NSCT based road extraction from remote sensing image using directional prior

Hua Zhong, Dan Zhu, Licheng Jiao, Kun Liu

Abstract: Road extraction has become an important topic in remote sensing image processing. As a linear target, road has a simple geometric appearance and can be well represented by the Nonsubsampled Contourlet Transform (NSCT). This paper proposed a new road extraction method, which exploits the directional prior of the road in NSCT domain. Road prior includes not only the directional features, but also the context of intra-scale and inter-scale. The proposed method consists of three parts. The first two parts are context based including direction correction and direction enhancement; and the last part extracts road through non maximum suppression. Simulation results demonstrate that the proposed method shows an obvious improvement as compared with the method without the directional prior.