Toward a model of computational attention based on expressive behavior: applications to cultural heritage scenarios