Abstract: In video chat, people usually gaze at the display window of the other side instead of the camera which cause that people can’t gaze at each other’s eyes. Based on this, a method is proposed. First, a picture in which people stare at the camera is taken, and Harr-classifier is used to extract rough area of eye’s location from picture. Corners of the eye and eyelids are detected after removing the interference information in area, based on this, area of eye rectangle is extracted, which is standard template. Then, single image in video is in same treatment and area of eye rectangle is obtained. Last, the standard template replaces the area of eye rectangle. This method solves the problem of selecting threshold value successfully by introducing histogram specification into binary image processing. The experimental results show that this method has low cost, and it is effective and operationally simple, and accurate rate can reach more than 85%.

Key words: histogram specification, eye detection, replacement process