A road centerline extraction algorithm from high resolution satellite imagery

Traditional road extraction methods rarely consider the shape features and thus the extracted centerline is not smooth. To overcome the aforementioned limitations, a new road centerline extraction algorithm from high resolution satellite imagery using shape features and multivariate adaptive regression splines (MARS) is proposed in this paper. The method has three main steps. Firstly, an improved shape feature is developed to filter the segmentation result to extract linear features. The spectral information is then used to extract pure road segments from the linear features. Finally, road centerline is extracted using MARS. The proposed method is programmed by Matlab and tested on three high resolution imageries. The experimental results indicate that the improved shape feature can effectively extract linear features and the smoothness of road centerlines extracted by MARS is higher than that of traditional methods.