A labeling scheme based on Markov Random Fields and Gaussian mixture models for hyperspectral images

A new method about surface feature labeling for hyperspectral images is presented in this paper in the framework of Bayesian labeling based on Markov Random Field (MRF). After the dimension of the hyperspectral image is reduced by PCA, a kernel density estimator and a Gaussian mixture model (GMM) are respectively used to capture the non-Gaussian statistics of the dimension-reduced images and their difference images. Further more, one of components of GMM is chosen to describe the energy of difference images to improve classification accuracy. A Markov random field-maximum a posteriori estimation problem is formulated and the final labels are obtained by the simulated annealing algorithm. Additionally, the labeling result based on GMM is compared with Generalized Laplacian (GL) model. Experimental results show that it is an efficient and robust algorithm for surface feature labeling.