基于强度和图像梯度的瞳孔中心定位
Pupil Center Localization Based on Intensity and Image Gradient
程子豪 1裴玉瑶 1周义祥 2张文东 1王常青 1周璇 3王艳玲 3吴茜4
作者信息
- 1. 安徽医科大学 生物医学工程学院,安徽 合肥 230012
- 2. 安徽医科大学 卫生管理学院,安徽 合肥 230012
- 3. 合肥市第三人民医院,安徽 合肥 230022
- 4. 安徽医科大学 生物医学工程学院,安徽 合肥 230012;安徽医科大学 人文医学院,安徽 合肥 230032
- 折叠
摘要
瞳孔中心是眼动追踪、人脸识别等计算机视觉领域中的精细参数,实现瞳孔中心自动检测具有广泛的应用价值.论文结合Faster RCNN模型,提出一种细分虹膜形状特征与图像梯度法的人眼瞳孔定位算法.首先,对图像进行光照补偿预处理,在此基础上,利用改进的ResNet50 作为Faster RCNN模型的骨干网络来检测人脸和眼睛;其次,通过几何约束对眼睛区域进行选择,采用积分图像法实现虹膜区域检测;最后,通过图像梯度算法进行瞳孔中心定位.实验结果表明:该算法在GI4E数据集及自建的面部数据集上能够较精确地实现瞳孔中心定位,并且在归一化误差0.2 阈值内,分别达到了100%和99.46%的定位精度,具有较好的鲁棒性和实时性.
Abstract
Pupil center is a precise parameter in eye tracking,face recognition and other computer vision fields,and the realization of automatic pupil center detection has a wide range of application value.Combined with the Faster RCNN model,this study proposes a pupil localization algorithm for the human eye based on the segmented iris shape features and the image gradient method.First,the image is preprocessed with light compensation,and on this basis,the improved ResNet50 is used as the backbone network of the Faster RCNN model to detect the face and eyes.Then,the eye region is selected by geometric constraints,the iris region is detected by the integral image method,and finally the pupil center is localized by the image gradient algorithm.The experimental results show that the algorithm can achieve pupil center localization accurately on the GI4E dataset and the self-built facial dataset,and achieves 100%and 99.46%localization accuracies within the normalized error threshold of 0.2,respectively,with good robustness and real-time performance.
关键词
瞳孔中心定位/Faster/RCNN/图像梯度/ResNet50/积分图像Key words
pupil center localization/Faster RCNN/image gradient/ResNet50/integral image引用本文复制引用
基金项目
国家自然科学基金青年资助项目(62001005)
安徽省高校科学研究资助项目(2022AH050660)
出版年
2024
NETL
NSTL
万方数据