Background and Aim:
Visual attention is a high order cognitive process of human brain which defines where a human observer attends. Dynamical computational visual attention models mimic the human brain behavior in predicting the focus of attention when the observer watches a video. Several computational models have been proposed to provide a better understanding of dynamical visual saliency maps. In this study, we adopt different computational visual attention models to evaluate the multimedia designed for English learning. Methods: In this study, we use a designed multimedia and collect the eye-tracking data of English learners and use them as our ground truth. We select four computational visual attention models explained below and compare their results with the human behavior. The well-known model of Itti et al. (۱۹۹۸) uses low level visual features (i.e., color, intensity and orientation) and combines the corresponding saliency maps. Harle et al. have introduced a button-up graph-based visual saliency model (۲۰۰۶), which consists of two steps of forming activation maps on certain feature channels, and normalizing them in a way which highlights conspicuity and admits combination with other maps. Hu and Zhang (۲۰۰۷) proposed a simple method which uses the spectral residual of an image to reach the saliency of an image or frames of a video. Wong et al. (۲۰۱۸) developed a framework that uses visual attention models to extract salient objects in a video. Results: The use of multimedia for education is widespread today. One of the most important goals of
educational multimedia design is to attract the attention of the audience and to convey the educational concepts appropriately. For this reason, it is important to be able to predict the audience's attention areas and provide appropriate educational material in those areas. In this study, we apply the computational visual attention models to observe how well they can mimic the human behavior during watching an English educational multimedia. Hence, they can be used as the tools for the evaluation of any designed educational multimedia. Conclusion: The evaluation of above mentioned models based on NSS, CC and ROC criteria show that the GBVS model and Wong et al model can better mimic the human behavior. Multimedia designers can apply these models to evaluate their designed multimedia. Moreover, it can help them to modify and improve the
educational multimedia such that it attracts the focus of visual attention to the regions where the designer is interested in.
