FB pixel

Facial recognition research community going beyond the algorithm to improve accuracy

Facial recognition research community going beyond the algorithm to improve accuracy

The concept of ‘garbage in, garbage out’ applies to biometric systems just as easily as it does to coding or anything else. The importance of data quality in facial recognition was explored in detail at the International Face Performance Conference 2022 this week.

IFPC 2022 was held this week by the U.S. National Institute of Standards and Technology (NIST), with support from the European Association for Biometrics (EAB) and the Department of Homeland Security’s (DHS’) Science and Technology directorate.

The first day of the event focussed on face image quality and its assessment. Christophe Busch of the EAB hosted.

Presentations included two representatives of secunet, Patrick Grother of NIST and Yevgeniy Sirotin of SAIC along with several other prominent organizations in the field of biometrics.

S&T keynote urges broad view on system performance

In a keynote presentation, Arun Vemury of DHS S&T placed face image quality within the context of building and evaluating face biometrics systems.

Vemury reviewed the major gains in accuracy and tolerance to change across the field over the past five years, before moving onto the challenges remaining.

Facial recognition systems in the real world are complex, and made up of various components. Third-party testing is common and extensive for face biometrics algorithms, but not necessarily all of the other pieces that make up a given system.

Testing for scenarios and operations, not just technology, is therefore important to understanding how well the system works. Face image quality is just one example of an area beyond the algorithm where errors can be introduced into the system.

Evaluating biometric technology appropriately, meaning for the specific use case in consideration, however, is challenging.

Part of the reason for this is because, as Vemury points out, every vendor finds a way to suggest that their algorithm is the best. Furthermore, components that are highly effective may not work well together. It is important, therefore, to consider how each element in the system will affect the others.

Related Posts

Article Topics

 |   |   |   |   |   |   |   | 

Latest Biometrics News


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Most Read This Week

Featured Company

Biometrics Insight, Opinion

Biometrics White Papers

Biometrics Events

Explaining Biometrics