We present in this report a comparative study between models of line network extraction, within a stochastic geometry framework. We rely on the theory of marked point processes specified by a density with respect to the uniform Poisson process. We aim to determine which prior density is the most relevant for road network detection. The "Candy" model, introduced in [21] for the extraction of road networks, is used as a reference model. This model is based on the idea that a road network can be thought of as a realization of a Markov object process, where the objects correspond to interacting line segments. We have developed two variants of this model which use quality coefficients for interactions. The first of these two variants is a generalization of the "Candy" model and the second one is an adaptation of the "IDQ" model proposed in [13] for the problem of building extraction from digital elevation models. The optimization is achieved by a simulated annealing with a RJMCMC algorithm. The experimental results, obtained for each model on aerial or satellite images, show the interest of adding quality coefficients for interactions in the prior density.