In the Biometric Universe, Colour Matters

For most of the world, the idea of a “biometric state” is science fiction and often a Orwellian, but thanks to Aadhaar it’s already a reality in India. India’s facial-recognition system is larger and more ambitious than anything that has ever been attempted, and other countries where this technology is in its infancy are taking notice. It’s not an exaggeration to say that the success of Aadhaar will shape how the world thinks of facial recognition for years to come.

Of course, for any new technology to be effective, it has to work. That may be an obvious statement, but Aadhaar is so important and all-encompassing that there is no option for failure. And one of the key problems for any facial-recognition system is that they don’t work well for all skin tones. In a country as diverse as India, that could prove to be a real problem. The good news is that it’s an easy problem to fix.

The idea of racial bias in facial recognition has been getting significant attention in the US in the last year. In fact, many people are asking the question, “is facial recognition racist?” The answer is no - but it’s not really that simple. A number of solutions have been developed in the last few years, and their results have proven to be inaccurate when validating people with darker skin. That’s not only a technical problem, it’s also a social and political one. And it’s one that prestigious universities such as Massachusetts Institute of Technology (MIT) has noticed.

Theory versus Practice

The main problem is that most of the solutions being rolled out now by companies like Amazon were developed in a laboratory and have not optimized for the real world, as the ACLU has pointed out. In a lab, it’s easy to control for things like light, face samples and recognition settings, but on the streets things are a lot more complicated. But the reality is a lot different because skin colour doesn’t have to matter when it comes to facial recognition.

That may seem like a crazy thing to say, but hue is probably the least important factor when it comes to properly identifying a person. Things like distance between the eyes or the ratio of the forehead to the mouth are far more difficult to hack or depth of noxe and the contour of checks - and are therefore far more vital. In addition, lighting conditions are so unpredictable that most solutions that are in use today don’t work properly in bright sun. That’s a problem in a place like India.

That’s why the answer to many of the potential problems that Aadhaar may face lie in the three-dimensional world, not the universe of 2D imaging. Just about all of the common tricks that unscrupulous hackers use to fool biometric systems can only be done with flat images, and issues such as failure to recognise darker skin tones aren’t an issue with 3D technology. Like most facial-recognition systems in the world today (the Apple iPhone X being a notable exception), Aahdaar is built on 2D imaging technology. 3D recognition technology can even improve performance with existing 2D databases. Aadhaar is a massive undertaking, but it’s only going to be as good as its facial-recognition technology. In order to maximise the chance for success, the programme’s leaders are going to need to implement a tool that won’t discriminate based on skin tone. And it’s not as difficult as one might think.

George Brostoff is CEO of SensibleVision, which makes facial-recognition software. Brostoff has seven US patents and has developed technology used by Dell and other major companies. He can be reached at gbrostoff@sensiblevision.com.

George Brostoff
CEO, SensibleVision