In this paper, we propose a support vector machines (SVMs) method of classifying image regions hierarchically based on their semantics rather than on low-level features. First, image regions are segmented using the ...