Detection of buildings from high resolution satellite images using urban area knowledge

In this study, a new approach that utilizes urban area information to detect buildings from single high resolution multispectral satellite images is proposed. Unlike other approaches, the proposed approach exploits the urban area knowledge during the revision process of the shadow mask to reveal dark regions which do not belong to shadow areas. Thereafter, these regions are assigned to a different class in the graph based classification, and in this way, the rate of incorrect labeling of buildings is decreased dramatically. Proposed approach is tested for six test sites from two different sensors (Ikonos and QuickBird). The comparison of the results of our approach with two different shadow based method reveals the success of the developed approach.