Facial Recognition: How to Teach a Computer to Recognize Faces

While walking through a crowded train station, all manners of motion surround one such as changing destination signs, bustling travelers, flickering monitors, and blinking lights. In the middle of all these, one could notice a face that they recognize. Although this process seems natural, facial recognition is a system involving several stages and mechanisms that researchers are still trying to understand.

Moreover, while researchers continue to understand the way in which humans recognize each other, they are also trying to teach computers how to do it. The fact is that facial recognition proves to be more challenging than expected.

So, what is the reason for trying this, and how will it work, if it works?

A Convenient Truth

There are already several ways to identify people: DNA, fingerprint matching, retina, and iris scanning. Even voice recognition is in use and is highly common. However, humans do not use any of these techniques to recognize each other, and this can result in mistakes. Considering fingerprints, the level of detail usually used to take fingerprints, in fact, may look similar to a computer observer or human.

Certain more drastic errors have occurred as a result of depending on fingerprints as a reliable identity marker. DNA evidence, often considered to be flawless on cop shows, has led to very big problems in the United States justice system, creating confusion for jurors and sending the wrong people into jail. All systems are imperfect, but when people themselves cannot verify the results, there is a high risk that an identity system may have undetected flaws.

Victoria Station. In London, the Network Rail system covers 15 mainline stations with 1800 cameras that record continuously.

Victoria Station. In London, the Network Rail system covers 15 mainline stations with 1800 cameras that record continuously.

When compared to other methods, facial recognition has certain features for it to be recommended. First, it is convenient. Across much of the world, people do not cover their face completely, which makes face recognition largely applicable and relatively non-invasive. If it is carried out correctly, it does not even require the cooperation of a subject to gather their data.

However, in contrast to fingerprints and DNA, the face of a person is prone to changes. People get haircuts, age, and have a wide range of facial expressions that can change the shape of their face. Why is something easier not chosen? The fact is that when it comes to the vision system, faces have a unique role to play.

What’s in a Face?

In 1965, Alfred L. Yarbus, a Soviet scientist, proved what other scientists had long doubted: that the eye performs its vital function of seeing through motion. When an image is shown, the human eyes dart around on a quick fact-gathering task, filtering out the visual noise and focusing on the most important details—those that will define what (or who) the person is looking at.

If more time is given to look, the person’s attention does not turn to an unexplored corner, but spontaneously reinvestigates those elements that “allow the meaning of the picture to be obtained.” Mostly, the element which humans go back to is the face.

Yarbus tracked eye movement as he asked subjects to evaluate an image.

Yarbus tracked eye movement as he asked subjects to evaluate an image.

This is due to the fact that a lot of information can be collected from faces—the identity of the people seen and also other information such as age, gender, and how they are feeling. From childhood, humans look at facial expressions of others to get clues on how to react to strange situations. While looking at faces, similar visual patterns are followed, looking at (according to the order of importance) the eyes, mouth, and nose.

Although this task appears instinctive to humans as social animals, it was not until 2012 when Stanford University researchers could find a region of the brain responsible for recognizing faces exclusively—the fusiform gyrus.

When looking at faces, other parts of the brain are active initially, recognizing each feature as individual pieces, but it is the fusiform area that puts together the information comprehensively to understand the whole face. The link between these key features then aids in forming an essential “faceprint” in the mind based on a fragment of the available information.

As observed in mantis shrimp, it is possible to optimize the visual system to solve for both soft and hard constraints. The ability to limit attention to the decisive visual features of a face enables processing to take place more rapidly, as well as to overcome the biological bandwidth problem: human eyes can capture a greater amount of information than can actually be transmitted over the optic nerve.

The ganglion cells in the retina of the human eyes can form the equivalent of a 600-megapixel image; however, the nerve that connects to the retina can transmit only about one megapixel.

A face is detected using the brighter bridge of the nose and the darkness of the eyes. It can be learned from this video that although the detection method is not very advanced, computers are very fast, and hence, their speed can be used to identify faces in short time frames with a high degree of accuracy.

Making a Face

In order to develop face detection software, researchers were able to use what they had learned from the human biology to create innovative technology. All human faces share some properties in common, which can be used to confirm that ‘this is a face’ and then to find out which face it is from using an existing database.

Properties that identify a face are:

  • The eye region is darker than the upper-cheeks.
  • The forehead, nose, and cheeks are brighter than the eyes.

Properties that identify a specific face are:

  • Value-oriented gradients of pixel intensities.
  • Location and size of the following: eyes, mouth, and bridge of the nose.

This strategy, at its most basic, can be highly effective for static images using a significant database (see Facebook’s automatic face search and Google’s face-matching Arts & Culture app); however, things begin to crumble in more real-world applications such as surveillance.

In spite of its clear success in television shows and movies, rough camera footage usually has changes in angle, lighting, frame rate, image noise, resolution, and even variations in subjects from day to day. All of these make real-time recognition highly challenging, although new technologies are still being developed that may eventually end up justifying CSI detectives.

Maybe I was right all along.

Maybe I was right all along.

A Good-Looking Industry

Invention, research, and innovation will continue to emerge to overcome these challenges (and tomorrow’s challenges). Analysts predict that the global facial recognition market is estimated to grow from USD 4.05 Billion in 2017 to USD 7.76 Billion by 2022.

Companies are highly intrigued by the possibilities of facial recognition technologies, and global security concerns are driving interest in better biometric systems. As both sophisticated services and independent technologies, security organizations, retail corporations, governments, healthcare institutions, and militaries will be looking for ways to advance.

This information has been sourced, reviewed and adapted from materials provided by Teledyne DALSA.

For more information on this source, please visit Teledyne DALSA.

Tell Us What You Think

Do you have a review, update or anything you would like to add to this article?

Leave your feedback