Land Cover Classification of Very High Resolution Image using Graph-Cuts

Land cover classification from very high resolution (VHR) image (e.g. QuickBird, IKONOS) is an extremely challenging task due to high spatial heterogeneity and low spectral separability. Markov random field (MRF) based classification methods have proven to be effective for solving the above problem by integrating contextual information into classification process. Existing energy minimization approaches for the MRF-based image classification model cannot balance classification accuracy and computational efficiency. Focusing on efficient energy minimization of the MRF-based image classification model, we apply two graph cuts algorithms, called the swap move algorithm and the expansion move algorithm to classify land cover using VHR image. The proposed methods turn out to be fast, simple, and robust. A set of experimental results of QuickBird images demonstrate that the proposed methods can obtain the land cover classification results similar with manual interpretation with homogeneity interior and smooth boundaries.