The Best Facial Recognition Algorithms Today
It’s not easy getting people to listen to what you have to say. After years and years of providing the masses with free research, you just might start getting a decent sized following of regular readers. Or not. It just depends on how long you can get away with paying MBAs peanuts and convincing them “it’s good for their careers.”
When it finally happens that people are listening to what you have to say, you can then start doing clever things to get your readers to help promote your content – like “top X lists” – where you present the world with a list of startups that are supposedly doing something better than everyone else. The best way to produce these lists is to rely on someone else to create them. In case the list is blatantly wrong, you can just blame it on someone else. That’s why when we came across a Face Recognition Vendor Test by the National Institute of Standards and Technology (NIST), which talks about the best facial recognition algorithms out there, we thought it was worth a closer look.
Face Recognition Vendor Test (FRVT)
The first thing to note here is that this is an ongoing test which looks at facial recognition algorithms from a variety of different “vendors” – 39 at the moment – and then evaluates them against eight different image categories including things like:
- Visa images
- Mugshot images
- Selfie images
- Webcam images
- Child exploitation images
The second thing to note is that we are evaluating this ongoing study at this current point in time. That means these results will change over time, and if your darling algorithm is now number one on their list, we’re not going to update this article. This is simply our evaluation of this report at the present time, using the most current information provided by the NIST. For each of the eight dimensions, we looked at the best performing algorithms which gave us five vendor names. We then compared this to the overall leaderboard seen below:
So, what we ended up with was a consolidated list of 11 vendors. Let’s take a brief look at each of these vendors who – as of today – are doing better than a bunch of other vendors according to a very complex 277-page report published by the NIST – which is constantly being updated.
The Best Facial Recognition Algorithms Today
Founded in 2012, Shanghai startup Yitu Technology has taken in just over $355 million in funding from a long list of investors, which includes Sequoia Capital on the roster of investors. The company first came across our radar earlier this year when we wrote about 8 AI Startups Doing Computer Vision, half of which hailed from China. That’s no surprise when you consider that half of China’s biggest AI startups – at least the last time we looked – are involved in some form of biometric authentication (facial recognition or voice recognition). Just last week, the company showcased their cancer-detection AI algorithms that are being used in a large-scale cancer screening program, called “AI Map for Cancer Screening,” which recently launched in China.
Founded in 2012, Russian startup VisionLabs has taken in $5.5 million in funding so far to develop their facial recognition platform. The company first came across our radar in an article we published on The Top-10 Russian Artificial Intelligence Startups in which they managed to secure the top place in the list having taken in the most funding of any other AI startup in Russia at that time. Over the years, they’ve been partnering with major banks and corporations – like Google and Facebook – to provide their Luna biometric identification platform as a service. Some of the applications for their technology are quite innovative, like their facial identification solution that will “allow a vehicle to recognize its owner from a distance, confirm identity and unlock the doors as they get closer.”
Update 12/26/2018: VisionLabs reached out to us and said that they are a Dutch company that is headquartered in Amsterdam. We originally listed them as a Russian company based on outdated information in Crunchbase.
Founded way back in 1990 when Vanilla Ice was rocking the mic like a vandal, Lithuanian startup NEUROtechnology has taken in an undisclosed amount of funding to develop algorithms and software development products for biometric applications, computer-based vision, and object recognition. Over 3,000 system integrators, security companies, and hardware providers in more than 140 countries integrate NEUROtechnology algorithms into their products.
If all you people in ‘Murica think your country has problems with voting technology, try moving to DR Congo where NEUROtechnology recently completed a project. There, they deduplicated 46.5 million “multibiometric” voter records and found that 5.3 million votes (more than 10%) were duplicates and 900,000 votes were from underage individuals.
Founded in 2007, Idemia is a privately held French company with 14,000 employees and more than $3 billion in revenues from the sales of security and identity solutions. For a company most people have never heard of, they have some impressive stats. They’re the world’s largest supplier of biometric terminals, the number one issuer of U.S. drivers licenses, and they shipped 1.2 billion SIM cards in 2017 – a year in which they spent $200 million in R&D. The company’s facial recognition solution, IDEMIA Face Expert, has applications in law enforcement and authentication.
