Russian startup plans commercial launch of scarily accurate facial recognition tech

Kelly Hodgkins
Russian startup plans commercial launch of scarily accurate facial recognition tech
With 30 successful tests under its belt and 300 pending orders, NTechlab is ready to cash in on its scarily accurate facial recognition system. It plans to make its cloud-based system and software development kit available to clients worldwide.

NTechLab is only a year old, but the Russian startup is making headlines with its controversial facial recognition technology. The company rocketed to the top of this nascent industry when it beat Google in the “MegaFace” facial recognition competition held last year in Washington state. With 30 successful tests under its belt and 300 pending orders, the company is ready to take its facial recognition system to the world.

The company plans to make its cloud-based facial recognition system available to corporate, government, and law enforcement clients. This cloud-based service allows an entity to upload a database of photos and use it for facial recognition purposes. Later this year, NTechLab will release a software development kit for third-party developers and will roll out a factory security system powered by its facial recognition tech.

NTechLab sets itself apart from its competitors with its high level of accuracy and its ability to search an extensive database of photographs. At the MegaFace Championship, NTechLab achieved a 73 percent accuracy with a database of 1 million pictures. When the number dropped to 10,000 images, the system achieved a jaw-dropping accuracy of 95 percent.

Related: Big Brother is watching: New York to fight terrorism with a facial recognition system

“We are the first to learn how to efficiently handle large picture databases,” said NTechLab founder Artem Kukharenko to Intel iQ. “This advantage is the key to solving real-world problems, such as finding a criminal in real-time or identifying a regular customer from store surveillance cameras.”

The 20-person company was able to achieve these results using both deep learning and a neural network-based architecture. According to Kukharenko, the most difficult part of the process is the initial facial recognition, which is system-intensive. This step is the bottleneck in the process, and the team is working hard to improve the facial recognition algorithm so it can process a face at a faster rate and with fewer resources.

Once a face is scanned, the system builds a feature vector with 80 numbers that represent detailed information about the face. Each person has a unique collection of numbers that distinguishes each profile from other people. In this stage, the system also identifies features in the face that’ll remain constant even when a person ages, grows facial hair, or changes their appearance by wearing glasses or putting on a baseball cap. The final stage compiles this information and uses it to search for a match in the picture database. NTechLab built its architecture to scale efficiently, allowing it to increase the database size tenfold, while only slowing down the detection process by 1.5 times.

RelatedWall-mounted Lucy will control your smart home through voice, facial recognition

NTechLab has tested its technology on crowds, using it at an Australian amusement park and at the Alfa Future People Festival held this summer in Russia. At the music festival, concert attendees were invited to share their selfies with the NTechLab system. The system would scan the photos, store them, and use facial recognition technology to find the concertgoers in the crowd. When a match was identified, the system would send the matching concert photos directly to the person’s phone.

In another public demonstration of its facial recognition prowess, NTechLab earlier this year released FindFace, a free app described as Shazam for people. The app allowed users to snap a photo of someone on the street and submit it for identification. The facial recognition technology identified the unknown person by searching the Russian social network Vkontakte and matching the submitted photo with a person’s social network feed