Genuine or Impostor: How Researchers Are Improving the Accuracy of Biometrics Systems Using Ensembles
By Lori Cameron
 

Biometrics systems that use facial recognition, iris scans, voice recognition, and fingerprints are used all the time by banks, law enforcement, and employers.

Biometrics also give you a convenient and somewhat fail-safe way to log in to your electronic devices.

Now that more and more people are doing financial transactions and storing personal data online, service providers are looking at biometrics systems to improve security. However, recent studies show that biometric features can change over time—an effect called “template aging.”

Adaptive biometric systems that use query samples classified as “genuine” have been proposed to deal with this problem. However, despite some good results, researchers are concerned about their robustness.

That’s why a group of computer scientists from the University of São Paulo propose the use of a stacking ensemble to improve the way biometric systems classify certain traits.

Stacking ensembles combine several different algorithms that run on the same set of data increasing the precision of a biometric match.

“Several adaptive biometric systems have been proposed for one-class algorithms. Some are better at reducing false match, while others are better at reducing false non-matches. Combining individual techniques into ensembles can produce more accurate and stable decision models,” say the authors of “Adaptive Biometric Systems Using Ensembles.” (login may be required for full text)

Division of samples used for training
Division of samples used for training. The system includes one genuine user and three impostors.

Ensembles teach biometrics systems how to better identify you.

They can improve the recognition performance of decision models, providing a more stable classification decision that will increase accuracy and security.

The authors—Paulo Henrique PisaniAna Carolina Lorena, and André C.P.L.F. de Carvalho—explore questions regarding the application of ensembles to adaptive biometric systems using one-class classification algorithms that will automatically adapt the meta classifier over time.

 

Research related to biometrics in the Computer Society Digital Library

Login may be required for full text.

 


 

About Lori Cameron

Lori Cameron is a Senior Writer for the IEEE Computer Society and currently writes regular features for Computer magazine, Computing Edge, and the Computing Now and Magazine Roundup websites. Contact her at l.cameron@computer.org. Follow her on LinkedIn.