You go up the stairs and enter the elevator. He knows which floor you need to go to. The doors to the apartment open themselves in front of you. The computer and phone "recognize" you and do not require a password. Cars, social networks, shops - everyone greets you, barely seeing you, calls you by name and anticipates your every step. This is how face recognition works. Like? Scary?
At first glance, it may seem that any organization that can afford this is watching your every step, collecting a dossier on you. But you can't even imagine how widely facial recognition technologies have spread throughout the world and what powerful prospects they promise. In addition to the above examples, face recognition systems allow you to do such simple and complex things:
- confirmation of the student's identity during online exams;
- identification of people from the "black list" at the entrance to stadiums and nightclubs;
- payment for goods;
- keeping your place in line when visiting an amusement park;
- unlocking your phone or computer.
What can I say, if in Moscow alone there is already a network of more than 150,000 outdoor video surveillance cameras. There is nowhere to hide from them, and this makes people think, but the scale of "surveillance" is not so great. The network uses a powerful facial recognition system, but it takes a lot of power to work, so only 2-4 thousand cameras work in real time. So far, mass surveillance of the population is only frightening, so it is worth focusing on the real advantages of this technology. But first things first.
Promotional video:
How does the face recognition system work?
Ever wondered how you yourself recognize a face, recognize it? How does a computer do it? Of course, human faces have certain properties that are easy to describe. The distance between the eyes, the position and width of the nose, the shape of the brow ridges and chin - all these details you unconsciously notice when you look at another person. The computer does all this with a certain efficiency and accuracy, because by combining all these metrics, it obtains the mathematical formula of a human face.
So how well is the facial recognition system currently performing? Not bad, but sometimes wrong. If you've ever come across face recognition software on Facebook or another platform, you've probably noticed that there are as many fun results as accurate ones. And yet, while the technology isn't 100 percent accurate, it's good enough to be widely used. And even make you nervous.
NEC's Paul Howie says their facial recognition system scans faces for individual identifiers:
“For example, many consider the distance between the eyes to be a unique characteristic. Or it could be the distance from the chin to the forehead and other components. In particular, we take into account 15-20 factors that are considered important, as well as other factors that are no longer so significant. This creates a three-dimensional image of the person's head, so even if it is partially covered, we can still get an exact match. The system then takes the facial signature and passes it through the database."
Should I worry about face recognition software?
First of all, face recognition is data. Data can be collected and stored, often without permission. Once the information is collected and stored, it is open to hacking. Facial-recognition platforms have not yet been heavily hacked, but as technology spreads, your biometrics are in the hands of more and more people.
There are also ownership issues. Most people don't know that when they sign up with social media platforms like Facebook, their data from that point on belongs to Facebook itself. As the number of companies using facial recognition is constantly growing, very soon you won't even have to upload your own photos to the Internet to be compromised. They are already stored there, and have been kept for a long time.
Speaking of software, they all work in different ways, but basically use similar techniques and neural networks. Each face has many distinctive features (it is impossible to find two identical faces in the world, and there were so many of them in the history of mankind!). For example, the FaceIt software defines these features as anchor points. Each face contains about 80 nodal points, similar to those we mentioned before: the distance between the eyes, the width of the nose, the depth of the eye sockets, the shape of the chin, the length of the jaw. These points are measured and create a numeric code - a "face print" - which is then fed into the database.
In the past, face recognition relied on 2D images to compare or identify other 2D images from the database. For efficiency and accuracy, the image had to be a face looking straight into the camera, with little light dispersion and no facial expression. Of course it worked pretty darn badly.
In most cases, the snapshots were not taken in a suitable environment. Even a small play of light could reduce the efficiency of the system, leading to high failure rates.
2D has been replaced by 3D recognition. This newly emerging trend in software uses a 3D model to provide highly accurate facial recognition. By capturing a 3D image of the surface of a person's face in real time, the software highlights distinctive features - where hard tissue and bone protrude most, such as the curves of the eye socket, nose and chin - to identify the subject. These areas are unique and do not change over time.
By using depth and axis of measurement that are not affected by lighting, the 3D face recognition system can even be used in the dark and recognize objects from different angles (even in profile). Such software goes through several stages, identifying a person:
Detection: Taking a snapshot by digitally scanning an existing photograph (2D) or video to obtain a live image of the subject (3D).
Centering: Once the face is detected, the system marks the head position, size and posture.
Measurement: The system measures curves on the face with millimeter precision and creates a template.
Representation: the system translates the template into a unique code. This code gives each pattern a set of numbers that represent features and facial features.
Matching: if the image is in 3D and the database contains 3D images, the matching will take place without changing the image. But if the database consists of two-dimensional images, the three-dimensional image is decomposed into different components (like two-dimensional images of the same facial features taken from different angles), and they are converted into 2D images. And then the match is found in the database.
Verification or identification: during the verification process, a snapshot is compared to only one snapshot in the database (1: 1). If the goal is identification, the snapshot is compared with all snapshots in the database, which results in a number of possible matches (1: N). One or another method is applied as needed.
Where are face recognition systems used?
In the past, face recognition systems have found use mainly in law enforcement, as authorities used them to find random faces in crowds. Some government agencies have also used similar systems for security and to eliminate electoral fraud.
However, there are many other situations in which such software becomes popular. Systems are getting cheaper, their distribution is growing. They are now compatible with cameras and computers used by banks and airports. Travel agencies are working on a “seasoned traveler” program to conduct rapid safety screening for passengers who voluntarily provide information. Airport queues will advance faster if people go through a facial recognition system that matches faces to an internal database.
Other potential uses include ATMs and cash dispensers. The software can quickly check the client's face. After the client's permission, the ATM or terminal takes a picture of the face. The software creates a faceprint that protects the customer from identity theft and fraudulent transactions - the ATM simply will not give money to a person with a different face. Even a PIN is not required.
Magic? Technology
The development of face recognition technology in the field of bank transfers can be especially important and interesting. Recently, the Russian bank Otkrytie presented its own unique solution developed under the Open Garage technology brand: transferring money from a photo in the Otkritie. Transfers mobile application. Instead of typing in a card or phone number, you just need to take a photo of the person who needs to make a transfer. The face recognition system will compare the photo with the reference one (it is done when the bank issues the card) and will prompt the first and last name. All that remains is to choose a card and enter the amount. Most importantly, clients of third-party banks can also use this function for transfers to Otkritie clients - the sender of transfers can use a card of any Russian bank.
“Using a client's photo instead of a bank card number is a fundamentally new approach to online transfers, based on the use of a neural network face recognition system, which allows a client to be identified with a high degree of accuracy by his biometric data,” says the head of the Partner Systems Development Department of Otkritie Bank Alexey Matveev. - The service opens up completely new life scenarios for users to perform money transfers. Currently, none of the financial market participants in the world offers such a service to their clients."
Ilya Khel
