基于角点特征的眼睛轮廓提取

Incorporating Corner Information for Extraction of Eye Features

提出了一种新的基于可变形模板的提取眼睛轮廓的方法。在新的方法中,利用了眼睛所有角点的信息来构造能量函数。同时,采用了SUSAN(smallest univalue segment assimilating nucleus)中的角点强度函数来描述角点的特性。通过结合角点的信息,眼睛轮廓能够得到更为鲁棒地提取。考虑到角点能量函数的局部特性,采用了三步优化的方法来对能量函数进行优化。为了测试新的能量函数的表现,随机设置眼睛模板的初始参数以模拟各种预处理方法得到的结果。实验结果表明,该方法能比以前的方法更精确地提取出眼睛的轮廓。

A new eye feature extraction method incorporating comer information with deformable template is proposed in this paper. In the energy function, the comer information of all comers in the eye template is used. The corner strength being used in SUSAN （ smallest univalue segment assimilating nucleus ） is introduced to define the comer energy term. Incorporating with comer information, the eye template is located more robustly. Considering the local property of eye comer energy term, three steps optimization process is adopted to deform the eye template. To test the robustness of the new energy function, the initial parameters of eye template are set randomly to simulate results of various preproeess. Experimental results show the new method is more accurate for eye feature extraction.