Paravision emphasizes importance of FMR Max in fight against biometric bias

U.S. face biometrics developer Paravision has released a study framing its results on the latest evaluation of demographic differentials among face biometrics algorithms performed by the National Institute of Standards and Technology (NIST). The study’s main focus is showing how these results can solve the problem of algorithmic bias towards certain demographic groups, a problem that has been hitting headlines and stirring debates.

Paravision compared its results in NIST’s Face Recognition Technology Evaluation (FRTE) 1:1 Demographic Differentials with the top 25 vendors as measured by overall false non-match rate (FNMR), as of June, 2024. The next best vendor’s Maximum False Match Rate (FMR Max) is more than five times higher than Paravision’s.

FMR Max measures the system’s ability to match people in the demographic with which it is least effective. The metric is crucial when assessing how a facial recognition algorithm performs when looking at different demographics, the company says.

Differences in results between demographics can introduce multiple risks, including perpetuating discrimination and inequality, poor customer experience, financial losses and organizational as well as legal risks. Bias in facial recognition has been a hot topic with rights groups decrying the technology’s use in policing and jurisdictions increasingly introducing legislative measures against it.

The company also notes that it is one of three vendors, alongside Regular Biometrics Solutions and Viante, in the top 25 achieving an FMR Min of zero, meaning that the best-performing demographic group has no errors whatsoever.

The report also highlights NIST’s role in evaluating the performance of facial recognition algorithms and calls on biometric companies to address risks that could come from bias and misidentification.

Last year, Paravision’s algorithm scored the highest accuracy in a new category of NIST’s Face Recognition Vendor Test 1:1 Verification called Visaborder Yaw≥45 degrees. The company recorded a false non-match rate (FNMR) of 0.0025 percent with the false match rate (FMR) set to 0.000001 percent. The algorithm is part of its Gen 6 facial recognition software.

Article Topics

accuracy  |  algorithms  |  biometric bias  |  biometric matching  |  biometrics  |  face biometrics  |  Face Recognition Technology Evaluation (FRTE)  |  facial recognition  |  NIST  |  Paravision