5: Eye Biometrics: Iris and Retina Scanning in Software

Create Code 128B in Software 5: Eye Biometrics: Iris and Retina Scanning

for those code lengths However, perhaps additional header information or changes have since been added to the process as Iridian now describes the iris as being processed into a 512-byte IrisCode (For more information, go to http:// wwwiridiantechcom/) Regardless of the code size, the theory and technique are unchanged Hamming distance can also be expressed as the fractional difference between two binary sources of equal length Dr Daugman s mathematical analysis of IrisCode comparisons shows iris recognition to have a very low error rate The odds of two different irises generating a sufficiently similar code to produce a false match is theoretically 1 in 12 million5 In practice, iris recognition systems perform extremely well A recent real-world technology test with 200 subjects reported iris recognition performance as almost flawless Iris recognition performed better than the fingerprint, hand geometric, voice, and face systems involved in the test There were no decision errors (no false accepts or rejects), and the single error reported for iris recognition was one of the subjects failing to enroll due to a blind eye6

Initial commercial markets for iris scanning were mostly geared toward physical access applications These applications are characterized by having a relatively small population of enrolled users and fairly infrequent identification events Thus, a single modern PC easily provides sufficient processing power and storage capacity to support them An access control unit produced by IrisScan is shown in Figure 5-2 With growth in security markets, equipment manufacturers and security integrators have aggressively pursued the broader space of authentication and identification applications, including desktop authentication, secure banking, biometrics on access cards, passenger screening, and even remote surveillance In 2000, the Charlotte/Douglas International Airport in conjunction with EyeTicket of McLean, Virginia, piloted iris recognition for pilot access and passenger screening (For more information, go to http://wwweyeticketcom) As with any biometric, the advantages and disadvantages of iris identification depend heavily on the application and the intended target environment From a pure technology standpoint, iris representation and matching techniques are elegant and refined and the systems have been well tested in a variety of environments ranging from cozy offices to ship decks in the military Iris identification is a known accurate technique with transaction time for trained

J Daugman, 1998, How Iris Recognition Works, University of Cambridge, http://wwwclcamacuk/users/jgd1000/irisrecogpdf Tony Mansfield, 2001, Biometric Authentication in the Real World, The National Physical Laboratory (NPL) Center for Mathematics and Scientific Computing, Middlesex, United Kingdom

Figure 5-2

users, being as fast as 4 seconds, and the majority of that time is spent by the subject aligning his or her eye Once the image is captured and processed, the actual computer matching is fast, on the order of micro-seconds Recent tests observed average transaction time for iris verification to be about 10 seconds, where that time also included entry of a four-digit PIN7 From the standpoint of the trait itself, it is an advantage that the iris is visible (hence, readily scanned) but also a naturally protected part of the eye While eyelids, eyelashes, contact lens, size variation, and natural movements present challenges for isolating iris samples from images, processes to accommodate these difficulties have been worked out for most environments The natural motion of the iris is said to be the basis for liveness detection, although this has not been established through independent testing The small size of the iris and its curved surface are easily accommodated for in cooperative access control applications using short-range frontal scans However, the small size of the iris presents significant challenges for longer range, noncooperative surveillance applications where image resolution, camera angles, occlusion, and isolation and tracking increasingly become factors

Tony Mansfield, et al, 2001, Biometric Product Testing Final Report, The National Physical Laboratory (NPL), Center for Mathematics and Scientific Computing, Middlesex, United Kingdom