Does biometric facial recognition work?

We are our face. More than any other piece of information, such as our full name, social networking profiles, ID number, or Social Security number, our face and other biometric information identify us definitively.

As novelist Frederick Buechner wrote, “I am forced to conclude that, to an alarming degree, I am my face.”

And he was not wrong: our biometric data is the most difficult, if not impossible, to falsify. Even altering your face with surgery so that you cannot be recognized has a prohibitively high price tag for most mortals, and some algorithms already apply factors that recognize certain surgeries that could have been undertaken to modify the appearance.

In contrast, data based on numbers and letters can be falsified, with more or less effort or skill. Thus, it’s not surprising that facial recognition based on biometrics is increasingly used in more areas. From unlocking your laptop or cell phone, a system that is limited to facial recognition, to identity verification, when used to identify an unknown person, that is, to connect the face to a name listed in a particular database.

Let’s set aside the legal issues, which vary significantly from country to country, and ethical concerns, which depend on the application of the technology and not on the development of the technology itself. What we are considering is whether this technology really works.

Although there are limits to facial recognition, they are related to the environment, how to identify a person walking down the street, through a video recording. For onboarding processes, to give another example, the conditions increase the trustworthiness of the identification, since they cross-reference different identity verification tests with existing databases and with authentication of document security measures.

Is biometric facial recognition really reliable?

As we have seen, a face is indisputably linked to a person’s identity. But, can the technology associate the face with an identity or a name with a sufficient degree of confidence?

Honestly, this technology is not 100% reliable, but it is between 95% and 99% accurate. Facial recognition studies conducted on the Labeled Faces in the Wild database found a margin of error lower than 5%. Currently, systems reach reliability of up to 99%, depending on the document.

In any case, it is a much higher reliability percentage than recognition carried out by humans. In 2014, a study of border control officers with specific education and training in facial recognition found that fraudulent photographs were accepted in 14% of cases.

The latest advancements in biometrics: liveness testing

Facial recognition technology has advanced by leaps and bounds in recent years. To mention it briefly, in 2014 the Android facial recognition system could not distinguish a live person from a photograph, and even positively matched people who looked similar enough.

Spoofing fraud, one of the most common schemes online, has even managed to falsify a blink by superimposing images with eyes open and closed to create a sequence.

Fortunately, technology is advancing at the same rate or faster: modern systems detect the entire movement, and therefore a sequence is not sufficient to return positive validation.

How far will biometrics identification go?

And this technology has not just fine tuned facial recognition. In 2015, Facebook announced that its algorithm could identify people even in unclear images or images in which people were not looking. Recently, according to its AI department, it can even recognize postures and gestures.

We could even go farther than that: in Japan a professor used facial recognition technology to identify emotions. More specifically, he used it to identify his students’ level of interest in their classes.

In summary, identity verification technology is much more reliable than manual identification by a human, since it can reach reliability indexes greater than 95%, depending on the system. Also, systems combine biometrics with cross-referenced third-party databases, which makes identity falsification or theft much harder, since it requires two levels of authentication.