Researchers are hard at work on systems for identifying people based on their physical characteristics. The aim is to do a better job of verifying who you are for situations like crossing national borders or conducting financial transactions, or to ferret out your identity if you refuse to identify yourself to police. There are already huge biometric databases full of fingerprints, face scans, and genetic material. As iris and palm-vein scanners become more widespread, so will the number of database records based on those features.
Last year, Anil Jain, a Michigan State University computer science professor and coauthor of ”A Touch of Money” in the July 2006 issue of IEEE Spectrum , expanded the field of biometrics when he announced he had begun work on an automated tattoo recognition system that will allow authorities to identify people more quickly and accurately based on the ink embedded under their skin. Jain points out that tattoos and other marks cannot uniquely identify a person. But he says that his system will still be valuable because of its ability to help authorities narrow down the list of identities to which they’ll compare a suspect they have in custody.
How can tattoo matching be effective when, unlike fingerprints or iris patterns, body art is not something everyone has? A 2006 report from the Journal of the American Academy of Dermatology reported that about 36 percent of U.S. residents between ages 18 and 29 have at least one tattoo. Perhaps that doesn’t sound like a lot, but authorities use tattoos and other marks to properly identify murder victims and criminals, who use aliases frequently enough to make tattoo recognition worth the effort.
Instead of having a witness sit in a room for hours, thumbing through huge books filled with yellowing photos in an attempt to identify a criminal, police using Jain’s Tattoo-ID system will be able to dramatically narrow the parameters of the search by looking in one of eight digital bins labeled according to an American National Standards Institute/NIST tattoo classification standard. These categories include ”human forms and features,” ”animals and animal features,” ”flags,” ”insignias & symbols,” and the ever popular ”other.” Each of these categories is further broken down into subclasses. This represents a dramatic improvement over letting database software file away and subsequently call up images based on keywords that are manually input and not standardized, which increases the chances of failing to find a relevant record.
Jain's is team is conducting its research using the Michigan State Police’s 70 000-image tattoo database. Just like the returns that Google or Yahoo provide for a query on their search engines, Jain’s system ranks the images pulled up by a particular query based on points of similarity to the new image. According to Jain, if he and his colleagues have a photo to go by, it takes 2.3 milliseconds for a computer with a 2.66-gigahertz, 3-gigabyte Intel Core 2 Duo processor to decide which key points it will compare. Calling up potential matches from among the Michigan Police database’s 640-by-480-pixel color images takes just over three-tenths of a second.
But fast is not a synonym for good . Though the system discards points that have low image contrast and performs several other steps to improve accuracy in determining the points of similarity between two images, Jain and his team have achieved retrieval accuracy of only 77.2 percent. A 20 percent error rate plus the fact that 20 percent of the images in the database are copies of the same tattoo—some from the same person arrested multiple times, and some images that are nearly identical but on different people—creates a strong chance that someone could be falsely accused of a crime or misidentified because he, like the perpetrator, happens to have, say, a bald eagle or a spider tattooed on his arm. This raises an important question: Is a tattoo is enough of a unique identifier to put someone under suspicion?
Have your say on these other new technologies: