A variational Bayes algorithm for joint-sparse abundance estimation

Paris V. Giampouras, Konstantinos E. Themelis, Athanasios A. Rontogiannis, Konstantinos D. Koutroumbas

Abstract: This paper presents a variational Bayesian scheme for semi-supervised unmixing on hyperspectral images that exploits the inherent spatial correlation between neighboring pixels. More specifically, a hierarchical Bayesian model that promotes a joint-sparse profile on the abundance vectors of adjacent pixels is developed and a computationally efficient variational Bayes algorithm is incorporated to perform Bayesian inference. The benefits of the proposed joint-sparse model are demonstrated via simulations on both synthetic and real data.