A Markov random field integrating spectral dissimilarity and class co-occurrence dependency for remote sensing image classification optimization

This paper develops a novel Markov Random Field (MRF) model for edge-preserving spatial regularization of classification maps. MRF methods based on the uniform smoothness lead to oversmoothed solutions. In contrast, MRF methods which take care of local spectral or gradient discontinuities, lead to unexpected object particles around boundaries. To solve these key problems, our developed MRF method first establishes a spatial energy function integrating local spectral dissimilarity to smooth the initial classification map while preserving object boundaries. Second, a new anisotropic spatial energy function integrating the class co-occurrence dependency is constructed to regularize pixels around object boundaries. The effectiveness of the method is tested using a series of remote sensing data sets. The obtained results indicate that the method can avoid oversmoothing and significantly improve the classification accuracy with regards to traditional MRF classification models and some other state-of-the-art methods.