Biometric Identification using Song-Based Blink Patterns Tracy Westeyn, Peter Pesti, Kwang-Hyun Park & Thad Starner GVU, College of Computing Georgia Institute of Technology Atlanta, GA 30332 {turtle, pesti, akaii, thad}@cc.gatech.edu Abstract In this paper we describe a system that uses patterns of eye blinks as a biometric. Each user chooses a personal blink pattern based on a song (for example “Jingle Bells”). To establish their identity, the user looks at the system’s camera and blinks to the cadence of their chosen song. Computer vision is used to detect the blinks and to compare the blinked cadence to a database of stored blinked patterns to determine which song is being blinked, and therefore which user is performing the blinking. To make the system more secure, the system compares the characteristics of the blinking itself, the “blinkprint”, to the user’s stored blinkprint. This provides a verification check to help protect against the theft and subsequent use of a user’s blink pattern. We discuss the possible use of an enrollment process that alerts new users when their new blink code is similar to other codes already in the database, and report on the results of a preliminary experiment using this new enrollment procedure. 1 Introduction After September 11 th , security measures in public places, such as airports, have increased. Keys, identity badges and numeric keypads are utilized to restrict access to sensitive areas, allowing only authorized personnel to gain entry. These physical systems rely on possessing specific knowledge (numeric codes), specific devices (keys and badges) or a combination of the two. In either case, physical systems rely on the manipulation or possession of physical devices and/or access to special knowledge and can easily be compromised by an unauthorized person obtaining that knowledge (learning the numeric codes) and/or acquiring the physical device (stealing a key). Physical devices give the illusion of security, as possession is the only thing a person needs to gain access to a restricted area. Biometrics rely on physiological and/or behavioral characteristics, rather than associating access with the possession of special items or knowledge. This, in turn, associates access more directly to a person by using personal characteristics that cannot be easily duplicated. Biometrics are the use of intrinsic physiological and/or behavioral characteristics to verify the identity of an individual. Fingerprint identification, DNA identification and face recognition are three examples of well-known physiological biometrics. An example of a behavioral biometric is recognizing someone by the way they walk, known as gait recognition. Biometrics offer several security advantages. Unlike keys and passwords, physical biometric traits cannot be lost or forgotten by an individual. Most biometrics are also very difficult to steal. While one could easily steal a person’s key, it would be more difficult to mimic someone’s gait. While biometrics provide several advantages, they are not without fault. Some biometrics are susceptible to spoofing. For example, 2D face recognition can be fooled using photographs (Ross, Jain & Qian, 2001). Biometrics can also suffer from lack of generality, as not all individuals possess certain biometrics indicators. For example, while everyone has fingerprints, not everyone’s fingerprints are capable of conveying uniquely identifying information. Due to the frailties of using a single biometric indicator, research in the past few years has seen a heightened interest in the fusion of multiple biometric indicators to authenticate identity (Ross et al., 2001). Combinations of biometric indicators, such as face recognition and iris recognition (Wang, Tan & Jain, 2003), can help boost accuracy of identification. In addition, as the number of biometric indicators used increases, the chance of defeating the system decreases.