Automatic Road Extraction from Remote Sensing Images Based on Rectangle Marked Point Process

Abstract

Aiming at solving inaccurate and incomplete extraction of road in remote sensing images, this paper proposes an automatic extraction algorithm based on Rectangle Marked Point Process (RMPP). First, the RMPP is designed to model the road surface, which aims to obtain the centerline and width of road correctly. Secondly, under the framework of Bayesian theory, the proposed road extraction model is built by combining network reconstruction model and spectral measurement model. The former is to constrain the relationships between rectangles according to the structure characteristic of the road. The latter is to constrain the consistency between rectangle and road body in image. Also, to obtain optimal results, related transfer kernels of RJMCMC (Reversible Jump Markov Chain Monte Carlo) based simulation algorithm are designed accordingly. Finally, testing of the proposed method and comparing methods are carried out with different remote sensing datasets. Experimental results from the proposed algorithm show that the completeness, correctness and quality can reach 98%, 94% and 92%, respectively. Compared with the results from the comparing method qualitatively and quantitatively, it can be verified that the proposed method can not only extract the high-quality road networks from different datasets but also can obtain the width of the road simultaneously.