Automated extraction of road networks from IKONOS data in urban area

In this paper, one of knowledge-based approaches to extract road networks automatically is presented. The roads are extracted with several spectral and geometric characteristics of them. Seed areas to trace road networks are investigated by spectral and geometric information using linear-shape kernel. Then the road networks are traced by knowledge-based methods from the seed areas. Main difficulties to trace roads are occluding materials such as cars, buildings, casting shadow of buildings, and vegetation area. Vegetation area is masked using NDVI. The frame of road networks is searched, and the interrupted parts of networks are connected to build complete networks using knowledge about the global network characteristics of roads. The results of the extraction are evaluated by comparison with the road networks in the topographic map.