Automated road network extraction from high spatial resolution multi-spectral imagery

Abstract

In this paper, a new approach to road network extraction from multi-spectral (MS) imagery is presented. The proposed approach begins with an image segmentation using a spectral clustering algorithm. This step focuses on the exploitation of the spectral information for feature extraction. The road cluster(s) is automatically identified using a fuzzy classifier based on a set of predefined membership functions for road surfaces and the corresponding normalized digital numbers in each multi-spectral band. A number of shape descriptors from the refined Angular Texture Signature are defined and used to reduce the misclassifications between roads and other spectrally similar objects such as parking lots, buildings or crop fields. An iterative and localized Radon transform is then performed on the classified and refined road pixels to extract road centerline segments. The detected road segments are further grouped to form the final road network, which is evaluated against a reference dataset. Our experiments on Ikonos MS, Quickbird MS, and color aerial imagery show that the proposed approach is effective in automating road network extraction from high resolution multi-spectral imagery. Results from two different evaluation schemes also indicated that the proposed approach has achieves a performance comparable to other methods.