Unsupervised line network extraction from remotely sensed images by polyline process

This article presents a new stochastic geometric model for unsupervised extraction of line network (roads, rivers,...) from remotely sensed images. The line network in the observed scene is modeled by a polyline process, named CAROLINE. The prior model incorporates the topological properties of the network considered through potentials on the polyline shape and interactions between polylines. Data properties are taken into account through a data term based on statistical tests. Optimization is realized by simulated annealing using a RJMCMC algorithm. Some experimental results are provided on aerial and satellite images.