Underlying Science Behind Biometrics Requires More Rigor, Report Says
By Stew Magnuson
The biometrics industry has seen rapid growth in the post-9/11 world with numerous companies touting products that they say can confirm a subject’s identity based on his physical or behavioral characteristics with reasonable accuracy.
The physical measurements range from finger and palm prints, to irises, facial and speech recognition and other modes. Behavior can derive from the way a person types to the way he walks.
But none of these methods is infallible, and much of the underlying basic research that can confirm the utility of the devices has not been carried out, said a National Academy of Sciences report, “Biometric Recognition: Challenges and Opportunity.” This is occurring even as the technologies are becoming more ubiquitous, it added.
“Users and developers of biometric systems should recognize and take into account the limitations and constraints of biometric systems — especially the probabilistic nature of the underlying science,” the report said.
Because a person’s physical traits change over time, or can be intentionally altered in ways to fool a machine that reads them, no single method for collecting, analyzing and confirming whether the subject is who he says he is, can be called 100 percent accurate, the report concluded.
Human recognition systems are “inherently fallible,” the report said. Faces, voices, and other modes change over time.
There needs to be more studies on how populations of test subjects interact with biometric systems, particularly if they are connected to programs of “national importance,” the report stated. Presumptions and burdens of proof arising from technologies need to be based on solid, peer-reviewed studies, it added.
Laboratory evaluations of technology are useful, but they often do not reliably predict field performance, the report said.
“As biometric recognition is deployed in systems of national importance, additional research is needed at virtually all levels of the system including sensors, data management, human factors and testing,” the report recommended.