Recent studies show that image classification techniques which use both spectral and spatial information are more suitable, effective, and robust than those that use only spectral information. Although late model support vector machines (SVMs) have been providing accurate results in the studies, this method is intrinsically non-contextual, which represents significant limitation in image classification. In this paper, we propose a rigorous framework which integrates SVMs and Markov random field models in a unique formulation for spatial contextual classification of various species of forest trees, ground vegetation, and water using polar metric synthetic aperture radar data. Genetic algorithm is employed for selecting appropriate features and automatic estimation of optimal parameters. Comparison of the accuracy of the proposed method with baseline methods was performed. Comparison of the accuracy of the proposed method with some other methods was carried out. The results show that this algorithm allowed approximately 19%, 14%, 11%, 5% and 3% increase in overall accuracy with respect to the Wishart, WMRF, SVM, aMRF and MSVC methods, respectively.