Gaze Classification on a Mobile Device by using Deep Belief Networks

In this paper, we introduce a gaze classification method which classifies the locations of human gaze on a display of a mobile device. For example, when the user see the upper part of a display, our method classifies the gaze as the upper part among the available choices: upper, middle, and lower parts of the display. Our method uses appearance-based gaze estimation and gray-scale images captured from a camera of a mobile device. This method does not require any personal calibration. We train gaze classifiers by using Deep Belief Networks with considering head poses in general environments for a mobile device. The gaze classification method is applied to human-computer interaction and various application programs.