Convergence analysis of a class of non-negative matrix factorization methods for general graph regularization with applications

Abstract

In this paper, based on graph regularized non-negative matrix factorization (GNMF), we introduce and study a class of non-negative matrix factorization methods for general graph regularization (GGNMF), which includes GNMF and non-negative matrix factorization (NMF) as special cases. Furthermore, we analyze convergence of iterative sequences generated by GGNMF. As applications, we dispose unmixing of hyperspec-tral image via using NMF, and establish hyperspectral image of the original remote sensing satellite image decomposed by Jasper ridge and Hubble. Finally, we present some work for our future research.