Minimum variance block-based nonnegative matrix factorization algorithm for hyperspectral unmixing

In this paper, we present a new minimum variance constrained nonnegative matrix factorization (NMF) algorithm using the block principal pivoting optimization strategy for hyperspectral unmixing, named MVBPP. The proposed approach has two advantages. One is that the BPP-based NMF algorithm has shown good convergence property and better computational speed than other NMF ones. The other is that the proposed approach does not need the existence assumption of pure pixel of each endmember in the scene. Experimental results on highly mixed synthetic data confirm the accuracy of the developed algorithm.