Facial recognition isn’t a new technology. But it has become more and more relevant to businesses and government institutions thanks to recent advances in AI which has greatly sharpened its effectiveness. Facial recognition technology is a point of focus in our AI-powered future. The application of facial recognition ranges from detecting the possibility and reducing the risk of Covid-19 spread in public places to software that improves the identification process in airports, public buildings, and offices, in addition to in-person verification, improving trust and safety in every space.
Facial Recognition is already here!
Fintech, real-estate, and crypto companies have all been at the forefront of using facial recognition to complement other technologies used for identification and to reduce the need for manual reviews. Each of the large scale organizations belonging to these sectors may be closing contracts, deals or customer reviews in the thousands every day. They need facial recognition technology they can trust. That brings us to the discussion on racial bias.
Racial bias and the growth of face recognition
There has been an ongoing discourse on the racial bias involved in facial recognition technologies. While there is no clear evidence, some studies have shown that there is a bias against dark skinned females in particular with error rates 34% higher than for other groups. In fact some groups believe that facial recognition technology for surveillance threatens rights such as freedom of expression and association.
Though companies like IBM too had earlier raised an alarm when it came to AI-driven facial recognition only two years ago, AI has since then grown by leaps and bounds and today’s facial recognition systems are more reliable. The facial recognition market was valued at 3.8 billion U.S. dollars in 2020. The market is expected to grow rapidly, reaching 8.5 billion U.S. dollars by 2025.
The process of face recognition
Despite these so-called observed lesser risks to privacy, there is no denying the fact that facial recognition is absolutely essential to maintaining security at all levels in society. Added to this is the fact that today’s AI-powered facial recognition is more accurate, thanks to complex mathematical representations and processes that assist it in comparing facial features to stored data.
So it is clear that facial recognition involves inspection of facial features of a person. Now this happens through a three-step process. Capture, Analyze, Compare, for any person, irrespective of race or gender. In the first step of facial recognition, reliable video or imaging devices capture the image of the person. The second stage then transforms the analog information in a face, the shape of the nose, the depth of the eyes, the shape of the lips and the presence or absence of a beard etc. into a set of data and vectors, to distinguish between them. The facial features are also assigned coordinates marking their relative closeness or distance from one another. In the last step, the given face is compared against a set of faces in the repository derived from government IDs or against a selfie.
How HyperVerge overcomes racial bias
Given the advanced AI that drives modern facial recognition technology, it is also not surprising that nearly 18 of the 24 surveyed federal agencies use facial recognition technology across the United States. Racial bias is still a problem with some facial recognition engines and scientists are working hard to resolve the problem. HyperVerge facial recognition technology is NIST certified for facial recognition, which means the racial bias, if any, is below 1%. It is also iBETA certified for passive liveness detection for facial samples across a diverse group of people.
HyperVerge has worked with clients in diverse geographies, and so has a unique advantage when it comes to facial recognition, that of having been trained on facial samples belonging to several ethnic groups. Because of the geographies HyperVerge is present in, its AI-powered facial recognition system can capture all the variations that one can think of.
Once HyperVerge moved from India to ASEAN, ethnicity variations came in and HyperVerge had to extend the model. But the organization has been cognizant and has developed a flexible architecture that easily also integrates into client or other third party systems with ease from the data and modeling point of view.
And continues to overcome racial bias..
Also, a lot of data is coming in from several parts of the world, irrespective of the person’s race or gender. Such data augmentation ensures that the bias is low in facial recognition. In addition to this, HyperVerge also continuously benchmarks against competitors to improve the accuracy of its own AI models, which translates to better identification. Today, this has helped us make the model neutral and well rounded. In fact, the HyperVerge facial recognition model is so trustworthy that manual reviews are not required in as many as 95% of the cases, even when the subject is in poor lighting or the image capture happens over a poor connection.
The future is digital, and there is no alternative but for organizations to follow suit. Businesses everywhere are using facial recognition for onboarding customers at scale. But racial bias can be an impediment to growth for many of these companies as they have to comply with government regulations and improve the customer experience at the same time. While a digital KYC promises a better experience, it must also be backed by real-time analytics which weed out bias and the possibility of fraud. HyperVerge hopes to bridge the existing gaps in technology at a cost that is affordable to businesses. Talk to us here to know more!