Road extraction from high-resolution remote sensing images based on spectral and shape features

Road extraction from high-resolution remote sensing images has always been a significant but very difficult task. In this paper, a new approach integrating spectral and shape features into the object extraction process is proposed. The method mainly consists of three modules: spectral classification, morphological segmentation and shape feature recognition. The principal work is to refine the road network using three shape indices (distribution density, eccentricity and precision), which can remove the spectrally similar non-road objects. Experimental results demonstrate that this method is efficient to extract the central road network.