DECEMBER 23--Iridian Technologies (Moorestown, NJ; www.iridiantech.com) has announced the results of a new iris-recognition study, based on 983 million matches, that further confirms false accept rates of approximately one in 1 million in extremely large enrollment databases. Together with the low false reject rates of Iridian's imagers, these results place iris recognition as the most accurate of all biometrics.
The anonymous study was conducted by Iridian and was based on nearly 120,000 iris templates collected from operational installations around the world that included physical access, corrections, border crossings, and other applications. These templates were used in a "probe vs. target" match scenario to generate an imposter distribution composed of 983,726,000 matches. With this large number of samples it was possible to determine the false accept rate for large database applications without relying on statistical modeling of the distribution.
"Private and public sector organizations responsible for border crossing, simplified passenger travel, and national identity programs have been clamoring for detailed iris recognition performance metrics when dealing with databases numbering in the millions of records," said Jim Cambier, vice president of research for Iridian Technologies. "These results support Iridian's long-held statistical projection of false accept rates of 1 in 1 million. Because those estimated error rates are orders of magnitude better than all other biometrics, iris recognition has achieved the reputation of most accurate biometric. These low false accept rates enable one-to-many or one-to-all identification applications and detection of duplicate enrollments."
False accept rates for iris recognition have been measured regularly on ever-larger database sizes since 1996. Previously false accept estimates were obtained by compiling imposter distributions composed of up to 9.1 million matches, fitting a binomial model to these distributions, and predicting operational false accept performance from the "best fit" binomial distributions.
