摘要：Extracting the building rooftops from the satellite image has been an active research subject with a wide range of applications. However, most of previous approaches have some drawbacks: being sensitive to initial seed points which are defined by the operator according to his experiences; threshold selection is also finished by such interactive works; unexpected results for color image; needing extra data source. In order to solve those problems, this paper proposes an automatic approach that extracts building rooftops from a single satellite image, and furthermore makes the rooftops be of rectilinear contour in order to supply modeling data for 3-D building reconstruction. Firstly, an automatic method is proposed to obtain the initial seed points for the region growing method. Secondly, by over growing strategy aided with Canny edge map, the building areas can be departed from the non-building areas avoiding the difficulty of threshold selection. Thirdly, the rooftops are later refined and fitted by a block operation algorithm to get the rectilinear boundaries, and then the rectilinear contours are extracted by boundary tracking. It shows the quick and considerably precise results, and the tree occlusion problem is also solved in a degree.
关键词：building extraction; region growing; rectilinear fitting; satellite images
