A novel eye location method based on information fusion

Automatic facial feature location is an important step in human-computer interaction and automatic face recognition. In spite of the considerable past research effort, precise and robust facial feature location remains a challenging problem. In this paper, we deal with the eye location problem from the perspective of information fusion and propose a novel precise eye location method. Several different eye detectors, instead of one, are trained using Gentle AdaBoost. Their detection results in face region are fused using voting method. A response image is devised for practical implementation of the fusion scheme. Then we locate eye centers directly from the response image. The experimental results demonstrate that our method achieves good location performance comparable to other newly proposed methods, while our method is faster than them.