Extracting road from high-resolution satellite images with the combination of automatic and semi-automatic methods

In this paper, we propose strategies for road extraction from high-resolution satellite images, which combine automatic and semi-automatic methods. It performs road extraction according to following steps: firstly, extracting road automatically in Region Defined by User (RDU); secondly, if the result of automatic extraction is acceptable, repeating the first step in another RDU, if not, performing semi-automatic road extraction; thirdly, after automatic or semi-automatic road extraction, correcting inaccurate and incomplete road lines if those exist; finally, a complete road network is constructed. The program, which is conducted based on the abovementioned strategies, has been tested by SPOT images in China. The experimental results show that it performs more reliably than automatic road extraction and more efficiently than semi-automatic road extraction. ? 2005 IEEE.EI