Visual attention on human face

Abstract

Human faces are always the focus of visual attention since faces can provide plenty of information. Although some visual attention models incorporating face cues work better in scenes containing faces, no visual attention model is particularly designed for faces. On faces, many high-level factors will influence visual attention distribution. In practice, there are many visual communication systems in which faces occupy the scenes, such as video calls. Specific visual attention model designed for face images will be of great value in these circumstances. In this paper, we conduct research on visual attention analysis and modelling on human faces. To facilitate this research, we collect 120 face images and perform eye-tracking experiments with these images. Eye-movement data shows that detailed visual attention allocation exists on faces. Using face detection and facial landmark localization, we find that some facial features are highly effective for visual attention prediction. The performance of many visual attention models can be improved by incorporating those facial features.