What is Facial Recognition?

Biometric systems with facial recognition are used to measure and analyze an individual’s physical or behavioral characteristics.These systems, which are basically based on using person-specific structures, can also use physiological features such as fingerprints, iris, palm prints and face, as well as behavioral features such as signature, walking style, speech patterns and facial dynamics.

Recently, the use of facial data in biometric systems has become one of the most studied areas.The diversity of usage areas as well as the reasons that no special permissions are needed and that it does not violate a person’s private space were also effective in this.So is this actually the case?

Which biometric feature is used?

Biometric systems are widely used for person recognition and verification in new generation technologies.One of the most basic biometric features used is face.Thanks to the human face, information about the person’s identity, age, gender, race, and emotional and mental characteristics can be obtained.The analysis of the human face and facial movements is considered an interdisciplinary research field that includes psychology, neuroscience and engineering.

Recognition systems, especially used in person tracking and finding applications, can be based on both the physical and dynamic features of the face.Digitally obtained facial images are processed by complex algorithms and compared with faces in existing databases, and within seconds, the identity of the person in the image can be learned.

Does it always give accurate results?

Of course, finding the result completely accurately is not that easy under all circumstances.Many different factors such as the angle of the face, lighting, age, race, facial expression, make-up, beard and facial accessories create obstacles to the effective functioning of recognition systems.Similarities between twins and relatives can also cause facial recognition systems to give erroneous results.

On the other hand, detecting faces can be accomplished successfully by humans in very short periods of time.The complex neural network in the human visual system is capable of processing the static and dynamic features of faces very quickly.However, when it comes to computers, this seemingly simple task becomes quite complex.

Cornerstones of Facial Recognition Technology

The history of facial recognition systems dates back to the 1950s.It is accepted that the foundations of automatic facial recognition systems were laid in the 1970s.Facial recognition, which was attempted to be carried out by using the distances between important parts of the face in the first studies, has reached a different dimension with technological developments.Although the classification of facial recognition systems made as a result of the developments between 1990 and 2000 is far from today’s classifications, the studies and developments in these years form the basis of today’s facial technologies.It is possible to list the important historical development stages of facial recognition technologies as follows.

• 1964: American researchers Woodrow Wilson Bledsoe and his colleagues developed a semi-automatic face recognition method that could query twenty variables, such as the person’s mouth and eye dimensions, with the computer programs they developed for face recognition.

• 1977: The facial recognition system was further improved by adding 21 variables (such as lip width, hair color).1988: Artificial intelligence began to be used in facial recognition processes.

• 1991: Alex Pentland and Matthew Turk of the Massachusetts Institute of Technology (MIT) introduced Eigenfaces, the first successful example of facial recognition technology, using statistical principal component analysis.

• 1998: Within the scope of the FERET Facial Recognition Technology Program managed by the Defense Advanced Research Projects Agency (DARPA), it shared its database consisting of 2400 images of 850 people with the world in order to accelerate studies on the subject.While the program provided an expanding database for use in research, it also offered researchers the opportunity to compare different facial recognition algorithms on a common basis.

• 2005: The Face Recognition Grand Challenge (FRGC) competition was organized to support existing facial recognition technologies used in the USA and to accelerate their development.

• 2011:Deep learning, a machine learning method based onartificial neural networks, opened a new dimension in facial recognition technologies.In this method, the computer itself selects the points to be compared in facial analysis, and the more images the computer is provided to analyze, the better the computer learns.

• 2014: “DeepFace”, a deep learning facial recognition system, was developed by Facebook researchers.It was reported that the system that recognizes faces in digital images reached the performance of the human eye with an accuracy rate of approximately 97%.On the one hand, developments in facial recognition technology continue, on the other hand, new areas of use of this technology are being added.Facial recognition methods are used in many areas such as banking transactions, money transfers, telephone and operator applications, and smart tracking systems.

Different Approaches for Better Matching

It is possible to consider face recognition as an identification, matching and verification problem.An unknown face is analyzed and compared with all identified faces in the database, and as a result, the system makes a decision.

In simpler terms, the system is faced with a matching problem.As a result of comparing the queried face with those in the database, the match is confirmed and the person’s identity is determined in this way, or the person is not recorded in the database and no match is made.

Facial recognition systems are divided into two main groups: image-based and video-based.While image-based systems try to identify the person using the instantaneous physical appearance, video-based systems also benefit from changes in appearance and the dynamic structure of the face.Image-based face recognition methods are divided into three main groups.View-based (holistic) methods, model-based methods, and texture (partial view)-based methods.

Video-based face recognition methods are basically evaluated in two main classes.These are set-based methods and turn-based methods.Set-based methods treat the frames of a video as a collection of images without paying attention to temporal order.Sequence-based methods, on the other hand, use images by preserving their temporal order.Therefore, the dynamics of the face over time also play a role in recognizing the person.

According to studies, the general classification of facial recognition systems can be made in this way, but the diversity of algorithms and methods used and the overlapping features of some methods make it very difficult to make a clear classification.

How Does Facial Recognition Work?

The operation of facial recognition systems basically takes place in six stages.

1. First, the image of the face is obtained through photographs or video.
2. Then, the security of the system is ensured by using the face anti-spoofing module and the elements that prevent face recognition are eliminated.
3. In the third step, reference points on the face are detected from the image or video frames.
4. Then, a series of pre-processes are applied to the image or video, in which a wide variety of operations such as image alignment, video frame selection, noise reduction, and contrast increase are performed.Before moving on to the final stage, all features of the relevant face are extracted from the image or video by using different face recognition methods.
5. In the last step, identification and verification is carried out by comparing faces in databases.

ETHICAL DISCUSSIONS

Many researchers and ethical review authorities working in the fields of computer science and artificial intelligence do not see any problem in using publicly available data in facial recognition research without anyone’s permission.

However, this situation has begun to change gradually today, and new steps are being taken to conduct research in a more prudent and ethical manner.Scientists‘ use of personal data in their research without any permission poses serious problems.Recently, it seems that the scientific community and local governments have different views on the ethical evaluation of recognition technology.Due to these differences of opinion, in the last two years, some universities and companies have retired the databases of facial photographs they used to improve facial recognition algorithms.Most of the photographs used in the research are collected on the internet.

There is no problem in the use of these publicly available images, and ethics committees do not impose any sanctions against such use.People whose photographs are included in research without permission are not happy with this situation.Facial recognition studies and related articles, especially those based on vulnerable populations in some parts of the world, are considered unethical by most researchers.

Attempts to use faces as a measure of items that are not scientifically or ethically approved (such as criminality) and to trigger discrimination are also condemned by most scientists.On the other hand, incorrect person identifications and erroneous results using facial recognition technology are analyzed and evaluated by scientists.

It is expressed on various platforms that the use of facial recognition technology should be carried out in a more transparent and full cooperation, and it is deemed necessary to clearly draw the ethical framework for the development and use of this technology.

Scientists need to question the morality of research and studies that collect and use large data sets containing various faces without people’s consent.Thanks to a growing number of responsible researchers, there are beginning to be re-evaluations of how to collect and distribute facial recognition datasets and other ethical issues.

Some institutes have started to take positive steps.Last year, some academic journals and an academic conference announced that they required facial recognition systems studies to comply with ethical criteria.

There is no guideline regarding the ethics of studies on facial recognition systems.The commercial goals of biometric technology companies that provide research funds and data sets to scientists often override ethical concerns.

Although it is considered very important for some scientists to express and discuss the ethics of facial recognition technology on different platforms, these initiatives alone are insufficient to take the necessary precautions.