摘要
为了能在正面人脸图像上对人眼位置进行检测和精确定位,提出了一种新颖高效的分级策略。利用Gabor变换计算显著极值图,得到若干具有最大显著极值的候选人眼区域;通过PCA(Principal Component Analysis)重构对候选区域进行验证,将具有最小重构误差的两个区域选定为眼睛区域;通过两级邻域运算对瞳孔进行精确定位。该方法对面部表情变化不敏感,同时具有非迭代和计算简单的优点。通过在JAFFE数据库上的对比实验,检测精度达到99.6%,验证了该方法的有效性。
We propose a novel and efficient hierarchical scheme, which can locate the accurate positions of the eyes from frontal face images. First, Gabor transform is used to calculate the salient map and a number of rectangular regions with the maximum saliency values are selected as the coarse eye-region candidates for further verification. Second, the two eye windows with the minimum PCA(Principal Component Analysis) reconstruction errors among the eye-candidate regions are selected. Finally, the pupil centers are localized by applying two neighborhood operators within the eye windows. The proposed algorithm is non-iterative, computationally simple and robust to different facial expressions. Experimental results on JAFFE database show that this algorithm can make the detection accuracy of 99 . 6%, and can achieve a superior performance compared to other state-of-the-art methods.
出处
《吉林大学学报(信息科学版)》
CAS
2014年第3期223-228,共6页
Journal of Jilin University(Information Science Edition)
基金
吉林省科技发展计划重点基金资助项目(20071152)
青年科研基金资助项目(20140520065JH)
关键词
人眼检测
显著极值
PCA验证
邻域运算
eyes detection
saliency values
principal component analysis (PCA)-based verification
neighborhood operators