Semi-automatic building extraction from high resolution imagery based on segmentation

Recently, more and more high resolution remote sensing images appear, and they provide new data source for building extraction. Some building extraction methods are proposed to adapt to this trend. This paper addresses a semi-automatic method that combines segmentation and region selection. First, mean shift segmentation is applied to the image, and then the region extraction is implemented through the interactively selection of building parts. Edge detection is also involved to get the boundary of buildings. At last, the paper gives the contrast of the unsupervised classification ISODATA result and this extraction method.