High level Bayesian analysis of spatial and image analysis requires sophisticated prior models from stochastic geometry. Due to the high complexity of such models inference has to rely on MCMC methods. However, the efficiency of such algorithms is often a bottleneck. MCMC is particularly complex in this setting due to huge data sets which exhibit intricate spatial dependence structures and lead to multi-modal posterior models. In this project we examine methods based on spatial aggregration and multireolution approaches that may improve the mixing for models in stochastic geometry. This project is funded to an EPSRC first grant scheme.