Iris recognition in less intrusive environments
暂无分享,去创建一个
The human iris is considered one of the most accurate biometric characteristics. Iris recognition algorithms are shown to have extremely low false acceptance rate under ideal iris images. However, most commercial iris recognition software requires substantial user cooperation in order to get ideal images. In the research for this dissertation, we developed a comprehensive framework in less intrusive environments, including both hardware architecture and software algorithms. Our system is able to: 1) Enlarge the working volume of the iris capturing device; and 2) Recognize low-quality iris images captured from moving users at different standoff distances.
We first designed an iris capturing system that is able to detect and track 3D eye positions in real time and capture iris images from users moving at a walking speed. In order to estimate accurate 3D eye positions, we proposed an algorithm to calibrate a narrow field of view camera which contains a telephoto lens.
As users are allowed to move with a large capturing volume, the captured iris images could be easily blurred. In order to restore clear iris images for recognition, we developed an iris image restoration algorithm. The key feature of our algorithm is that we use the domain knowledge inherent in iris images and iris capture settings to improve the performance. The additional information includes iris image statistics, characteristics of pupils or highlights, and even depth information from the iris acquisition system itself.
Another problem caused by the large capturing volume is that iris images have different resolutions and noise levels. We developed a robust varying-resolution iris recognition algorithm. Our algorithm consists of two parts: 1) a modified segmentation method for images with a large variation of iris diameter and eye gaze changes; and 2) a new feature encoding method that is robust for non-ideal iris images due to noise, blur, occlusion, and down-sampling. We showed that our iris recognition algorithm can achieve state-of-the-art performance on very low resolution images.
KEYWORDS: Biometrics, Iris Recognition, Camera Calibration, Image Restoration, Segmentation