School of Science, Northeastern University, Shenyang, China

---

Abstract—Eye location is one fundamental but very important problem for face recognition. In this paper, we proposed a new eye location method based on Adaboost and Random Forests. The proposed method consists of three main steps. Firstly, we apply Haar features and Adaboost algorithm to extract the eye regions from a face image. Secondly, we highlight the characteristics of eyes by Gabor filter, then segment the pupil from the eye regions based on intensity information. We compute the coordinate of center of the pupil as the position of the eye. Lastly, the eye location result is judged and adjusted by symmetry-axis of the face and Random Forest. Compared with the existing eye location approaches, the proposed method use the symmetry-axis of the face and Random Forests to judge and adjust the eye location result, which enhance the accuracy of eye location remarkably. The proposed method has been tested in the CAS-PEAL-R1 database and CASIA NIR database respectively, the simulation results demonstrate that the location accuracy rate is 98.86% and 97.68% respectively.

Index Terms—Adaboost, regional features, eye location, Haar feature

[PDF]

Cite: Xiangde Zhang, Qingsong Tang, Hua Jin, Yue Qiu, and Yang Guo, "Eye Location Based on Adaboost and Random Forests," Journal of Software vol. 7, no. 10, pp. 2365-2371, 2012.