Facial recognition: what it is, how it works

Facial recognition systems are part of our routine life. The face is already used in different parts of the world to unlock mobiles, withdraw money at ATMs, pay at establishments, carry out checks at airports or identify suspects in mass events such as football matches or concerts. In 2018 the American singer Taylor Swift used, without warning the attendees, a facial recognition software at a concert in Los Angeles to detect stalkers among her audience. What is the facial recognition and how it works? Here’s everything you need to know about facial recognition. Let’s get into it!

What is facial recognition?

Facial recognition - one of the "technological heroes" of our time - is a biometric technique designed to uniquely identify a person by comparing and analyzing models based on his "facial contours."

There are several facial recognition techniques, such as "generalized" facial recognition and other adaptive facial recognition. The system of the facial recognition is based on the different nodal points of a human face. The values ​​measured concerning the variable associated with the points of a person's face help to identify or verify the person uniquely. With this technique, applications can use the data captured by faces and can accurately and quickly identify the individuals. Facial recognition techniques are rapidly evolving with new approaches (such as 3D technology), helping to overcome the problems present with existing techniques.

Advantages and disadvantages of facial recognition

Facial recognition is "non-contact" in nature. Facial images can be captured remotely and can be analyzed without ever requiring any interaction with the person concerned. Facial recognition has a lot of security measures for tracking the time and participation of the individual at a given event or place. Facial recognition is also a standard technology, as the processing is less "expensive" than other biometric techniques.

There are some disadvantages associated with facial recognition. In fact, this technique can identify people only when the light conditions are favorable (good lighting or focus of the face): in other words, it could be less reliable in case of insufficient light or in the presence of a face even partially obscured. Another disadvantage is that facial recognition is less effective when different human facial expressions vary.

How facial recognition works

It is essential to understand how facial recognition works - from an eminently technical point of view.

Below are three applications of this technique, listed from the most "basic" to the most complex.

·   "Basic" facial recognition: Taking the Instagram or Snapchat filters as an example, the device's camera (e.g., smartphone) "looks for" the features that define a face, and in particular the eyes, the nose, and the mouth. Once his research is done, the camera - via the Social - uses algorithms to hook up a face and determine in which direction the person is looking, if his mouth is open, etc. In this case, we are in the presence of software that "looks for faces," and we cannot speak of real face recognition.

·   Unlocking the device with the face: if you want to unlock the device with the face, the device takes a photo of the face and measures the distance between the facial features. Then, whenever you go to unlock the phone, "look" through the camera to measure and confirm the identity of the subject. Note: here, the difference between devices can be "abysmal"; just think of the technological level reached by Apple's "Face ID."

·   Identification of a stranger: In the act of identifying a face for security purposes, for advertising or police purposes, there is the use of algorithms that "angle" in a large database of faces, associating from time to time different profiles with the one in question.

The technologies underlying facial recognition

Most facial recognition software is based entirely on 2D images. The vast majority of cameras take 2D photos, and the vast majority of pictures on social media are in 2D (e.g., Facebook profile picture). However, the faces "in 2D" are not qualitatively accurate since they are flat images - or without depth - of the face, in that it does not lead us to have identification features.

With a 2D image, a device can measure pupillary distance and mouth width; however, it is not able to recognize the length of the nose or the prominence of the forehead. Moreover, the 2D facial image is based on brightness: this means that this technique does not work in the dark, and can be unreliable in low light conditions (with a high rate of false positives.

3D facial recognition is obtained through a technique called "lidar," which is similar to sonar (used in the maritime field). In essence, the facial scanning devices (e.g., the iPhone) project a sort of laser pulse on the face; this pulse is reflected on the face and is taken up by an IR (InfraRed) camera in the device.

The phone's IR camera measures how long it takes for every bit of IR light to bounce off the face and return to the device. Of course, the light reflected from the nose will have a shorter path than the light reflected from the ears, and the IR camera uses this information to create a unique depth map of the entire face. 

When used together with 2D technology, 3D technology can significantly increase the accuracy of facial recognition software, while decreasing the possibility of incurring false positives.

To solve the problem of 2D face recognition "in the dark," it is possible to use a thermal imager. Unlike the lidar, thermal imaging cameras do not emit IR light, but simply detect the IR light emitted by the objects. Hot objects emit a great deal of IR light, while cold objects emit a negligible amount. The most technological thermal cameras can detect even subtle differences in temperature on a surface, which is ideal for facial recognition.

There are several ways to identify a face using a thermal imaging camera. All these techniques are incredibly complicated, but share some fundamental similarities, mentioned below.

·   Multiple photos are required: A thermal imaging camera takes multiple images of a subject's face. Each picture focuses on a different IR light spectrum. 

·   They determine the "mapping" of blood vessels: IR images are used to extract the formation of blood vessels in a person's face. They can be used as unique facial fingerprints. They can also be used to find the distance between the facial organs (in the case where the typical thermal images produce poor images) or to identify contusions and scars.

·   The subject can be identified effectively: A composite image is created using multiple IR images. This composite image can then be compared with a facial database to identify the subject.

The thermal facial recognition is usually used in the military. Furthermore, the thermal image does not work well during the day (or in generally well-lit environments, the exact opposite of the 2D technology that "avoids the night"), so it does not have many potential applications outside the military sphere.