摘要
为了提高人眼检测算法的鲁棒性和实时性,提出了一种层级化的人眼检测方法。首先,该方法通过融合人脸检测器和肤色模型在图像大范围内进行快速人脸检测。然后,根据人脸面部拓扑结构几何关系的先验知识粗定位人眼得到人眼的粗定位区域。最后通过融合人眼检测器和人眼模板在人眼粗定位区域实现人眼的精确定位。大规模比较实验表明所提出方法能够准确定位人眼,并能够适用于人脸大小和旋转的变化,具有更强的鲁棒性。
This paper proposes a hierarchical method of human eye detection to improve robustness and speed.First,the method detects the face in current frame by fusing the decision of both the face detector and the skin model.Based on the prior knowledge of topological structure of human face,human eyes can be roughly localized.Finally,human eyes can be localized in a finer level by merging the decision of both eye detector and eye template.Extensive experimental results on several public face datasets.The experiment demonstrates that the proposed method can accurately localize human eves and has high robustness during the scale changing or the rotational variations.
出处
《电子测量技术》
2013年第11期39-42,共4页
Electronic Measurement Technology
基金
国家自然科学基金(61100124
61202168
61170239)
天津市应用基础与前沿技术研究计划项目(10JCYBJC25500)
2010和2011天津大学自主创新基金项目