基于改进GIPF和形态学滤波的快速人眼定位方法

摘要

针对传统积分投影方法易受眉毛、睫毛、阴影、遮挡及噪声等干扰的问题，提出了一种改进梯度积分投影（GIPF）和形态学滤波相结合的快速人眼定位算法。该算法首先采用二维梯度算子计算人脸图像的行列积分投影，完成人眼粗定位，减少眉毛和光照等的干扰；其次，利用形态学滤波平滑图像边界的特点，准确分离出瞳孔区域；最后采用不规则矩求质心的方法精确定位人眼。该方法在耶鲁大学的YaleB图像库、ORL及日本表情库（JAFFE）上表现优异。实验结果表明，本方法不易受眉毛及噪声干扰，对不同光照条件及头部姿态都有良好的鲁棒性，并且运算速度更快。%To overcome the effects of eyebrows,eyelashs,shadows,occlusion and noises in traditional integral projection methods,this paper presents a fast algorithm for eye localization combined an improved gradient projection function (GIPF) method with morphological filtering. First of all, it uses a two-dimensional gradient operator to calculate the integral projection of a human face to locate the human eyes roughly. Secondly, in order to separate the pupil area accurately, morphological image filtering algorithm is used to get a smooth boundary. Finally, the precise position of human eyes can be obtained by centroid calculation with irregular shape image moments. We have compared this method with the Japanese expression database (JAFFE), the YaleB database and the ORL database. This method is much faster for eye localization, without influences by the eyebrows, furthermore, it shows good robustness to different light conditions and head poses as well as other noise interferences.