Automatic road extraction from high-resolution images applied over urban areas

Road extraction plays an important role in many applications such as car navigation. However, the manual extraction of roads is a laborious and tedious task. Road extraction from satellite images has drawn considerable attention in last years due to the recent availability of commercial high-resolution optical satellite imagery. Road extraction strategies are usually classified into two categories according to the degree of human interaction: semi-automated and automated extraction. Different strategies for road extraction require specific algorithms. For semi-automated, initial seed points, sometimes with directions, must be provided to an algorithm that attempts to connect these points using various search path criteria. For automated extraction, salient roads, called road seeds, must be detected automatically and tracked or linked to form the road network. These approaches involve many techniques such as template matching [1,2,3], heuristic reasoning [4,5], dynamic programming[6,7] or stochastic tracking[8,9]. In this paper, we propose a new method for automatic road extraction from high-resolution images applied to urban areas (fig1). In order to deal with the high complexity of this kind of scenes, we integrate detailed knowledge about roads and their context using explicitly formulated models. The knowledge about how and when certain parts of the road and context model are optimally exploited is expressed by an extraction strategy. Scale space and Edge-detection techniques are used as preprocessing for segmentation and estimation of the road width. The detection of road is based on the Energy minimization techniques. The estimation of the Energy depends on many parameters including variance, direction, length and width of the road in consideration. The use of width and variance information for road extraction expects high-resolution images. The contribution of this paper consists of Energy minimization and path following by using at first pre-processing through scale-space and canny edge detector adapted to road extraction. The paper also handles tracing of road junctions.