Automated thresholding for real-time image processing in video-based eye-gaze detection

This paper describes the improvement of the method for removing the eyeglass reflection and for accurately detecting the center coordinates of the pupil and corneal reflection images, which determine the eye-gaze in the video-based eye-gaze detection method. Under eyeglass wearing conditions, the reflection light in the glass, caused by the infrared light sources irradiated for the eye, disturbs the pupil and corneal reflection images. To eliminate the glass reflection image and to detect the pupil image, the following parameters were automated: the thresholds for high brightness elimination and enhancement, the thresholds for extraction of the pupil and corneal reflection, the sizes of windows for the pupil and corneal reflection detection, and the sizes of the structuring elements for the morphological operations (dilation and erosion), and so on. This paper also describes the algorithm for detecting the eye-gaze immediately after blinking.