06 Jul 2024

Buyers Guide

Complete playbook to understand liveness detection industry

Learn More

What is FMR (False Match Rate)?

Author: admin | 06 Jul 2024

Facial recognition is the only unconstrained primary biometric identity verification standard as tested by NIST. Although the technology can have multiple errors and limitations for some reason, these errors don’t fully nullify the need for facial recognition. Rather they are testing mistakes that can be reduced over time with technological advancements and removing environmental constraints.

In this Knowledge source, we will explain False Match Rate (FMR) as one of the major errors that can occur in biometric facial identity verification compromising the credibility and robustness of a facial recognition solution.

False Match Rate (FMR) Meaning and Overview

It is the estimated error of a biometric authentication system in which it incorrectly matches two entirely different individuals and identifies them as the same person. It is closely related to the False Acceptance Rate (FAR). It is also known as False Positive.

False Match Rate formula for calculating the False Match Rate is given below:

False Match Rate (FMR) is the total number of false matches divided by the total number of identity matching attempts multiplied by 100.

It calculates the number of imposter attempts over the total attempts of identity verification through a biometric system. Particularly, in the case of facial recognition, the FMR is of high importance as facial identity matching is complicated and requires high levels of accuracy to differentiate between a spoofing attempt and a genuine face. 

Why is False Match Rate (FMR) Identification Important?

Detecting and calculating the False Match Rate (FMR) is highly important for two main reasons:

  1. To identify and detect deepfake images falsely matched with a true identity.
  2. To maintain the accuracy and precision of a facial recognition system.
The graph illustrates the False Match Rate (FMR) and False Non-Match Rate (FNMR) with an intersection between the False Acceptance Rate (FAR) and False Rejection Rate (FRR) as Equal Error Rate (EER). Both curves show the Imposter attempts and Genuine user attempts accordingly while a biometric solution finds its current performance standing at the decision threshold.

False Match Rate (FMR) Calculator in 1:1 and 1:N Face Matching

FMR can occur in both 1:1 and 1:N face matching.

The zoom-in illustration of False Match Rate (FMR) and False Acceptance Rate (FAR) that are used interchangeably to indicate the same factor of wrongly accepting an imposter identity as a legitimate one.

False Match Rate (FMR) in 1:1 Face Matching

1:1 or 1 to 1 identity matching in facial recognition refers to matching the facial photo of an individual with another one to verify that the image presented is exactly the person he claims to be. If FMR is shown in the results of 1:1 matching, it means that:

  • The face presented to the system is an imposter claiming the wrong identity.
  • The identity in the system of an individual is incorrectly matched with a person due to a technical error.

Other reasons for False Match Rate occurring can include:

  • Identical Twins trying to spoof in.
  • Deepfakes, mask attacks, and video replay attacks. 

False Match Rate (FMR) in 1:N Matching

In 1:N matching the facial identity matching is carried out taking 1 image and matching it with a database of multiple face images. If a False Match Rate (FMR) is observed in 1:N matching, it means that:

  • The face presented to the system is a spoofing attempt.
  • The database contains incorrect images or a technical error in the identity-matching algorithm.

How to Reduce False Match Rate (FMR)?

Firstly, you need to understand that biometric identity verification can’t produce a 100% accurate result in different environments. False Match Rate (FMR) occurrence during a biometric facial recognition solution testing shows the margin of error which can be corrected through the following measures:

 

Improve Image QualityUsing high-resolution cameras, ensuring appropriate lighting, and capturing clear images. 
Enhance FRT AlgorithmsImproving algorithms and machine learning techniques to better differentiate between individuals.
Facial Data AugmentationTraining diverse datasets that include various angles, expressions & lighting conditions.
Regular UpdatesContinuously update and refine models to adapt to new data and patterns.
Multi-factor Authentication (MFA)Combining facial recognition with other biometric authentication methods will also help in the reduction of FMR.
Liveness DetectionFacial liveness detection prevents spoofing by ensuring that the subject is a real and live person.
Threshold AdjustmentSet appropriate matching thresholds to balance false matches and false non-matches (FMR).