Multi-Spectral Data Fusion Using a Markov Random Field Model : Application to Satellite Image Classification

I n this paper, we present a method of classifying multi-spectral satellite images. Data fusion of the multi-spectral images is achieved using a Markov random field approach. Classification is expressed as an energy minimization, problem and solved using Simulated Annealing with the Gibbs Sampler fo r label updating. The results of two digerent methods of class training, supervised and unsupervised, are shown. The proposed fusion method improved the results over those with only a single input channel.