Visual saliency based automatic road extraction from high-resolution multispectral satellite images

Abstract

With the development of remote sensing satellites, radar imaging, and unmanned aircraft, how to utilize the modern image processing technology to extract road network has been a hot research topic of remote sensing and geographic information technology. Therefore, a visual saliency based automatic road extraction method from high-resolution multispectral satellite images is proposed in this paper. Firstly, the color, intensity and orientation features are extracted to construct visual attention model for road saliency region detection. Then road network is extracted by adaptive region growing algorithm after automatically setting the pixels with larger saliency as seed points. At last, the non-road parts and noises are removed from road network by using morphological operations. The experimental results show that the proposed automatic road extraction method can achieve a superior performance in completeness and correctness than other traditional methods.