Computational versus psychophysical bottom-up image saliency: A comparative evaluation study