It’s hard to imagine now, but for most of history, there hasn’t been a reliable way to identify a person. It wasn’t a big problem though, because most people lived in small communities and rarely interacted with strangers. Fingerprints were used in ancient Babylon as seals on contracts, but they weren’t widely adopted or accepted as identification for individuals until the mid-1850s, when they quickly grew in popularity and eventually became the dominant biometric modality. Unlike the face and iris biometric modality, latent fingerprints are left at crime scenes (and everywhere else you go), so they are uniquely suited for criminal investigations. But fingerprints have some disadvantages that new, emerging biometric modalities do not have.
Surprisingly, not every person has a usable fingerprint – people who work a lot with their hands can actually wear them off. Fingerprint sensors also have a difficult time reading fingers that are dirty or overly dry. However, the biggest disadvantage is usability – people must stop and touch the sensor in order to use a fingerprint reader. In a post-COVID world, people are much more aware of the things that they touch. Face recognition systems are faster and more hygienic, because people don’t have to stop and they don’t have to touch anything.
Fingerprint sensors are inexpensive and widely available. For many years, face recognition technology was expensive and didn’t scale well. But that has changed recently. Face recognition systems are now affordable, reliable and easy to use.
Impact Of COVID-19
Contactless face recognition systems are especially important today thanks to the COVID-19 pandemic. Face masks are required in almost all public places. Initially that requirement caused problems with commercial facial recognition systems but vendors quickly updated their solution to meet the new expectations. Updated algorithms focus on “periocular recognition” and focus on the area of the face around the eyes. Unlike people, computers can make this adjustment easily. Many vendors have also updated their algorithms to detect whether a person is wearing a mask in the first place, which makes for an easy way to verify compliance without putting people at risk.
Facial recognition algorithms are evolving at a rapid pace. According to the National Institute of Standards and Technology (NIST), the top face identification algorithms tested in 2020 have an error rate of 0.08% compared to 4.1% for the best algorithm in 2014. To put that in perspective, that means that the current algorithms are more accurate than humans are in recognizing faces.
According to the NIST report, errors were caused mainly by image quality variations in pose, illumination and expression (PIE). In 2018, face recognition software was at least 20 times more accurate than it was in 2014, and in 2019 most programs were found to be “close to perfect” for the highest performing algorithms. These improvement have come from the integration of deep convolutional neural networks, which increase performance for poor quality images. Notably, Artificial Intelligence (Al) and deep learning are also key elements of the latest-generation algorithms. Facial recognition technology is finally reaching the same level of performance as fingerprints.