Building extraction of urban area from high resolution remotely sensed panchromatic data of urban area

Abstract

With the recent availability of commercial high resolution remote sensing panchromatic imagery from sensors such as IKONOS and QUICKBIRD, it is possible to extract individual objects such as buildings from the imagery. However, traditional extraction methods cannot get the desired accuracy, because knowledge is not utilized. In this paper, we put forward a texture-based approach to get building information from the panchromatic imagery. Firstly, the image is segmented based on texture of variogram feature. Building corner structure knowledge is also combined to detect and connect building edges. Then we fill interiors of buildings through seed filling algorithm. In the final stage, point noises and linear noises are eliminated from the imagery through area or shape index value. The accuracy assessment adopted in this paper is GIS overlay analysis, which shows that 93.9% of building information is extracted correctly. The result indicates that the approach supplies another new technique for interpreting high spatial resolution remotely sensed imagery.