A Semi-automatic Road Extracting Method for High Spatial Resolution Remotely Sensed Imagery

It is one of the important work to extract linear features,e.g.,roads,from remotely sensed imagery in the field of remote sensing information extraction.A semi-automatic method to extract roads from high spatial resolution remotely sensed imagery is proposed.The main steps include:1)some basic profile features,e.g.,the starting road direction,width,and radiometry distribution are obtained with the user-specified starting road seed couple;2)a searching fan is then created,within which several 'scan snakes' on several directions are dispatched,which contains themselves' several snake joints,i.e.,the scan profiles.Within each scan joint of each snake,a pair of edge points(gradient extremes of the pixel values along each side of the road)which satisfy the road profile model will be searched.For every finding within every joint of a snake,its votes will be added.The best snake is the one which carries the most votes,which then denotes the next searching direction.The searching is carried out from the starting position until reaching some finishing conditions,e.g.,the boundary of an image;3)these edge points are then connected to form a double-side road.The main road network can be extracted under a lot of complex conditions,such as distinguishing changes of road directions and radiometry distributions,road broken and intersections.Several experiments on Beijing-1 panchromatic imagery(with spatial resolution 4m)are given,which validate the adaptive ability and practicability of our method.