Making computers look the way we look: exploiting visual attention for image understanding