Top-to-down segment process based urban road extraction from high-resolution remote sensing image

Abstract

Aiming at improving the quality and efficiency of urban road extraction from High-Resolution Remote Sensing Image (HRRSI), this paper focuses on the top-to-down modeling framework which is constructed from network level to pixel level, and further proposes the segment process based road extraction algorithm. Firstly, segment process is defined to model centerlines of road. Based on that, network structure model is constructed according to the network-topology feature of urban road system. Secondly, the local geometry models characterizing road area and non-road area are constructed by marking network structure (characterized by segment process) with rectangles and buffers, respectively. Then according to the mixed spectral feature of pixels covered by road area and non-road area in HRRSI, mixed spectral measurement constraining model is constructed by using Gaussian mixture model and Kullback-Leibler divergence. Finally, Reversible Jump Markov Chain Monte Carlo (RJMCMC) simulation algorithm including birth, death and transformation transfer operations is designed to solve the parameters of road extraction model constructed by combining the above models. Different HRRSIs (resolution: <1 m) are tested and different comparing algorithms are experimented to illustrate advantages of the proposed algorithm. By analyzing and comparing the testing results qualitatively and quantitatively, it proves that the proposed algorithm can extract the urban road efficiently, especially in completeness and connectivity.

Keywords

• High-resolution remote sensing image (HRRSI)
• Reversible jump markov chain monte carlo (RJMCMC)
• Segment process
• Urban road extraction