非学习模式下的类Haar特征快速人眼定位

Fast eye localization without learning using Haar-like feature

中文关键词  人脸检测  人脸识别  眼睛定位  类Haar特征

英文关键词  face detection  face recognition  eye localization  Haar-like feature

中文摘要

      本文研究非学习模式下的快速人眼定位算法. 首先, 在已检出的人脸区域中, 根据人脸几何特征的先验信 息, 设置一定的人眼候选区域, 通过高斯差分滤波消除光照影响. 其次, 定义一种类Haar的特征作用于二值化后的图 像, 在人眼候选区域计算该特征的评价值获得精确的人眼位置, 实现人眼的快速搜索定位. 该算法的处理方法简单 快速, 并且对于眼镜、眉毛以及发髻的干扰都有一定的鲁棒性. 算法通过人脸几何先验知识, 减少了训练和学习过 程. 实验结果表明, 该算法能够快速准确实时地完成眼睛的定位, 为后续的人脸识别提供必要的前提条件.

英文摘要

      This paper investigates the fast eye-localization algorithm without learning. First, in the detected face region, we setup a certain eye candidate region according to the prior knowledge of the face geometry, and reduce the illumination on it by using Gaussian difference filter. Next, we define a Haar-like feature to realize the fast eye localization on binary image, and calculate the evaluation value for the eye candidate region to determine the accurate eye location. This method we proposed is simple in computation and robust to the disturbances from eye glass frame, eye brow and hair. Because this algorithm makes use of the prior knowledge of the face geometry, training and learning processes are thus reduced. Experiments demonstrate that our method can localize eyes fast and accurately, providing the necessary condition for the face recognition followed.