CORC  > 自动化研究所  > 中国科学院自动化研究所  > 毕业生  > 博士学位论文
题名远距离虹膜识别研究
作者董文博
学位类别工学博士
答辩日期2010-01-10
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师谭铁牛
关键词生物特征识别 虹膜识别 虹膜图像获取 智能人机交互 图像序列处理 特征表达 biometrics iris recognition iris image acquisition intelligent human-machine interaction image sequence processing feature representation
其他题名Study on iris recognition at a distance
学位专业模式识别与智能系统
中文摘要虹膜识别是一种新颖的生物特征识别技术,其唯一性好、稳定性强、识别率高,因此受到广泛关注。虹膜识别的算法已经逐渐成熟,但影响虹膜识别应用的最大瓶颈在于其易用性问题——虹膜图像获取困难,需要人积极配合才能完成识别,使用不方便。为了此关键问题,我们开始远距离虹膜识别的研究,目的就是设计一种能在两三米外、无需用户配合即可自动完成虹膜识别的仪器。围绕这个课题,我们分别从虹膜图像获取、智能人机交互、图像序列处理、虹膜特征比对等几个关键难点进行了研究,其主要内容和成果包括: (1)解决了远距离虹膜图像的获取问题。我们分别从虹膜成像的原理、成像的光路设计、传感器选择和光源安全与设计等几个关键问题进行了分析,并总结了虹膜图像获取装置设计的要点,最终采集和构建了远距离虹膜图像数据库。 (2)解决了远距离虹膜识别的人机交互问题。虹膜摄像机的拍摄范围和景深非常小,在远距离的情况下,几乎很难让人的眼睛对准摄像机。我们提出了三种主动人机交互方法:基于双目视觉和旋转云台的主动跟踪方法,基于自动瞄准和自动对焦的主动交互方法,以及基于摄像头阵列的图像获取方法。这些方法使远距离虹膜识别系统能够主动适应不同高度和位置的用户,使得远距离虹膜识别切实可行,并大大减少人的配合程度。 (3)提出了远距离虹膜图像序列的处理方法。远距离虹膜识别采用的是高像素、高帧速的摄像头,我们采用特殊的眼睛检测和图像质量判断方法,从序列中实时地挑选清晰的虹膜图像。在对虹膜图像序列进行识别时,我们提出了两种新的思路:一种是基于质量的动态接收阈值的虹膜比对策略;另一种是基于质量的自适应虹膜-人脸融合方法。通过这些方法,系统可以对低质量虹膜图像进行识别,从而大大提高图像序列的识别通过率。 (4)提出了虹膜特征比对的新策略。我们提出了一种基于个性化加权图的虹膜特征比对策略。每一个虹膜类别都通过多幅注册图像生成一个加权图,加权图反映了虹膜特征在不同区域的稳定性,并有许多很好的特性。利用优化后的加权图求取虹膜比对分数,可以大大地提高系统性能,尤其对于低质量的图像更加有效。其理论还可以扩展到其他生物特征识别领域。 在以上研究的基础上,我们开发了国内第一台远距离虹膜识别仪器,此仪器可以在三米外实现主动的虹膜识别,性能快速稳定,适合在大规模的人群中使用,大大改善了虹膜识别的易用性能。
英文摘要Iris recognition is biometrics technology with high accuracy and strong stability. Iris recognition algorithms are becoming mature, but the bottleneck of iris recognition is that iris image acquisition is not very convenient and users need actively cooperate with the machine to perform the recognition. For the purpose of enhancing the usability of iris recognition, we begin the study of iris recognition at a distance and try to complement non-cooperative iris recognition at 2 or 3 meters. We investigate key points involving the iris recognition at a distance such as image acquisition, human-machine interface, image sequence processing and iris feature matching. The main contributions include: (1) Implement the iris image acquisition at a distance. It includes iris texture imaging, optical design, sensor selection, illuminator safety and so on. Finally we design an iris image acquisition system and establish the image database of iris recognition at a distance. (2) Realize the active human-computer-interaction at a distance. We propose three active human-computer-interaction methods: the active tracking method based on the stereo vision and pan-tilt-zoom camera, the active tracking method based on a single wide-angle camera and pan-tilt-zoom camera, and the image acquisition method based on the camera array. With those methods, the capturing range and depth of the iris recognition system is enlarged and the system can actively cooperate with different heights and positions of users at a distance. (3) Iris image sequence processing. Since the high-resolution camera is used in the iris recognition system at a distance, we develop special image processing method to select clear iris images from such high-pixel real-time sequence. Moreover, we propose two new strategies for iris recognition. One is of quality-based dynamic accept threshold for iris matching, and the other is of quality-based adaptive iris-face fusion. Those strategies make more low-quality iris image recognized, so greatly improve the recognition rate of the sequence. (4) Improve the method of iris feature representation and iris matching. We propose an iris matching strategy based on the personalized weight map. This weight map is generated from multiple registered images and reflects the different stability of different iris regions. By calculating the iris matching score based on optimized weight map, this method greatly improves the classification performance, especially for those l...
语种中文
其他标识符200618014628035
内容类型学位论文
源URL[http://ir.ia.ac.cn/handle/173211/6231]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
董文博. 远距离虹膜识别研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2010.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace