Hyperspectral unmixing based on the sparisity of spectrum´s gabor transform coefficients

Nonnegative matrix factorization (NMF), which has shown success in blind source separation, has also been applied to hyperspectral data unmixing. Many constraints have been introduced to render better estimates. However, most algorithms in this respect have not explored the inner characteristics of the spectrum and the formation of the spectrum itself. In this paper a novel sparisity constrained nonnegative matrix factorization (SCNMF) is proposed, which investigates the sparisty characteristic of the spectrum in its gabor transform domain. Results obtained with synthetic and real data are used to illustrate the effectiveness of the proposed method.