Facial recognition with Vision AI: How it works and business applications 

Face recognition has become an indispensable part of our lives. Many of us use phones that have a face recognition feature instead of password security. So, instead of repeatedly entering your password, you can just take a glance at your phone and it automatically unlocks. This is probably one of the most commonly seen instances of facial recognition technology. However, there are so many more uses for it.

If you’re a MasterCard holder, you would probably know about their selfie pay app that they launched in 2016. When it first came out, I was excited to see how they would incorporate facial recognition to facilitate contactless payments. In my opinion, they did an excellent job with the application. Customers would simply need to open the app to confirm a payment using their camera, and that’s it. Simple, fast, and hassle-free.

But it’s not just about the end-users. Technology like this comes as a huge boon to businesses as well. Today, facial recognition technology has garnered widespread popularity in business due to its ability to find applications in several key industries. There are several reasons for recent increased interest in face recognition, including rising public concern for security, the need for identity verification in the digital world, face analysis, and modeling techniques in multimedia data management and computer entertainment. Facial recognition applications can be seen in several industries like security, marketing, retail, health, etc.

Initially, when I developed an interest in face recognition, I assumed that it was a relatively new concept. Imagine my surprise when I realized that researchers have looked into automatic face recognition for decades and experiments with this technology have been conducted ever since the 1960s. Even back then, people saw immense potential with facial recognition.

A study published in June 2019 estimates that by 2024, the global facial recognition market would generate $7billion of revenue, supported by a compound annual growth rate (CAGR) of 16% over the period 2019-2024. These figures are a testament to the growth of this innovative technology.

But how much do we really know about face recognition? How does it work? And are these facial recognition systems always right or can they make mistakes too?

These are the questions that would constantly run through my mind, and I would like to share my perspective on the topic in this article.

So, let’s begin at the basics. What is facial recognition?

Facial recognition refers to the process of identification and verification of the identity of an individual using their facial features. Systems that possess a face recognition feature will be able to capture, analyze, and compare patterns based on the concerned individual’s facial structure.

The process of face detection is considered to be the most crucial step as it detects and pinpoints human faces in images and video footage. After this step, we can usually observe face capture, which transforms analog information (a face) into a set of digital information (data) based on the person’s facial features. Finally, the last step is the face match, which involves the verification of two faces that belong to the same person.

Applications of facial recognition systems

Facial recognition systems can see a wide variety of applications in several business sectors. A few of the most common applications are listed below: 

• Retail crime prevention: Face recognition systems can immediately identify shoplifters, criminals, or people with a criminal history as soon as they enter the premises. Images of these criminals can be matched with an existing database of criminals. This can ensure loss prevention and notify authorities in case of a threat. 
• Phone security: Many mobile phones are currently using face recognition technology as security measures to protect personal data. These systems can unlock phones by recognizing the face of the user.
• Missing person detection: Face recognition can be used to find missing children and victims of human trafficking. Provided that the missing individuals are added to a database, law enforcement authorities can be notified as soon as they are recognized by face recognition
• Disease diagnosis: Such systems can be used for disease diagnosis. Some diseases often cause minute changes in appearance. As an example, the National Human Genome Institute Research Institute uses face recognition to detect a rare disease called DiGeorge syndrome, in which there is a portion of the 22nd chromosome missing. Face recognition has helped diagnose the disease in 96% of cases. 
• Contactless attendance: Educational institutions have started using facial recognition technology to track the attendance of students and log them into an attendance sheet. This considerably reduces the time taken during roll calls. 
• Access control: Facial recognition can be used to ensure that only authorized individuals are able to enter secure areas like labs, boardrooms, bank vaults, training centers for athletes, and other sensitive locations.

How does it work?

In order to carry out facial recognition, various technologies exist, including 3-D, vascular and heat-pattern, and skin texture analysis. The most common type of facial recognition involves algorithms that identify certain points on the face, such as a person’s nose structure. These algorithms go on to create a template for that person.

The accuracy is said to be at its maximum when users voluntarily submit their images into a database, for applications like face access control. As soon as an individual comes in front of a facial scanner, their live image is captured and converted into a template, which is then compared to the templates stored in the database. A match allows the user to undertake some kind of activity, such as passing through a door or logging into a computer network.

In the case of facial biometrics, a 2D or 3D sensor will be able to capture a face. It then transforms it into digital data by applying an algorithm before comparing the image captured to those held in a database.

These automated systems can be used to identify or check the identity of individuals in just a few seconds based on their facial features: spacing of the eyes, bridge of the nose, the contour of the lips, ears, chin, etc. They can even do this in the middle of a crowded area and in dynamic and unstable environments.

Understanding the working of facial recognition systems has always intrigued me. Eventually, as I started reading more about it, I started conducting my own experiments with the technology as well. Most of the time, when I’ve worked with such systems, I tend to follow a few key steps that I zeroed in on through trial and error.

The first stage usually involves face detection in the concerned image, irrespective of scale, and location. Most of the time, I would make use of an advanced filtering procedure to distinguish locations that represent faces. I will then filter them with accurate classifiers. Throughout this process, I have observed that all translations, scaling, and rotational variations have to be dealt with in the face detection phase.

In the next step, an anthropometric data set‐based system predicts the approximate location of the principal features such as eyes, nose, and mouth. Of course, the whole procedure is repeated in cycles to predict the subfeatures, relative to principal features, and verified with collocation statistics to reject any mislocated features.

Generally, dedicated anchor points will be produced as the result of geometric combinations in the face image. After this, the actual process of recognition will begin. This will be carried out by finding a local representation of the facial appearance at each of the anchor points. The representation scheme depends on the approach, to a large extent.

From what I observed during deployment, the face recognition system worked very well under ideal conditions. However, under conditions of changing illumination, expression, resolution, distance, or aging, its performance would decrease significantly. Unfortunately, these systems don’t work very well while encountering any deviations from the ideal face image. I also found issues with effective storage and access granting to facial code stored as a set of features from an image or video. However, the most significant problem would be the amount of time taken by the system to complete the process of facial recognition.

These challenges inspired me to work harder on optimizing the applications that I have developed so far. With the outbreak of the COVID-19 pandemic, facial recognition applications like contactless attendance and mask detection are gaining a lot of attention, as there is no physical interaction required by the end-user. This calls for increased sophistication in the systems that we have today.