The Accuracy of BioID's recognition

event_note June 29, 2017

The accuracy of biometric systems is measured with two rates:

  • False acceptance (FAR): how often an impostor is incorrectly recognized as a legitimate user
  • False rejection (FRR): how often a legitimate user is not recognized

The two rates are interrelated: as one is decreased the other increases (though not necessarily proportionally). BioID typically adjusts its system according to NIST recommendations to operate at an FAR of 0.1%; at this rate FRR is around 1-3% depending on the quality of the samples.

It is important to note that test results are highly dependent on the test data used and how the test is performed – for instance high-resolution photos in ideal lighting vs. low-resolution images in poor lighting. In the same way, every application is different, and real-life results can vary based on user behavior, environment, hardware, and many other factors.

BioID begins with extremely accurate face positioning. It’s widely acknowledged by the industry that a 1% error in eye position can mean 10% reduction in recognition accuracy; precise positioning is one of many reasons our face recognition is so accurate and reliable. Meanwhile, BioID’s voice recognition is highly trusted and has been used to testify in court. Finally, BioID’s multimodal face and voice recognition has been proven in banking, government and military applications.