Saliency estimation using a non-parametric low-level vision model