Optimization of cluster coverage for road centre-line extraction in high resolution satellite images

This paper presents a new approach for road centre-line extraction from high-resolution images, based on cluster coverage. The idea is first to segment the image; in order to obtain road candidates. Then we apply the k-means algorithm to cluster the pixels of the segmented image. The resulting clusters cover the road and the noise generated by the segmentation process. We introduce the concept of cluster coverage to remove noise and to approximate the optimal number of clusters that fit the shape of the road.