Building extraction from high resolution satellite imagery based on multi-scale image segmentation and model matching

In this paper, we established a new general semiautomatic building rooftop extraction method applied for high resolution satellite imagery. Based on investigation of the current existed methods for building extraction and its feature extraction, a general framework of building rooftop extraction is proposed. To extract the precise building roof boundary, an seeded region growth segmentation or localized multi-scale object oriented segmentation is applied to extract small and simple rectilinear rooftops from its background; to delineate the precise position of complex rooftop, the pose clustering is applied for building locating, and model matching techniques based on node graph search is used for finding the correct building rooftop shape. Integration of these two methods makes extraction of buildings from simple rectangle rooftop to complicated building more practical. Preliminary experimental results on QuickBird imagery show that the proposed method can successfully extract about 75% of the regular building rooftops.