Fingerprint-Matching Accuracy Over Time
Written by Dale Garrison   

Study Shows a Critical Assumption for Fingerprints to be Valid

Anil Jain, Ph.D. admits there was a sigh of relief when he and former student Soweon Yoon, Ph.D. finished analyzing longitudinal data for a massive and groundbreaking study of fingerprints. Depending on what they found in research unveiled this spring, the validity of fingerprints as courtroom evidence could have been shaken to the core.

Labeled with the innocuous title of “Longitudinal Study of Fingerprint Recognition,” the work sought to verify a long-held assumption that individual fingerprint recognition accuracy does not change over time. Although taken for granted for more than 100 years in law enforcement, scientific evidence has been lacking to prove that a fingerprint taken from a 23-year-old suspect will match his or her prints at age 35. As long as that door stood open, huge numbers of fingerprint identifications were realistically open to question.

“Fingerprint recognition has been used for over 100 years,” Jain said. “We see it in law enforcement and, more recently, it’s been used for access control, mobile phones, and even some financial transactions. We have generally been led to believe that every finger has a different friction ridge pattern, and as we age, our fingerprints stay the same. But there has never been a real in-depth study.”

Regarding his and Yoon's work, Jain put it this way: “Although the uniqueness of fingerprints has been investigated by developing statistical models to estimate the probability of error in comparing two random samples of fingerprints, the persistence of fingerprint recognition has remained a general belief based on only a few small-scale case studies.”

Beyond Assumptions

Although most assume that fingerprints remain consistent over time, assumptions often do not stand up in science or in court. In fact, some evidence to the contrary has always existed in this very area.

“We know that the condition of the finger does change over time, whether due to occupation or injuries,” Jain noted. “There are other issues where the condition of the finger can change. The finger may be dry or wet, and even the act of acquiring the fingerprint itself causes some distortion. There has always been room for questions. If there is distortion in the print or if the image is faint, then the similarities between two prints of the same finger can become less evident.”

Jain, his students, and others have studied related issues of fingerprint uniqueness for more than 25 years. “But we’ve always been wanting to study the persistence of fingerprints over time,” he added. “It’s an area that has not really received adequate attention.” There have even been a few studies that found increased error rates in fingerprint matching when the impressions are more than five years apart in age.

As those in law enforcement are aware, such questions are increasingly finding their way into law enforcement and courtrooms. In Daubert v. Merrell Dow Pharmaceuticals (1993), the Federal Rules of Evidence superseded formerly accepted tenet. The Daubert ruling established a guideline for admitting forensic evidence that included the original Frye standard of general acceptance, plus empirical testing, peer review and publication, known or potential error rates, and the standards controlling the operation. The new Daubert standard immediately brought challenges to forensic evidence, including the admissibility of fingerprint friction ridge evidence.

In their report, Jain and Yoon noted, “Although all of about 40 such challenges resulted in a decision that friction ridge analysis is acceptable as forensic evidence, the Daubert case highlighted a lack of scientific basis of persistence and uniqueness of fingerprints and standards that can be universally referred to in friction ridge analysis.” In other words, the time for assumptions was ending.

One of the hurdles for improving fingerprint validity was access to a large dataset that included multiple fingerprints from the same individuals over a long period of time. Without such data needed to show a statistically significant result, the study would be, once again, relying on assumptions. “The only source for such data is law enforcement,” Jain concluded. “They’re the only ones who encounter the same individual—and capture those fingerprints—multiple times over a long period of time.”

Fortunately, Jain had worked previously with Capt. Greg Michaud, director of the Forensic Science Division, Michigan State Police. “He’s always been willing to help support our research,” Jain acknowledged. “But this time he was exceptional.”

The challenges were also difficult because Jain and Yoon, who is now with the National Institute of Standards and Technology, needed to be precise in their data requirements. Because they needed a large quantity of prints from multiple individuals, they set as a minimum standard that individuals must have been fingerprinted at least five times over a span of no less than five years.

Michaud and his staff at Michigan Forensic Science Division came through. They produced tenprint records for almost 16,000 individuals, who were each fingerprinted five or more times over a span of five to 12 years.

“This is what made our study scientifically valid,” Jain said. “We had a large amount of data, not something that is easily available. I can’t think of a similar study with that amount of data over a long span of time.”

Critical Methodology

Statistical models—multistage or hierarchical models in particular—noted Jain, are necessary to analyze the longitudinal data. In plain English, adding time as a factor is necessary to reach any certain conclusion about persistence of fingerprint recognition.

“This kind of longitudinal data is often used in many studies in health sciences to study the impact of drugs over time,” he explained. “That’s essentially what we are doing, except we’re looking for how fingerprint accuracy might change over time. Until now, that is a property that has not really been studied.”

A preliminary examination of the longitudinal fingerprint data showed some anomalies. Individuals involved in heavy manual labor, welders, or people who work around chemicals are known to have fingerprints that alter over time. Jain and Yoon found one individual, in particular, who had very deep cuts on his fingers that caused significant changes to his fingerprints. But—and here’s the good news—within the 16,000 individuals, the vast majority presented no fingerprint changes that would render their prints unusable during the time frame studied.

