December 18, 1997

High-Tech Security: The Eyes Have It

Warren Webb, Technical Editor

In the race to create the ideal personal-identification system, design engineers are turning to the human iris as the biometric of choice.

With bank fraud increasing annually, passwords and personal-identification numbers (PINs) just don't seem to cut it anymore. So, financial institutions are looking for new ways to protect customers and their money by developing mechanical and electronic systems to identify individuals by some unique biological characteristic instead. One of the latest areas of research has been iris identification.

Fingerprints, speech patterns, handwriting, and hand geometry are just a few of the biometrics that researchers have considered for identification systems. However, the iris provides an ideal means of biometric identification for several important reasons. Its randomly formed textural variation is unique to each individual, and it remains unchanged throughout life. Even the left and right irises differ. In addition, each eye has more than 170 discriminators that distinguish one eye from the next, compared with about 30 discriminators for the fingerprint. Al-though the iris is a membrane between the cornea and the lens, its image can be captured from a distance. And because the iris is an internal part of the eye, any attempted modification of the iris carries an unacceptable risk of vision loss and therefore reduces the possibility of fraud.

Many identification-system vendors, such as Sensar Inc, believe that all these factors make the iris the ideal biometric for identification. And, many of Sensar's customers, including NCR's Financial Systems Group, Citicorp, LG Electronics of Korea, and Oki Electric Industry Co Ltd of Japan, are betting that the iris will eventually replace PINs at ATMs worldwide.

Iris-identification systems aimed at access control and general authentication have been available for several years. The drawback of these systems, however, is that the user must accurately position his eye in front of a camera. A video display or other alignment device gives feedback to the user so that he can adjust his position for an accurate video image.

The key breakthrough in the development of Sensar's IrisIdent system was the ability to capture a detailed, well-focused image of the iris during the 3- to 5-sec period when the customer approaches the ATM. That is not a simple task. The IrisIdent contains a complicated distributed-processing system to capture video signals from three cameras, analyze the video images, and drive a robotic pan/tilt/focus camera control based on the results--all in real time (Figure 1).

The IrisIdent System has two main elements. The primary processing element, the IrisIdent Processing Platform (IPP), consists of a Pentium-class motherboard, a custom vision-accelerator board, frame storage, an 8-bit video digitizer, a Texas Instruments (Dallas) TMS320C3x-based DSP board, and serial or Ethernet interfaces for a user interface. The other main element, the IrisIdent Optical Platform (IOP), contains the optical elements necessary to capture the iris's image. Two cameras scan a wide-angle view of the scene to locate the customer's head and finally the eye in 3-D space. A third narrow-field-of-view camera with microprocessor-controlled pan, tilt, and focus motors captures the iris's image. The IOP also contains a controlled illumination source to ensure that the corneal reflection falls on a predictable location in the video image. The system can then exclude from analysis the portion of the iris obscured by the corneal reflection. The location of the light source also eliminates reflection from eyeglasses or sunglasses that the customer may be wearing.

Two stereo cameras feed signals to the vision-accelerator board, which does the preliminary calculation using a multiresolution-stereo approach, explains Marcus Salganicoff, PhD, Sensar's director of algorithms. Sensar uses proprietary ASICs developed by Sarnoff (Princeton, NJ) along with a crossbar switch that Sensar developed to dynamically configure the pipeline and to perform the required filtering and matching that is necessary for the stereo. The company also uses a template phase-detection algorithm that looks at the relative intensities of different regions of the wide-field-of-view image to detect the face and then the position of the eye within the face. The template routine returns the x and y value of the eye in the wide field of view. The system then computes the z value from the stereo information. The system then has the x, y, and z information to direct the narrow-field-of-view or zoom camera to the right point in space and also to set the focus value appropriately. Using these steps, IrisIdent can capture a very well-focused iris image.

Synchronize mechanical motion

When a system tries to capture a clear image of a 1/2-in.-diameter iris from as far away as 3 ft, any vibration or movement in the optics system could be disastrous. Sensar's engineers solved this problem by synchronizing all pan, tilt, and focus motion to occur only during the vertical retrace of the three videocameras. Using a trajectory profile on the mechanical motion, the system minimizes any excitation of mechanical resonant modes in the pan/tilt unit.

Once the system captures the narrow-field-of-view or zoomed image of the iris, it uses a proprietary algorithm developed by John Daugman, PhD, of Cambridge University (UK). The algorithm performs a wavelet decomposition of the iris, which results in a 256-byte iris code, no matter what the illumination conditions. The system then compares that iris code with the iris codes already stored in the system's database. The system quickly compares the codes by exclusively ORing them together. Based on the percent agreement with that compared iris code, IrisIdent can quickly make a high-accuracy determination of a person's identity, according to Salganicoff.

Daugman's patented algorithm uses a circular grid as a guide to select data from the iris image. The algorithm selects data points on the circular pattern at various angles, omitting those points that are normally obscured by the eyelids or by reflections from the illuminating light source. The circular pattern is stretched or compressed, depending on the dilation of the pupil to yield the same number of data points. A key finding in Daugman's work is that all irises can be mapped into a 256-byte code, regardless of the apparent amount of iris detail (Reference 1 and Figure 2). IriScan owns the worldwide rights to Daugman's patent; Sensar licenses Daugman's algorithm for financial applications.

So, how soon can you expect to see iris identification at your bank's ATM?

"You're probably not going to bump into them on the street corner. But if you're in the right test market, you'll be able to see these things in operation," says Michael Negin, PhD, Sensar's vice president of technology.

Sensar expects to manufacture approximately 500 IrisIdent systems in 1998. The company is establishing pilot programs with ATM vendors and their banking customers in Europe, Japan, and the United States for the most part. South America and Korea are also under consideration for these customer-validation and -reaction trials.

Sensar is also working on the next-generation product that it plans to offer in early 1999: a smaller, cheaper optical platform that can be directly integrated into ATMs as they are manufactured. The units available today must be mounted on top of an ATM, because they are too big to fit inside.

More iris-ID applications

ATMs are not the only place that iris-identification technology is being used. Other applications include hospital-patient-identity verification, access control, and law-enforcement bookings and prerelease verifications. The Lancaster County Prison in Pennsylvania uses an iris-recognition system to provide positive identification of prisoners. The Japan Racehorse Association is even testing iris identification for racehorses to eliminate microchip implants or tedious manual recording of physical appearance.

In the future, cameras will verify your iris as you log onto computer networks and periodically reverify your identity to ensure that no one else uses your computer while you're out for coffee. Automobile antitheft devices will identify authorized drivers via the iris and eliminate the need for the ignition key. The iris and other biometrics are also being investigated for immigration control and national identification cards. Whatever the application, iris identification offers a new level of security for users, who will be thankful for having fewer passwords to remember.

Reference

1. Daugman, John G, "High confidence recognition of persons by a test of statistical independence," IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 15, 1993, pg 1148 to 1161.

Acknowledgments

Thanks to Michael Negin, Marcus Salganicoff, and Gary Greene of Sensar Inc for their assistance.

| EDN Access | Feedback | Table of Contents |