Eye localization using a multi-scale Gabor feature vector is provided to acquire similarity to each Gabor jet registered in an eye model bunch and determine a position with the most similarity as eye coordinates, thereby improving the performance of the eye coordinates without increasing time required to detect the eye coordinates. An eye localization method comprises a model generation process and a detection process. The model generation process comprises the following steps of: detecting a facial region; normalizing the size of the detected facial region; normalizing the position of a detected face; generating multi-scale images; and generating an eye localization model based on a multi-scale Gabor feature vector. The detection process comprises the following steps of: calculating Gabor jet at the initial point of right and left eye coordinates on a normalized new facial image; calculating Gabor jet at each position of the neighborhood including an initial position; acquiring similarity to each Gabor jet registered in an eye model bunch; and determining a position with the most similarity as the eye coordinates.
展开▼