“It was like a sigh of relief,” he laughed. “Using this well known statistical model, we showed that over the 12-year lifespan for which we had data, the fingerprint accuracy essentially does not change. What we’re saying is that one of the main premises for fingerprint recognition is indeed true. At least with this operational data, we can conclude that fingerprint-matching accuracy does not degrade over time.”

Published in the prestigious Proceedings of the National Academy of Sciences June 2015, the study is not the last word, Jain insists. “We need more studies like this to put forensic science on a firm footing. There are a lot of things being used with force of law, but the science really isn’t there. Fingerprints were used for more than 100 years without these questions being asked. These issues were clearly and strongly stated in the 2009 landmark report by the National Research Council (NRC), ‘Strengthening Forensic Science in the United States: A Path Forward.’”

For this and other reasons, Jain sees this study as a beginning, not an end. The most immediate issues involve fingerprint quality, including both the results obtained in the typical police station and with latent prints acquired in the field.

“The quality issue (of fingerprints) is important,” he noted. “The condition of the finger is important. Are there scars or other factors that may affect the quality? But the second thing is what kind of fingerprint reader is being used? Some aren’t so good. These are questions we need to address.”

While none of these concerns may lead to immediate solutions for law enforcement, they could result in a dismissal or erroneous match in later years. “We found that in our statistical analysis, the quality of the image was more important than the time gap,” he said.

Related Areas

Jain already sees the need for similar studies in other areas. “We’re doing something on the same lines with face recognition,” he said, noting that adding time gap to that equation has even more impact than in fingerprint analysis. “We all know that our faces age; no one will claim that how we look is the same over time.” But how that can be measured and quantified in term of its impact on face recognition accuracy has rarely been examined.

In this work, Jain and his Ph.D. student, Lacey Best-Rowden, are using a large dataset of face images provided by the Pinellas County Sheriff’s Office in Florida and fitting a similar model to the fingerprint study for analyzing mugshot data collected over 20 years. “We agree that face recognition will degrade with time,” he said, “but we want to know at what time gap the accuracy will drop significantly. Should we update photos of a subjegct in the database after five years? Ten years? At what age will the accuracy drop significantly?”

Whether faces, fingerprints, or other biometric traits, such fields increasingly involve more than law enforcement, and all eventually will require refined benchmarks, including reliable time frames. “Our driver’s licenses are good for five years; our passports for ten. But there is no formal methodology to arrive at this time interval; maybe it should be seven. And, should it depend on the gender?”

Here, too, fingerprints are a good example. Increasingly used to unlock smartphones, gain access to secure rooms and buildings, or for financial transactions, fingerprint science and technology is not likely to sit still. Whether it holds up in court or to ensure fingerprint products will be successful in the market, research is needed.

“There is a great deal of forensic science that is based on assumptions about the distinctiveness and impact of time, but we need access to large data to really determine that,” Jain concluded. “Until we do, we can’t really say for certain about fundamental premises of forensic science.”

Study Highlights

Some of the conclusions of the “Longitudinal Study Of Fingerprint Recognition”

  • Genuine match scores tend to decrease as the time interval between two fingerprints being compared increases and as the subject’s age increases, or when the fingerprint image quality decreases. Despite this downward trend, the probability of true acceptance remains close to 1.0. However, if either of two fingerprints in a comparison is of poor quality, uncertainty of acceptance increases significantly.
  • The time interval, the subject’s age, and fingerprint image quality can explain the variation in genuine match scores, but a subject’s sex and race have marginal impact.
  • Fingerprint image quality explains the variation in genuine match scores better than time interval and subject’s age.

About Dr. Anil K. Jain

Jain is a computer scientist and University Distinguished Professor in the Department of Computer Science and Engineering at Michigan State University. Known for his contributions in the fields of pattern recognition, computer vision and biometric recognition, he is among the world’s most-cited computer scientists.

Born in India in 1948, Jain earned his Bachelor of Technology in electrical engineering from the Indian Institute of Technology, Kanpur, in 1969. He received his Master of Science and Ph.D. from Ohio State University in 1970 and 1973, respectively. He taught at Wayne State University from 1972 to 1974 and joined the faculty of Michigan State University in 1974.

In 2007, Jain received the W. Wallace McDowell Award, one of the highest technical honors given by the Institute of Electrical and Electronic Engineers (IEEE), the world’s largest professional association for the advancement of technology, in recognition of his pioneering contributions to theory, technique, and practice of pattern recognition, computer vision, and biometric recognition systems.

Jain served as a member of the U.S. National Academies panels on Information Technology, Whither Biometrics and Improvised Explosive Devices (IED). He also served as a member of the Defense Science Board and currently serves as a member of the Forensic Science Standards Board (FSSB).

About the Author

Dale Garrison is a freelance writer in Liberty, Mo.



< Prev   Next >

Lifting Latent Fingerprints from Difficult Surfaces

ALMOST ANYONE can find, process, and lift a latent print that happens to be in a logical and obvious place like a door handle, a beer can, or a butcher knife. But sometimes, a latent print is not just sitting there in a logical and obvious place. Sometimes, you have to use your imagination to find the print and your skills to lift it.