Webpage Saliency Prediction Using a Single Layer Support Vector Regressor

Websites play an important role in human lives due to virtual communication, online learning, social networks, and online business. Studying the pattern of human visual attention while navigating web pages is beneficial. Considering such information in web design helps increase users’ satisfaction in browsing and designers’ satisfaction in arranging the content. Webpage saliency detection is more challenging than finding conspicuous areas in the natural images because of items such as buttons, input boxes and menus. The natural saliency prediction methods fall short of delivering acceptable performance. In this article, we proposed an algorithm to predict a saliency map corresponding to any webpage. This method consists of two phases of low-level and high-level feature extractions and using them in a group of support vector regressors. We benefitted from two deep neural net-based algorithms to find text areas and natural saliency. Next, this information is fed to a single-layer support vector regressor to estimate the test web page saliency map. Experiments show that the output of the proposed method has the highest correlation with the corresponding ground truth compared to almost all the state-of-the-art algorithms.