printer friendly version

Your face tells a story

There are many companies offering facial recognition today, most of them incorporating some form of artificial intelligence into their identification and authentication offerings. Unfortunately, facial recognition has not gained the best reputation over the years, although there are some success stories in controlled environments.

NEC XON held its seventh annual summit at Sun City in October and mixed in with the various discussions on Industry 4.0, safe cities and artificial intelligence (among other topics), facial recognition gathered more than a few mentions as a non-intrusive and reliable identification mechanism of the future.

Bertus Marais

Following the summit, Hi-Tech Security Solutions spoke to Bertus Marais, GM, public safety and security at NEC XON about facial recognition and NEC’s work in this regard.

When considering the previous lack of reliability of facial recognition, due in large part to issues such as uncontrolled lighting or people not looking directly at the camera, or wearing caps and other clothing that hides part of the face, Marais says environmental difficulties have always played a role and will continue to do so.

However, he adds that the algorithms behind facial recognition have improved so remarkably that things like the angle of a person’s face or the lighting make much less difference today. The camera hardware has also improved and continues to do so, but the biggest improvements have been in the algorithms that work behind the scenes to interpret the data they receive.

Indirect and old views

The issue of lighting and whether the subject is looking directly at the camera, or if he/she is wearing sunglasses or a cap (and even ageing) is no longer as big a hindrance to facial biometrics. Marais says NEC has made huge inroads in this regard.

“Performance will always be best if you have a fully visible and well-lit frontal image and an image pair within five to 10 years of each other, for example a passport and source that are maximum 10 years apart. However, the real world does not always work like that.

“In the real world people wear hats and glasses, their faces are often partially obscured, often at an angle and presented in various lighting conditions. The more the image of the face deviates from perfect, or the bigger the age gap becomes, the lower the match scores. A weak algorithm will deviate widely between good and poor quality images, while a strong algorithm’s accuracy will tail off, but with a gradual and predictable drop.”

When a face is enrolled into the NEC facial biometric system, he says the AI-enabled technology automatically simulates how the face would look, not only in ideal conditions, but also in a variety of lighting conditions and angles. Additionally, the technology does not need to see all of the face to get a good match.

Ageing does impact the match scores, but the tail-off is very predictable; we have many examples of a 30+ year age gap still yielding successful results. “As a general rule, if a human could identify someone from the facial picture, our algorithm is highly likely to as well; this usually translates to about 70% of the face being visible. Unlike a human, the algorithms can do this against many thousands of faces in real-time.”

Privacy and facial biometrics

If we assume that facial biometrics work well, the privacy issues of using it in public spaces needs to be addressed as there will be nothing to stop anyone from identifying you anywhere if there are no rules to manage the use of the technology.

Facial recognition offers a unique platform that can help in many areas of society, notes Marais, from providing benefits such as national security, law enforcement, and more. Other examples include speeding up and streamlining a traveller’s airport experience, to reducing fraud in the financial sector, to ensuring that aid reaches the intended recipient after a disaster, to ensuring that a self-registered problem gambler can get the help they need, all the way through to hospitality, VIP solutions and beyond. All of these areas offer a unique opportunity to provide benefits to society, but each has a very different context and use-case. These must be addresses with a privacy model appropriate to the use-case.

“The key to this question is, “What is the context of the use-case?” It is reasonable to assume that a national security agency acting on firm intelligence responding to a potential life-endangering scenario will take a very different view to a local retail outlet using the technology to provide a VIP customer experience,” states Marais. “Different use-cases will take different approaches to areas such as where the cameras are placed, what they are used for, who is captured, what records are stored, what records are enrolled, data retention periods, encryption and security, alerting mechanisms, audit, etc.

“As with all technologies, robust regulatory frameworks and policies should be encouraged to drive ethical and responsible adoption,” he continues. “The public, governments, corporations and customers have a role to play in this area. From a technological standpoint, the system is taking the exact same data as a regular CCTV camera; in fact, it is storing less information as most facial recognition systems tend not to store the entire video, but rather just the faces. The underlying point is that use of any facial recognition system should be subject to an appropriate level of control; this comes down to context of the use case and the governing frameworks.”

Facial biometrics closer to home?

Most of the concepts of facial recognition we hear about are in a safe-city scenario or being used in airports or public transport, but what about identity authentication in the workplace or even at home? Could your laptop’s webcam act as a touchless authentication mechanism to make online shopping or logging into your corporate network as simple as a glance?

Liveness detection is a crucial issue in this regard, adds Marais. Some specialised webcams have depth of field sensing, which is crucial to liveliness detection, but the technology is not widely adopted because it is expensive and not as mature as we would like for tasks such as identity-based access privileges to physical and virtual environments.

“Your average laptop camera or webcam is essentially a 2D view of the world. You can trick most of those into facial recognition with a good quality photograph or video of a person. The camera has no way of knowing if it’s looking at a live human being.

“Current high-end camera technology that gets a real 3D view of the world is far better and cannot be so easily tricked. But it’s expensive so it’s unusual to deploy it today. We prefer not to position facial biometrics for access control, but rather for surveillance.”

He continues, “we prefer to use a different, two-factor system for access control. For example, if the facial recognition system sees me walking past the restrooms it knows I’m not at my desk. It can check my laptop or desktop system and, if it’s still active, lock it down so nobody can access it. Or we use a fingerprint and a face check, or a normal magnetic access card combined with fingerprints or facial recognition.”

Real-world usage

Facial biometrics are being used worldwide for a number of applications, at ports of entry and exit they are used to scan hundreds of thousands of faces daily. They are used for public surveillance, such as is the case with an African authority that runs it across urban CCTV networks. They use it for safety and incident management. The same technology will be used for safety, monitoring, and management at the upcoming Olympics to be held in Japan. It’s already being used to help control hooliganism at soccer matches across Europe and it is being used for border control at several locations across Africa.

“NEC’s NeoFace Watch facial recognition software is now so advanced that it is scanning tens of thousands of faces in minutes, with many successful frames per face to record a match. That’s how fast it is,” Marais says. “But it obviously requires the right infrastructure to support it, which many cities and facilities now already have.

“You have to network all the cameras, feed the data to servers in a properly maintained data centre, analyse the huge volumes of data intelligently, return usable results with high accuracy for processing by humans so they can respond – and all of that in real time. It’s sophisticated, accurate, and dependable and organisations, from countries to cities, installations, structures and public venues can definitely use existing technologies such as CCTV and networks. They only add what’s missing from the total picture.”

Share this article:

Further reading: