Rooftop detection using a corner-leaping based contour propagation model

Extracting building rooftops in satellite/aerial images is one of the most challenging problems in the application of computer vision for remote sensing. In this paper a new contour propagation model for rooftop boundary detection is proposed. It includes developing contour models that evolve by leaping on image corners and edge points while minimizing an energy function based on image corner responses, image color invariants and edge points. The proposed method is capable for coping with the complications associated with the gabled rooftop using Gaussian color invariance modeling. Experimental results for aerial/satellite images show that the average shape accuracy is above 90% for the test images of sub-urban areas.