Text this: A saliency-based bottom-up visual attention model for dynamicscenes analysis