This article presents an efficient approach for accurate registration of a building facade model "dressed" with dense semantic information. Localization sensors such as the GPS as well as vision-based methods are able to provide a camera pose in an efficient and stable way, but at the expense of low accuracy. We propose here to rely on semantic maps to improve the accuracy of a rough camera pose. Simultaneously we aim to iteratively improve the quality of the semantic map through the registration. Registration and semantic segmentation are jointly refined in an Expectation-Maximization framework. We especially introduce a Bayesian model that uses prior semantic segmentation as well as geometric structure of the facade reference modeled by Generalized Gaussian Mixtures. We show the advantages of our method in terms of robustness to clutter and change of illumination on urban images from various databases.