Israeli tech innovations showcased at Skolkovo1484
Robot-lawyer made by Skolkovo startup raises $1 million in investment1153
Sberbank to build housing for its staff at Skolkovo784
Skolkovo council looks to future with plans for Technopark part II733
InSpace 3.0: Roscosmos, space startups hash out plans for joint success310
Israeli tech innovations showcased at Skolkovo0
Skolkovo council looks to future with plans for Technopark part II0
Robot-lawyer made by Skolkovo startup raises $1 million in investment0
Sberbank to build housing for its staff at Skolkovo0
InSpace 3.0: Roscosmos, space startups hash out plans for joint success0
Face recognition technology went definitively mainstream this week, with the unveiling of Apple’s new iPhone X, which is unlocked when it recognizes the owner’s face. But the technology makes many people decidedly nervous. Last week, The Guardian published a story about a computer algorithm that can purportedly determine in up to 81 percent of cases whether a person is gay, based on an image of their face. The ramifications of the technology for people’s right to privacy are very serious, and not just for gay people. Computers can differentiate between faces – including identical twins – far more reliably than humans can, and many people dread the prospect of a privacy-free, “Minority Report”-like future.
Skolkovo resident VisionLabs, which has been ranked one of the world’s top-ranking face recognition technology companies, is keenly aware that with its technology comes great responsibility. Last week, the startup announced it had undergone testing by Echelon, a Russian information security company that tests services for their adherence to data protection laws. Echelon found that VisionLabs systems, which store photos in an encrypted form as a selection of symbols, do not process personal data and are in full accordance with Russian law.
Alexander Khanin, CEO of VisionLabs, a resident startup of the Skolkovo Foundation's IT cluster. Photo: Sk.ru.
“All companies working on similar products face the task of proving that they do not keep the personal data of clients, and that the leaking of data via their systems is impossible,” says Alexander Khanin, CEO of VisionLabs.
Face recognition systems require as much data as possible to continue to improve. VisionLabs, whose slogan is “Machines Can See,” trains its Luna SDK engine on new data that it gets access to from partners, but which it doesn’t store, ensuring there is no breach of data protection laws.
“Most of our data is not public, but databases that we have gained access to during our projects in Africa, Kazakhstan and other places,” says Khanin. “It’s very important that the images are not saved together with personal data. We don’t risk our reputation, or those of our partners and clients.”
Despite the concerns over privacy, face recognition technology offers a wealth of possibilities that no one can ignore. The facial recognition market is expected to reach $6.84 billion in value by 2021, according to data compiled by the MarketsandMarkets research company: twice the $3.35 billion it was worth last year.
The technology is already in use in many sectors, from detecting fugitives and stolen passports at airports or carrying out ID checks in banks, to bringing up a customer’s previous purchase history in retail and compiling statistics for stores on the gender and approximate age of their customers. Several credit card companies and retailers have started testing face data as a form of biometric authorisation for payments.
VisionLabs' technology can be used as a high-security password on corporate computer systems. Photo: Sk.ru.
VisionLab’s Luna system is mostly used in the banking and retail sectors, at more than 25 companies in Russia, as well as in China, Southeast Asia, the U.S. and Europe. The startup, founded by Khanin in 2012, has gone from strength to strength, raising $5.5 million last year from Sistema VC, as well as being selected for the ChallengeUp! international accelerator and joining forces with Facebook and Google to create an open-source computer vision platform. In global ratings such as Labelled Faces in the Wild, compiled by the University of Massachusetts, and the Ongoing Face Recognition Vendor Test compiled by the U.S. National Institute of Standards and Technology (NIST), VisionLabs is consistently in the top five algorithms in terms of the quality of recognition.
VisionLabs’ technology is in use at Russia’s Pochta Bank, where computers can now only be logged into once they recognise the rightful user, and where the faces of loan applicants are scanned to determine whether they have previously applied for a loan under a different name or passport. The technology saved Pochta Bank from issuing 1.5 billion rubles ($26 million) in loans to fraudsters in just one year, and a further 3.5 million rubles in phone bills: staff are no longer required to send verification codes by SMS to change their passwords, as they are identified by their faces instead.
Although many people associate the technology primarily with security, VisionLabs says its focus is firmly on efficiency.
“We want to change the system of interaction between companies and their clients,” says Khanin. “We want to improve business efficiency using computer vision technology. We want to give robots eyes.”
Teaching robots tolerance
There are about 1,000 developers working on face recognition technology around the world right now, with more than 100 companies in China alone, according to Khanin. The country’s dominance is evident both in the widespread use of the technology there, and in the level of investment the industry is seeing: there are more than 10 companies in China that have raised more than $50 million each in investment.
“In China there is already a legislative base for the use of face recognition technology as biometric data,” says Khanin, adding that he expects similar legislation to be introduced in Russia next year. “There are already cash machines in China where you can use your bank card, a QR code on your phone or your face to get out money.”
The company's Face.DJ app creates a 3D model of the user's face from a single selfie. Photo: Sk.ru.
Khanin and his colleagues make regular visits to China to work on VisionLabs’ ongoing projects to integrate their systems there. For despite China’s strength, VisionLabs still has an advantage: many Chinese algorithms only work well on Chinese faces.
“There’s a test for how well systems recognise different races. NIST says clearly that the VisionLabs algorithm is consistently by far the best at recognizing faces on an international level. Our algorithm is the most tolerant,” laughs Khanin.
“The Chinese want their systems to work on a global level, so they’re interested in working with us, in integrating our algorithms into their systems: mainly into consumer devices, but also into access control and video management systems.”
VisionLab’s main market, alongside China, is North America, followed by Southeast Asia and Latin America. The European market is more difficult because of more stringent personal data laws there, though the company is in the process of opening an office in Amsterdam, where it will do research.
Of those 1,000 developers around the world, no more than 10 have honed their technology to be able to identify faces from a partial view, or in difficult conditions such as at an angle or in poor lighting, says Khanin. VisionLab’s Face.DJ app, in the meantime, allows users to create a photorealistic 3D model of their face simply by taking a single selfie.
VisionLabs is headquartered in Moscow, where about 50 people work, and is soon opening an Amsterdam office. Photo: Sk.ru.
VisionLabs has in fact been so successful in developing effective face recognition technology that it is already looking ahead to the next challenge. It may be ahead of the pack right now, but Khanin believes that soon recognition indexes will reach 100 percent.
“If you look at the face recognition industry, we believe in the next few years, maybe even sooner, the difference between systems will be insignificant,” he says. “You can make a good face recognition algorithm now in a weekend.”
In the scientific community, it is generally accepted that the task of face recognition has already been solved, says Khanin.
“But there is a big trend for the analysis of facial dynamics: how people react to questions, how they behave, are they nervous? We already know how to recognise faces. Now we’re moving into the field where the dynamic of the face is important,” he says.
Recently, VisionLabs teamed up with the Antirabstvo (Antislavery) project to develop a system that analyses people’s behaviour during job interviews and then tells them what mistakes they are making that could put off recruiters. Users can upload a video presentation of themselves to the service, and receive feedback within 24 hours.
Behaviour recognition technology also has great potential value for use by recruiters to eliminate candidates who display a nervous disposition but who are applying for a job requiring strong nerves, for example. As with face recognition technology, there are obvious uses for security purposes, such as at airports to detect people acting suspiciously. For VisionLabs, face analysis is a new line of business and technology that has the same value for the company as the core business of face recognition.
“Recognizing emotions is just a small part of the process of face analysis, and emotions without other indexes don’t have any value for business,” says Khanin. “We recognize behavior patterns that lead to a result for the HR and security businesses.”