Gaze direction estimation by component separation for recognition of Eye Accessing Cues

This paper investigates the recognition of the Eye Accessing Cues used in the Neuro-Linguistic Programming as a method for inferring one’s thinking mechanisms, since the meaning of non-visual gaze directions may be directly related to the internal mental processes. The direction of gaze is identified by separating the components of the eye (i.e., iris, sclera and surrounding skin) followed by retrieving the relative position of the iris within the eye bounding box that was previously extracted from an eye landmarks localizer. The eye cues are retrieved via a logistic classifier from features that describe the typical regions within the eye bounding box. The simultaneous investigation of both eyes, as well as the eye tracking over consecutive frames, are shown to increase the overall performance. The here proposed solution is tested on four databases proving to have superior performance when compared in terms of recognition rate with methods relying on state of the art algorithms.