Regularized Nonnegative Matrix Factorization with Real Data for Hyperspectral Unmixing

Nonnegative matrix factorization has been applied in hyperspectral unmixing, while the accuracy of unmixing is closely related with the local minimizers. In this paper, we present a new regularized cost function of nonnegative matrix factorization by fully considering the real data information of the endmember signature. The endmember signatures can be easily found in the United States Geological Survey spectral library. The multiplicative update rules are employed to obtain the factor matrices, because it is easy to implement and often yields good results. We demonstrate the success of regularized nonnegative matrix factorization by applying it on hyperspectral unmixing.