Research On The Algorithm Of Road Extraction From High-resolution Remote Sensing Images

Researchers can extract many targets they interested at from remote sensing images because it carries a wealth of information. Road as the basic geographic information, is important for urban planning, cartography, car navigation, and the importance makes the road extraction from remote sensing images becoming a research hotspot. Since the seventies of last century, experts and scholars from domestic and foreign have done a lot of research on road extraction and proposed many valuable methods. But most of the achievements have some limitations because of the complexity and diversity of the feature on the images. The appearance of high-resolution images brings the road extraction an opportunity of the further development. In this article, the research is mainly about the road extraction from high-resolution remote sensing images.The main content of this article is:(1) According to the characteristics of the binary mathematical morphology erosion and dilation can eliminate small area and fill holes in the road; shape features can distinguish between non-road information and road information, this paper use mathematical morphology and the shape features of the road to extract the road.(2) The computational expense and memory required by traditional Hough transform are Large, so an improved method based on sampling window is proposed. The author uses the new method to extract the road and the experimental result shows that the new method requires less time.(3) Different objects may have the same spectrum in remote sensing images which affect the clustering result. The result of traditional FCM method is different from the real surface and will affect the road extraction. In this paper, the author use the shape feature of road as a factor to weighted the image pixel value, the contrast between the road and building is increased and a better clustering result is got. So the road can be extracted better. Meanwhile through the image histogram analysis, the selection of initial cluster centers of FCM is improved and the time is short.