An Adaptive Multifeature Method for Semiautomatic Road Extraction From High-Resolution Stereo Mapping Satellite Images

In this letter, an adaptive multifeature method for semiautomatic road extraction in high-resolution stereo mapping satellite images is proposed. First, the digital surface model (DSM) is generated from high-resolution stereo mapping satellite images by the semiglobal vertical line locus matching method. To combine the entropy feature and the spectral feature with the DSM, an adaptive method based on the image matching theory is introduced, which maintains a dynamic balance in the road extraction process and ensures the complementarity of each feature. Then, the geodesic road tracking method integrated with kernel density estimation and mean shift is used to extract roads from the feature fusion map. Experimental results have shown that the proposed method can extract roads both smoothly and correctly from high-resolution stereo mapping satellite images and especially performs better than other existing methods when there is an elevation difference between the road target and its neighborhood.