矩阵分解算法的目标函数改进为批量式的增量学习算法,在此基础上施加类间散度最大和类内散度最小的约束.最后,采用乘性迭代的方法计算分解后的因子矩阵.在ORL、Yale B和PIE等3个不同规模人脸数据库上的实验验证文中算法的有效性.%Incremental non-negative matrix factorization is an unsupervised learning algorithm based on subspace dimensionality reduction technology. In this paper, the idea of fisher discriminant analysis is introduced into incremental non-negative matrix factorization, and an incremental learning algorithm of non-negative matrix factorization with discriminative information and constraints is proposed. Firstly, prior information of original training samples is utilized to initialize the incremental coefficient matrix through an index matrix. Secondly, the object function of incremental non-negative matrix factorization is improved to be a batch-incremental learning algorithm with the constraints of maximizing between-class scatter and minimizing within-class scatter. Finally, the factor matrices are calculated by the method of multiplicative iteration. Experimental results on ORL, Yale B and PIE face databases show the effectiveness of the proposed method.