Iris Recognition
Iris is known as one of the most accurate, distinctive, universal, and reliable biometrics to determine the identity of an individual. However, the accuracy of iris recognition depends on the quality of data acquisition and it is negatively affected by several factors such as angle, occlusion, and dilation. Since standoff iris recognition systems are much less constrained than traditional systems, the iris images captured are likely to be non-ideal, off-angle, and dilated images.
In this study, we aim to improve performance of standoff iris recognition using deep learning techniques within the traditional and nontraditional iris recognition frameworks. First, we develop a deep learning-based frontal image reconstruction framework to eliminate the effect of the eye structures on standoff images before comparing them with frontal images in database. This approach unwraps non-ideal iris images in traditional iris recognition framework using a non-linear distortion maps and occlusion masks. Therefore, we pursue to develop a generalized eye model that includes all these eye structures by keeping the approximate geometry of the eye anatomy. Second, we develop nontraditional iris recognition frameworks based on deep learning algorithms to improve performance of standoff systems using additional biometric information in ocular and periocular structures. In this approach, we also investigate the effect of the gaze angle in iris/ocular/periocular biometrics and fuse the biometric information in different standoff images. Finally, we present the new two-camera based data collection platform to collect real standoff iris images.
Standoff Iris Data Collection Platform
• Dataset: 478,000 iris images from 111 different subjects,
• Camera: IDS-3240ML-NIR, 1280x1024 resolution, 60 fps,
• Lens: 108 mm and 40mm focal length and 5.6 f-stop.
• Filter: 720nm High-pass filter
• Illumination: ThorLabs 780nm LED