Founded in 2014, Silicon Valley startup Camvi Technologies is an “artificial intelligence company specializing in face recognition and biometrics technologies” which has taken in an undisclosed amount of funding in that pursuit. The company’s technology is remarkably fast, being able to search a database with a billion faces in one second. Their real-time video recognition system can identify multiple faces in live videos, even on a smartphone with no network connection. Their solution is being targeted toward all kinds of applications from healthcare to authentication. According to the company, the chance that a person shares the same facial biometric elements with you is 20 times lower than that of using a fingerprint.
There’s not a whole lot more we can say about Chinese startup Sensetime that we haven’t already said in our article on Facial Recognition in China with SenseTime. Since that article, Sensetime has now taken in an incredible $2.6 billion in funding. According to an article by Bloomberg last month, which said it was “the world’s most valuable AI startup” – maybe by funding taken in, but certainly not by valuation – about two-fifths of its revenue comes from government security contracts. The same article goes on to say that they’re not nearly as aggressive in pursuing government contracts as our next Chinese startup is.
Just a few months back, we wrote about the 8 Top-funded Facial Recognition Startups. First on that list was the aforementioned Sensetime, and second on the list was Megvii – another Chinese startup that’s taken in massive funding. Alibaba, Ant Financial, and Foxconn are just some of the giant companies that ponied up the $607 million in funding for Megvii. Their latest funding round is being used to push the company forward in the area of retail. That’s according to an article by Bloomberg which talks about how “Megvii’s software uses facial scans held in a Ministry of Public Security database drawn from legal identification files on about 1.3 billion Chinese.”
Update 05/8/19: Megvii has raised $750 million in new funding ahead of what Reuters says is a planned IPO in Hong Kong later this year. This brings the company’s total funding to $1.4 billion to date.
The Shenzhen Institute of Advanced Integration Technology (SIAIT) was established under the joint efforts of Chinese Academy of Sciences (CAS), the Chinese University of Hong Kong (CUHK), and the Shenzhen Municipal Government in 2006. They are the only academic institution on this list, though a number of other universities also submitted algorithms for evaluation. In a press release by the university touting their achievements, they talk about how “the performance of facial recognition technology has improved 80 percent compared to a year ago.”
Founded in 2015, Russian startup NTechLab has taken in $1.5 million in funding to develop a client-side facial recognition solution where no biometric data is transferred or stored by the company. They were also one of the companies that made our list of The Top-10 Russian Artificial Intelligence Startups . We were surprised to see that three Russian startups made our list of “the best facial recognition algorithms today,” the third being this next company.
Not a whole lot is known about Russian startup Vocord, aside from the fact that they took in an undisclosed amount of funding back in 2011. They offer a free facial recognition tool that tells you who you resemble more, mom or dad. (In the case where the result is neither, and you realize you’re adopted, they’ll provide you with free counseling – but only in Russian.) They have about 120 employees at work building “high-tech security systems” around their facial recognition algorithms.
Founded in 2013, San Francisco startup EverAI has taken in around $29 million in total funding to develop algorithms that they are training on an “ever-expanding private global dataset of 13 billion photos & videos.” They claim to be faster and more accurate than algorithms on offer from big corporates like Microsoft and Amazon, with the ability to detect additional factors like ethnicity and emotion. The company claims tens of millions of users in 95 different countries, and also made our list of the 8 Top-funded Facial Recognition Startups as of September this year.
The reason we put together these “top X lists” is not because the startups that get featured will share the article across social media (free exposure for us) and post it in their press sections (more site authority for us). We do them because it gives us a chance to highlight some of the success stories out there that may not make the headlines in other popular tech rags.
In this case, it was Camvi Technology that emailed us about the NIST report with the subtle intention of getting us to do free PR work for them (to be fair, they spent a lot of time helping us understand the convoluted report). We’re always excited to see smaller startups like Camvi scrapping to get ahead of well-funded startups like Sensetime, and it’s quite interesting to see who is developing the best facial recognition algorithms – at least as of today.