Numerous important databases contain images of our faces to identify us accurately. These databases include employment files, law enforcement, driver’s license and passport.
Images are also used when comprehensive criminal background checks are performed, something that has become a crucial practice in the recent years. (Here are the 7 vital components of background checks.) As we age, our faces change. For how long will old images still be able to identify us?
Anil Jain, a Professor of computer science and engineering at Michigan State University and a biometrics expert set out to explore to what extent aging affects the ability of automatic facial recognition systems. Depending on how much facial aging affect these systems, it could have huge implications on their ability to identify criminals successfully. Jain and his team also hoped to determine how often identification documents need to be renewed.
Jain explained that they wanted to find out if state-of-the-art facial recognition systems could recognize the same face many years apart, for example at age 20 and then again at age 30. He added that this is the first study of automatic facial recognition that used a large longitudinal face database and a statistical model.
Jain worked with doctoral student Lacey Best-Rowden and determined that up to 99 percent of the facial images could still be recognized six years later. After the six years however, recognition accuracy begins to drop due to natural changes that occur to a face over time as the person ages. The study also found that this reduction in face recognition accuracy depends on the person. Factors such as health conditions, lifestyle, genetics and environment cause some people to age faster than others.
Pete Langenfeld, a manager in the Biometrics and Identification Division at the Michigan State Police believes that the research highlights the importance of taking new images every four to five years. If this were done, it would reduce the chance of not finding a person in a facial recognition search and the number of false positives due to length of time between images. Image acquisition of criminals depends on the number of times a person is arrested, as the bulk of these are not required to update their image. Civil applications that use facial images should however look at reducing the time between captures if the required update is greater than every four years.
Mugshot databases are the biggest collections of facial aging photos available that have standards in place to make sure the photos are uniform. For this reason, Jain and his team looked at two police mugshot databases. They looked for repeat criminal offenders with each offender having a minimum of four images taken over a minimum of a five-year period. They found 23,600 candidates to study. It is believed that to date, these are the largest facial aging databases studied in terms of number of subjects, images per subject and times elapsed.
Brendan Klare, CEO of Rank One Computing, a major supplier of face recognition software noted that academic research has facilitated automated face recognition to play an increasingly big role in the criminal justice system. He does however feel that there has been a lack of research on the appropriate usage of these systems thus far.
Klare added that the study undertaken by Jain and Best-Rowden provides a unique body of knowledge on the subject of the limits of automated face recognition for the first time.
The study was done in collaboration with the National Institute of Standards and Technology and the results will be published in the IEEE Transactions on Pattern Analysis & Machine Intelligence journal.