Road extraction from high-resolution remote sensing images based on multiple information fusion

Road extraction from high-resolution remote sensing images has been considered to be a significant but very difficult task. Especially the spectrum of some buildings is similar with that of roads, which makes the surfaces being connect with each other after classification and difficult to be distinguished. Based on the cooperation between road surfaces and edges, this paper presents an approach to purify roads from high-resolution remote sensing images. Firstly, we try to improve the extraction accuracy of road surfaces and edges respectively. The logic cooperation between these two binary images is used to separate road and non-road objects. Then the road objects are confirmed by the cooperation between surfaces and edges. And the effective shape indices (e. g. polar moment of inertia and narrow extent index) are applied to eliminate non-road objects. So the road information is refined. The experiments indicate that the proposed approach is efficient for eliminating non-road information and extracting road information from high-resolution remote sensing image.