Since the advent of computers, there have been many technologies that have had a great impact on our lives. One of these technologies is face recognition, which analyzes and authenticates people by creating a face map and can have various applications. In this article, we will learn about this interesting and controversial technology.
Many of us first got acquainted with facial recognition technology in science-fiction movies; Where a laser would fall on the actor’s face and apparently be able to identify him. But most movies have not properly showcased this technology. In fact, different facial recognition systems work in different ways and each usually has its own algorithms. But before we see how this technology works, let’s first see what facial recognition is.
What is facial recognition?
Facial recognition is a relatively new technology that has become very popular today. From social networking platforms that can recognize your face in pictures, to mobile phones that can unlock your device with facial recognition, they all use the same general technology.
Facial recognition technology is a biometric tool that analyzes the features of the user’s face and by matching the received information with the information in its database, it can recognize a person’s identity. Facial recognition software uses artificial intelligence and machine learning to scan faces and compare input data and information from their databases.
Although this technology was initially used mainly in government agencies to identify criminals, now consumer products are also equipped with it. The consumer market’s need for this technology has caused technology companies to be pioneers in the development of these technologies, and sometimes they have even imposed restrictions on the government’s use of their technologies.
How does facial recognition work?
The face recognition process can be briefly defined in three basic functions:
- diagnosis; It is the process of finding a face in an image. If you have used cameras, you must have seen how the cameras look for people’s faces in the picture frame and mark its circumference with a box to auto focus.
- analyze; It is the stage that maps the face. This is sometimes done by measuring the distance between the eyes, the shape of the chin, the distance between the nose and the mouth, and then converting this information into a string of numbers or points.
- ID; It is an attempt to verify a person’s identity from a picture. This process is used for authentication and this is where the proponents and opponents of the technology differ.
Now that you are aware of the general operation of this technology, let’s take a closer look at how facial recognition systems work. These systems must first learn what a face is. This can be done by training an algorithm on a large number of images of faces in known positions. Each time an image is fed to the algorithm, the software estimates where the person’s face is located in the image frame.
This neural network has an unfavorable performance at first, but after performing several exercises, it gets better and can finally find the position of the faces. The computer then learns how to tell the difference between one face and another, usually using a secondary neural network. Some algorithms map facial features directly, but others map their faces using more abstract features. Finally, the neural network obtains a vector for each face, which is actually a string of numbers that can uniquely identify a person’s face among other faces in the training sample.
In live systems, the software works on video clips in real time. In other words, the computer scans the video frames, which are usually taken from densely populated areas. Then it first detects the faces in each frame, then gets the vector for each one. Finally, this vector is compared with the data in the database. Any matches that meet the desired threshold are then ranked and displayed. This threshold in the UK police force is a 60% match, but it can be raised to lower the chance of misdiagnoses.
How accurate is face recognition?
The best systems in this area work very well. Independent tests conducted by the US National Institute of Standards and Technology (NIST) show that between 2014 and 2018, facial recognition systems became about 20 times better at finding people’s faces in a database of 12 million portrait images.
The error rate of these systems has also increased from 4% to 0.2% thanks to the development of deep neural networks. But this stunning performance depends largely on ideal conditions; That is, the person’s image must be clear and transparent and compared with a database of clear and transparent images.
In the real world, images can be blurry, poor quality, or dim. The person’s face may be at an angle to the camera or their age may have increased significantly compared to the image in the database. All these factors, along with the quality of facial recognition algorithms, can affect the performance of these systems.
A brief history of facial recognition technology
The roots of this technology appeared in the 1960s; When Woodrow Wilson Bledsoe developed a system for classifying images of faces. The system could compare unknown faces with photos in its database. This software was very different from today’s examples, but it followed almost the same idea.
By 1967, the US government was apparently interested in the technology, and there were whispers of investing in Bledsoe’s software to develop a program to match images. However, the results of Bledsoe’s efforts were never published.
During the 70s, 80s, and 90s, more and more systems tried to make facial recognition software. Some of these programs improved the ability of this technology to recognize the position of faces and then identify the characteristics of each face to become the basis for the leap of today’s modern technologies.
The first leap of facial recognition technology to enter the American mainstream coincided with a major controversy. In 2001, the United States police force used this technology to monitor the Super Bowl game, and the decision was met with widespread criticism. Critics believed that the police had violated the rights of citizens in the 4th Amendment of the US Constitution and had investigated people without justifiable reasons.
About a decade later, the power of computers reached a level where facial recognition systems could be trained with powerful neural networks, and the situation changed. As early as 2014, Facebook publicly announced that it was using DeepFace software to recognize users’ faces in photos.
This technology first entered consumer devices in 2015 with software such as Windows Hello and Trusted Face on Android. A little later, Apple brought this technology to the iPhone X in 2017, so that gradually most phones will be equipped with this feature.
Advantages and disadvantages of facial recognition systems
Proponents of this technology say that facial recognition has made many people’s jobs easier. These systems can be used to unlock phones and laptops, make payments, categorize images, pass through airport and stadium gates, and so on.
On the other hand, some other supporters of this technology point to the use of this capability to detect criminals. This issue can reduce the risk of terrorist incidents, especially in large and crowded events. Government agencies around the world have made great efforts to implement these methods in recent years, and China can be cited as the largest consumer of these systems to monitor citizens.
However, opponents of facial recognition say that the advantages of this technology do not outweigh its disadvantages. They believe that these systems can violate the privacy of citizens and that the people who control these softwares cannot be trusted.
On the other hand, the incidence of errors in facial recognition systems is still too high to rely on this technology. These systems still have a lot of errors and sometimes wrongly identify innocent people instead of guilty people or they can make huge errors with biases in the data they receive.
Critics also believe that there is no way for people to evaluate the performance of government facial recognition systems, and that these algorithms are often proprietary, and researchers are not allowed to verify their performance.
The future of facial recognition technology
Many technology companies are constantly optimizing their systems to improve their accuracy and speed up their operations. These companies are trying so that this technology can recognize users’ faces even in dark environments and from inappropriate angles.
However, experts believe that there should be clear and appropriate rules for the correct use of this technology. This issue, both in the corporate dimension, in the government dimension and in the personal dimension, should be specified with fair laws so that there is no possibility of its abuse and everyone can benefit from the benefits of these systems in the same way.