We see hundreds of faces every day. They look at us from the screens of phones, on social media platforms, we see them at work, at home, on the tram, and in the park. Human faces are such an obvious visual element for us, that we don’t think about how our brain recognizes and identifies them. Cognitive science and neuroscience try to explain how our brain recognizes faces and then can tailor them to a specific person. Technological development allowed us to go a step further – we teach artificial intelligence to recognize and verify human faces. The use of facial biometric technology is one of the most popular methods of identity verification id verification service today. So what does this process look like with humans, and how do machines do it?
Despite many studies, we are not able to clearly determine when we learn to recognize faces. Research on newborn babies shows that faces are more interesting for them, and they stare at them longer than on the other objects, even an hour after birth. The brains of four-month-old babies recognize them on the same level as adults. We also know that our ability to recognize and identify faces grows with age. The part of the brain responsible for this in adults is even 12% larger than in children. The fusiform gyrus in the temporal lobe of our cerebral cortex is responsible for the ability to recognize the face.
The human face recognition spectrum is also worth mentioning. At one end are people with prosopagnosia, a disorder that makes them unable to recognize faces. We are talking not only about the faces of friends or relatives but also about their own. When getting to know a person, such people try to focus on other characteristic features, like clothes, hair color, skin, or voice. This disorder has a genetic basis, and prosopagnosics may also have additional dysfunctions like dyslexia. Among others, Steve Wozniak, co-founder of Apple, Stephen Fry, British actor and writer, and Victoria, princess, and heir to the throne of Sweden have prosopagnosia. At the other end of the spectrum are people called super-recognizers. The term was used for people who have above-average facial recognition abilities and can identify the people they have seen once in their life. They can also name the place where they saw such a person and what they were wearing on that day. About 2% of the world population possess this skill. Most of them are not even aware that it is something unusual. The middle of the spectrum is people with normal face recognition ability. This is by far the largest group. There are many tests created by scientists to recognize the level at which a person recognizes faces. Some of them are generally available, and you can test your skills yourself here or here.
Technological development made it possible to develop a face recognition system in which artificial intelligence recognizes faces. They are recorded in a digital image or video frame, and AI matches them with a database. The difference between face recognition and face verification is quite significant. We usually do not participate voluntarily in the facial recognition process, and such systems are used for street monitoring in London or in Chinese cities. Street face recognition systems are designed to help police identify criminals, but many people have serious objections to them. In China, these systems are being developed without implementing privacy rules that could prevent government surveillance in this area of life. The open use of these systems to monitor protests and identify protesters in Moscow may also cause concern. Several cities in the United States have banned the use of the so-called Big Brother technology on the streets, explaining that this system will not really help people, and may contribute to the use of the position of the government or companies that provide this solution. Regardless of the extreme emotions it evokes, the facial recognition market is expected to grow by 14.5% between 2020 and 2027.
Face verification is a system in which artificial intelligence creates a face map, compares it with a photo on an ID card, and can determine whether a person is really who they say they are. The user is aware that the face verification process is being carried out. Algorithms can capture the characteristic features of the appearance and recognize the face even despite the beard or strong makeup. We do not have to look far for examples of using this technology – it is present, for example, in Apple smartphones. By applying approximately 30,000 invisible points to the face, the algorithm creates a three-dimensional map of it. The system also works at night. Apple is constantly improving its technology – back in 2017, many make-up YouTubers tried to trick the system and see how much they could change their face so that the algorithm would not recognize them. One of them managed to do it only when she used prosthetic make-up that changed her nose and chin. Currently, improvements to the system are expected to include facial recognition in masks.
Face verification is also used in banks, financial and security institutions, health care facilities, and other companies where there is a need for an identity verification service. Many companies, however, ask users to take a photo (selfie) and optionally a photo of the document. However, such a solution is not compatible with KYC and AML regulations. Due to the low level of reliability and safety, especially for the high-risk industry, it is not recommended. The security system can be improved by the use of remote identity verification recorded on video, as Fully – Verified offers. Recording of the process allows us to see the user, so we know what he is doing, whether he is acting suspiciously, and how he fills the steps of the process. Above all, however, it is safe and builds customer confidence in the service. The most efficient form of identity verification is based on document control and a face verification system. The combination of this and an experienced operator checking or performing verification is the key. We must remember that AI can make mistakes. Complete reliance on AI judgment can lead to errors that could have been avoided by letting the human check the procedure.
How reliable are facial recognition systems? Algorithms that learn face recognition have made great progress over the past few years. In 2014, the best algorithm had an error rate of 4.1%, while the best algorithm rate in 2020 was only 0.08%. However, specialists note that algorithms must constantly improve, and the lack of a factor controlling their results may increase the risk of errors. It is worth mentioning that the face verification algorithms are much more effective than the face recognition algorithms due to environmental conditions. Verification is done in a controlled environment, and the users know they have to look straight into the camera so that the shadow does not cover their face. They also don’t try to hide. Facial recognition is much more complicated due to the uncontrolled environment and the inability to correct the person’s position or removing accessories from the face or head.
The use of biometric features in identity verification is one of the safest and effective procedures. Identifying a person by examining the iris of their eye or measuring the blood vessel system in their hand is currently one of the most reliable methods. The iris of the eye is individual for each person. Even identical twins have different irises. It is similar to the arrangement of the veins in our hands, which are also unique. However, such identification is difficult and requires perfect conditions. The identification of the biometrics of our faces is so much easier, and it can be touchless and performed remotely. We can see such technology in action at the airports (in the United States or Canada), and 93% of them plan to develop this technology by 2023. The pandemic has also contributed to the growing interest in remote face identification. The search for remote identification options and the introduction of advanced technological solutions to various facilities have made face verification more popular.
The development of face verification technology is important to us because it significantly increases the sense of security. The future of face verification is all about the constant development of technology and algorithms. Currently, to be sure of its results, AI needs human control, but in a few years, it may be absolutely reliable and independent. Although some scenarios seem a bit abstract to us now, many of these solutions are waiting to be implemented.
Fully-Verified was created as answer to its founders collectively losing over $150 000 to various types of fraud in their eCommerce businesses.