Combination of region-based and pixel-based hyperspectral image classification using erosion technique and MRF model

Image classification plays an important role in remote sensing applications. Current paper presents a new spectral-spatial classification of hyperspectral data. This approach is based on combination of region-based and pixel-based methods. Erosion technique is used for extracting uncertain pixels from initially segmented image. These uncertain pixels are classified using pixel-based classification method. In pixel-based classification stage, Markov random field (MRF) model integrates contextual information into a classifier under a Bayesian framework. Experimental results show that this method can perform better in comparison with the conventional pixel-based MRF method and maximum likelihood (ML) classification.