Integration of Gaussian process and MRF for hyperspectral image classification

In this paper, we propose a framework GP-MRF, which combines Gaussian processes (GPs) and Markov random field (MRF) for accurate classification of hyperspectral remote sensing image (HSI) data. This method exploits the relationship among adjacent pixels and integrates it into spectral information to obtain spectral-spatial classification. This framework consists of two steps. Firstly, a GP classifier (GPC) yields pixelwise predictive probability for each class. Secondly, an MRF is applied to extract spatial contextual information in the label map achieved in the first step. Then the classification results are inferred from the spectral-spatial information. By means of MRF regularization an enhanced classification result has been obtained. The experiments are performed on three hyperspectral benchmark datasets. The results from the GPC are compared with those obtained by state-of-the-art classification approaches and demonstrate that, GP model is a competitive tool for classification of HSI in terms of accuracy. Furthermore, the experimental results indicate that our proposed method GP-MRF improves the classification accuracy of conventional GPC.