A Unified information-theoretic framework for viewpoint selection and mesh saliency