Road extraction from high-spatial-resolution remotely sensed imagery by combining multi-profile analysis and extended Snakes model

Abstract

Extracting linear features, for example, roads, from remotely sensed imagery is an important topic in the field of remote-sensing information extraction. In this study, we propose a new road-extraction method based on multi-profile analysis and an extended Snakes model. Firstly, a gradient-based road-profile model is proposed. The method then iteratively carries out the searching of fitted profiles along many directions to form an initial road. Since the road may deviate from the actual road position due to the influence of noisy pixels, an extended Snakes model is used for positional optimization. Several experiments under different circumstances, including rural areas, suburbs and urban areas, were carried out for method validation. This shows that our method gives satisfying results in rural areas and suburbs with high identification and positional accuracy. However, in complex urban areas, its identification accuracy declines due to serious background disturbance to the profile model.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (No. 40871189), the Chinese National Programs for High Technology Research and Development (No. 2007AA12Z224, 2009AA12Z148) and the Qin Lan Project of Jiangsu.