Abstract: Aiming at the problems such as low accuracy and poor generalization ability of the existing human eye location algorithms under near infrared light, a human eye location algorithm based on Histogram Oriented Gradient（HOG） and Support Vector Machine（SVM） is proposed. HOG is used to extract the human eye features of iris images and the HOG features are trained by SVM classifier to locate human eyes. In order to further improve the accuracy and reduce the missing detection and false detection, a multi-level cascade SVM classifier algorithm is proposed. In addition, aiming at the unique grayscale distribution characteristics of iris images under near infrared light, an image preprocessing method is designed, which can significantly improve the positioning speed. The experimental results on MIR2016 and CASIA-IRIS-Distance dataset show that the human eye location algorithm based on HOG and SVM has high accuracy, strong generalization ability and high real-time performance.

Key words: iris recognition, eye location, Histogram Oriented Gradient（HOG）, cascade Support Vector Machine（SVM）, image preprocessing

摘要： 针对近红外光下现有的人眼定位算法普遍存在准确性不高、泛化能力不佳等问题，提出了一种基于方向梯度直方图（HOG）和支持向量机（SVM）相结合的双眼虹膜图像的人眼定位算法。利用HOG提取虹膜图像的人眼特征，并结合SVM分类器对HOG特征进行训练从而实现人眼的精确定位。为了减少漏检和误检，进一步提高定位准确率，又提出了多级级联SVM分类器算法；另外针对近红外光线下虹膜图像独特的灰度分布特点，设计了一种图像预处理方法，能够显著提高人眼定位速度。在MIR2016和CASIA-IRIS-Distance数据集上的实验结果表明，基于HOG和SVM的双眼虹膜图像的人眼定位算法具有高准确率、强泛化能力和高实时性。

关键词: 虹膜识别, 人眼定位, 方向梯度直方图（HOG）, 级联支持向量机（SVM）分类器, 图像预处理