Facial recognition: separating the hype from the reality with SensibleVision CEO George Brostoff
SensibleVision CEO George Brostoff takes something of an unusual perspective on the maturation of facial recognition technology and the biometrics industry. His excitement about the potential of facial recognition to solve real-world problems is evident, but so is his concern that technology providers could shoot themselves, and each other, in the foot.
Brostoff feels that the industry has at times done itself a disservice by applying innovative technology to uses that it is not effective enough for, or promising more than it can deliver, which has warped public perception of what it is, and how it can be used.
“You have to separate out the difference between the hype and the reality,” Brostoff told Biometric Update in an interview.
Applications of facial recognition that have had an obvious benefit for end users, like the convenience of Apple’s Face ID, have helped make people more comfortable with the technology. The opposite is also true, however.
“It’s so important to give something to the end user, because otherwise it really hurts adoption,” Brostoff point out.
Brostoff’s involvement with the industry extends well before the founding of SensibleVision in 2005. Back in 2001, he relates, many in the industry were out to, as he says, “find Bin Laden in the stadium.” While facial biometrics may be effectively deployed in the fight against terrorism, some of the promises made when the technology was in its early years were not helpful to establishing a reputation as reliable and effective, and have only bolstered the arguments of privacy advocates and other critics of its use.
“You have to apply technology when it’s going to be successful, not just because you hope it will work, otherwise that creates a tremendous amount of backlash,” he says. “The U.S. government, the Canadian government, governments all around the world were all pursuing technologies with biometrics and other innovations, and the majority fell flat on their face after lots of money was spent on them. The technology wasn’t refined enough yet for the kinds of things they were trying to do. That can hurt the industry because it sets expectations that can’t be fulfilled.”
This means that industry players have a responsibility both to customers and themselves “to apply the right technology in the right way with the right expectations.”
When asked if that responsibility extends all the way to SensibleVision and other companies telling customers that they cannot do things, or that customers’ projects will not work according to their expectations, Brostoff answers simply: “Yes.”
Eventually he adds, “That doesn’t mean that the technology can’t do wonderful things. It’s a matter of putting it into the right places.”
As the technology continues to mature, the number of right places is expanding.
“We’ve been waiting for this technology to mature, from being very big and somewhat inaccurate, to where now we’re able to do real 3D face recognition, meaning the actual point cloud, the actual individual pixels which we call depth points in this case, essentially we’re able to use those to make up a template that allows us to do real face recognition with that. It allows us to have improved accuracy because we’re getting a lot more data. It’s not just X and Y, it’s also Z.”
Brostoff notes that when he started working with 3D imaging using active cameras, it was with Canesta, which was later acquired by Microsoft in 2009. The camera was the size of a table, he says. While the technology has come a long way since then, many smartphone manufacturers are taking advantage of the hype around advanced imaging by providing 3D cameras, but without applying 3D processing for facial recognition. By using both active cameras for 3D image capture and 3D processing, SensibleVision’s 3DSafe can accurately verify the identity of individuals who may not be facing directly toward the camera.
The company recently deployed its technology for a major transit system in an Asian country, where elderly people can provide a card with a facial image to prove their age to use the system without paying a fare. Another facial recognition solution had previously attempted to provide the same service but had been unable to verify people consistently enough due to the range of angles their faces were captured at.
No adjustment of the false acceptance rate (FAR) and false rejection rate (FRR) could make up for the limitations of the facial recognition technology being applied in the circumstances of the use case, according to Brostoff. “The curve just doesn’t fit properly.”
This kind of challenge is typical of legacy facial recognition approaches, he says, which often prove much less accurate outside of the development laboratory in real-world conditions.
“The other challenge that they have is that these tend to be outdoors, in places where there’s bright sunlight, or no light, or strong shadows, and this is all where our technology tends to excel,” he says.
2D technology also tends to be easily thrown off by makeup, which 3D systems are not challenged by the same way because of the modulation of IR light to measure depth.
The danger of applying technology to situations where it will not be effective is that consumers will generalize from failures or inconvenient interactions.
“Our industry tends to overdo it so that people tend to find reasons not to use solutions,” Brostoff warns. “They do that because we didn’t set the right expectations.”
Being realistic about the capabilities of facial recognition technology, and applying it where it will be most effective, then, is a matter of protecting the reputation of the company, and the technology in general.
“Properly applied, we can start solutions that offer a higher level of transparency to the user while solving the problems that the security people are looking to address,” Brostoff says. “That’s historically been the problem. By the time the technology gets implemented because it hasn’t necessarily been the right balance of technology. It wasn’t convenient to use, or it wasn’t accurate enough, and we have to balance that better.”
From access control for livestock companies to a solution in development for border control which could be integrated with smartphones and used to apply facial recognition inside dark vehicles, true 3D facial recognition expands on what can be expected of facial recognition technology, according to Brostoff. If those capabilities are utilized to “give something to the end users,” acceptance of the technology could increase rapidly, as it has with the introduction of Face ID. If companies using 2D facial recognition promise the same results, or companies offering 3D image processing promise miracles, industry growth could be significantly hampered.