Extraction automatique des réseaux linéiques à partir d'images satellitaires et aériennes par processus Markov objet

This article presents a method for the unsupervised extraction of line networks (for example, road and hydrographical networks) from satellite and aerial images of medium and high resolution. We model the line network in the observed scene by a Markov object process, where the objects are interacting line segments. The prior model, the Quality Candy model, is designed to exploit as fully as possible the topological properties of the network under consideration, while the radiometrical properties of the network are modelled using a data term based on statistical tests. Two techniques are used to compute this term, one accurate and the other efficient. Optimization is performed using simulated annealing with a RJMCMC algorithm. We accelerate convergence of the algorithm by using appropriate proposition kernels. The results obtained on images coming from different sensors are quantitatively evaluated with respect to manually generated ground truth.