Biometric technology is becoming devastatingly accurate at matching a sample from within databases of millions of possible candidates. But what happens if that matching process is disputed? What if a justice system calls for a higher level of proof than just relying on a sophisticated algorithm's say so?
With fingerprints the process is well established. Fingerprint database searches still require a human to make a final decision from a set of potential matches. But what about other biometrics used in large-scale matching, such as iris recognition? Can human examiners be expected to provide decisions in these cases?
Researchers from the Department of Computer Science and Engineering at the University of Notre Dame have performed some fascinating research into just this area.
In a paper just released, the team focused on the question of how accurately a human examiner can determine if two iris images come from the same iris.
Although automated iris recognition technology is already very accurate and continues to improve, say the researchers, there will always be some small rate of false matches and false nonmatches. In the case of disputed results, human examiners may be called upon to make a final decision. In addition, the American justice system would likely require some level of human expert verification of a match in order to use biometric information in a courtroom setting.
Results suggest that novice examiners can readily achieve accuracy exceeding 90% and can exceed 96% when they judge their decision as “certain”. Results also suggest that examiners may be able to improve their accuracy with experience.
Professor Kevin W. Bowyer, Chair, Department of Computer Science and Engineering, University of Notre Dame told Planet Biometrics: "Humans aren't as good as the Daugman algorithm, of course. But I think it will be important that there is a human check on computer accuracy in disputed cases."
One interesting possibility is that human examiners could be making different sorts of errors when compared with than automated matching. Therefore, say the researchers, hybrid matching may achieve greater accuracy than either alone.