Emergency response ought to be rapid, reliable and efficient in terms of bringing the necessary help to sites where it is actually needed. Although the remote sensing techniques require minimum fieldwork and allow for continuous coverage, the established approaches rely on a vast manual work and visual assessment thus are time-consuming and imprecise. Automated processes with little possible interaction are in demand. This paper attempts to address the aforementioned issues by employing an unsupervised classification approach to identify building areas affected by an earthquake event. The classification task is formulated in the Markov Random Fields (MRF) framework and only post-event airborne high-resolution images serve as the input. The generated photogrammetric Digital Surface Model (DSM) and a true orthophoto provide height and spectral information to characterize the urban scene through a set of features. The classification proceeds in two phases, one for distinguishing the buildings out of an urban context (urban classification), and the other for identifying the damaged structures (building classification). The algorithms are evaluated on a dataset consisting of aerial images (7 cm GSD) taken after the Emilia-Romagna (Italy) earthquake in 2012.