Scientists in Great Britain hope you may never have to worry about losing your keys or forgetting your password again.
University of Warwick researchers have unveiled a new fingerprint recognition technology, which allows them to "unwarp" distorted prints. The technology could prove especially important in mass-market biometric access systems, which have remained elusive because of small but significant rates of false positives and negatives.
Fingerprint recognition came into wide use in forensic investigations in the early 20th century. Ever since, sci-fi writers and scientists have dreamed of using the unique skin contours on our fingertips to tell our machines we really are who we say we are. The problem is that the number of errors has just been too high.
"In real settings, the best algorithms I've seen are still talking about an equal error rate of 3 to 4 percent," said Dr. Venu Govindaraju, director of the Center for Unified Biometrics and Sensors at SUNY-Buffalo. "In good settings, we're looking at 0.2 to 0.3 percent." (The equal error rate is when the sensitivity of a test is adjusted to the point where the proportion of false-positive results equals the proportion of false-negative results.)
Three percent may be good enough for a low-security location like a public library, but nowhere near good enough for military-grade installations or even airport security.
Most current technologies focus on what the experts call Level 1 and Level 2 features. Level 1 is the general pattern of your fingerprint (you remember: arch, loop, whorl). Level 2 includes the specifics of the way the contours end and split. The problem is that a host of environmental factors can throw noise into the data. One major noise source is that people mash their fingers onto sensors with varying amounts of pressure, generating non-linear stretching.
"The skin stretches differently depending on how hard you press, which moves the Level 2 features around," explained Govindaraju.
It's the same idea that web forms use to defeat spammers: Distort letters a certain amount and a person -- like a CSI or you -- can recognize the pattern, while most electronic systems are defeated. Good for the cleanliness of WiSci's comment section, but bad for biometric researchers.
Enter Warwick Warp, the brainchild of three British computer scientists. The company's received seed capital from Oxford Early Investments, and it's on the prowl for another million dollars of equity financing. The scientist have come up with a system that recognizes the distortions introduced by those disobedient individuals who don't treat the sensor like a lab instrument.
"We're trying to model the variations and therefore minimize them," said Dr. Li Wang, chief technological officer of Warwick Warp. The company's algorithm filters out the environmental distortion and then unwarps the contours of the skin. Drawing on our distorted-letter analogy, it helps reflatten the letter into a shape recognizable by any spambot.
It's a promising technology, but one that is still unproven. "There's some merit to that idea," said Govindaraju. "But I've never seen real numbers."
We might see some real numbers very soon: The company plans a commercial release within six months.