Layout-Driven Top-Down Saliency Detection for Webpage

Webpage is a significant part of internet content and a pervasive medium for information communication. Studying the distribution properties of eye fixation for webpage can help improve the quality of webpage design. In this paper, we investigate the web-viewing behavior caused by webpage layout and propose a novel webpage saliency method based on this top-down knowledge. We derive a layout saliency map by probability model as top-down knowledge according to different layout styles and incorporate it with the bottom-up saliency map computed by conventional models. Experimental results demonstrate that with the proposed layout-driven top-down saliency, webpage saliency detection can be improved for almost all the conventional saliency models.