EOH-Based Classifier for the Interpretation of Urban Images

We propose an algorithm for recognition of buildings and interpretation of outdoor urban scenes. Our method deals with diversity of shapes and appearances of buildings and their parts by considering local histograms of edge orientations. Local image patches are mapped onto classes man-made structure, vegetation or sky. The proposed method provides thus interpretation of an urban scene by partitioning the image into three semantically meaningful types of regions. In the context of recognition of buildings and its parts, our approach can also be viewed as initial focus of attention.