Influence of number, location and size of faces on gaze in video

Abstract

Many studies have reported the preference for faces and influence of faces on gaze, most of them in static images and a few in videos. In this paper, we study the influence of faces in complex free-viewing videos, with respect to the effects of number, location and size of the faces. This knowledge could be used to enrich a face pathway in a visual saliency model. We used eye fixation data from an eye movement experiment, hand-labeled all the faces in the videos watched, and compared the labeled face regions against the eye fixations. We observed that fixations made are in proximity to, or inside the face regions. We found that 50% of the fixations landed directly on face regions that occupy less than 10% of the entire visual scene. Moreover, the fixation duration on videos with face is longer than without face, and longer than fixation duration on static images with faces. Finally, we analyzed the three influencing factors (Eccentricity, Area, Closeness) with linear regression models. For one face, the E +A combined model is slightly better than the E model and better than the A model. For two faces, the three variables (E,A,C) are tightly coupled and the E +A+C model had the highest score..