Databases for Iris Biometric Systems: A Survey

Abstract

Due to latest advances in information technology, the research community has been exploring new applications for iris biometrics. For example, it has now been integrated into the cell phones, auto teller machines, walk-through portals, etc. Literature reveals that most contemporary iris biometric systems perform poorly for image date where a subject may wear cosmetics (e.g., contact lenses), not cooperate with system, stand at far distance, and/or be on the move. Moreover, research community has also started working on subject’s recognition from video, which is a challenging task. To validate the performance of new algorithms, researchers generally need databases. If they do not find relevant databases, then they need to develop databases demanding for funds and workforce. To address this issue, this study offers an extensive survey of public databases. These databases are classified into the ideal, less-constrained, and unconstrained classes using a subjective metric based on the system’s image acquisition range, level of constrains, and noisy factors in image data. Collectively, these databases offer noisy factors such as blur, non-uniform illumination, poor contrast, eyebrows, hair, off-axis and off-angle eyes, partially open and closed eyes, artificial eyelashes and cosmetic lenses, diseased pupils, light reflections, synthetic irises, fake irises, face images captured at a distance, short videos, and so on. Images in these databases are acquired using the near infrared (NIR) and visible wavelength (VW) illumination. No doubt, the research community could comfortably use these databases to simulate the ideal and/or non-ideal iris biometric conditions.