An Optimal Road Seed Extraction Algorithm

The proper extraction of road seeds is the premier step of road network extraction from high resolution remote sensing images. An optimal road seed extraction algorithm is proposed. Firstly, Canny-Deriche edge detection and spatial FCM (fuzzy c means) region extraction are performed separately to detect the details.Secondly, an averaged Hausdorff distance is introduced to evaluate the difference between the results of the two methods. Thirdly, in order to take advantages of both edge detection and region extraction, some constraints are loosen and each kind of information is corrected and complemented by the other through iterations, until the results are consistent and unnecessary details are eliminated. Finally, road seeds are extracted through iterative Hough Transform and grouping. The proposed algorithm is tested on high resolution Ikonos images and the results are proved to be favorable.