Face Detection

We all have smartphones here, right? Wondrous little devices we bring with us everywhere that helps us do all manner of things. They entertain us, bring the world’s knowledge to our fingertips, and make recording our memorable moments as easy as clicking a haptic-enabled glass pane (or the button on the screen, whatever works). This last one is what is of interest here. Whether taking a selfie or a picture of friends, we’ve seen the little boxes that frame our faces, giving us insight into how the camera is thinking about the image it’s looking at and how it knows to automatically focus and adjust the light to get that perfect pic ready for Instasnaplinkredditweetbook. The technology behind this is called, wait for it, face detection! Not a particularly interesting or thoughtful title, but you know engineer types, always naming things logically rather than following in the long tradition of the English language naming things as nonsensically as possible. But how does this eponymous face detection technology actually work? Well it operates under a concept called object-class detection and uses Eigenface and Genetic algorithms to then positively identify an object as a face.

First is this concept of object-class detection. Here a class would be a type of thing to be detected, and the object would be a thing fitting into the class. For example in our face detection, the class would be the face, and the object would be your face. Well how does it do that? That doesn’t explain anything. That’s because this is where the Eigenface comes in, that being simply the application of eigenvalues to the human face. Eigenvalues are a fancy math word used to describe the concept of a vector that changes by a single factor when a transformation is applied to it. Fancy math words aside, what this comes down to in face detection is the analysis of low resolution grayscale images so that the computer can “see” faces. Remember, the computer thinks in ones and zeros, and anything you want the computer to “see” needs to be translated into that language. Which generally means math. Lots of math.

So we’ve got the spooky math face, now what? Well now the computer starts to determine if it actually is a face. It does this using what is called a Genetic Algorithm. Once again, engineers naming something sensibly, this is an algorithm that works the same way genetics and natural selection does. The computer understands what a face basically looks like, and is able to generate solutions to the problem which it then alters, compares against each other, eliminates and alters again until it has found one that fits the problem at hand, in the process identifying features like your irises or the corners of your lips. Once it’s identified your facial features and mapped them accordingly the computer is willing to say, tentatively, that you do in fact have a face. Congratulations!

Now for the human question. Why does the computer need to be so sure of a face? Despite the computer not caring at all, it is an excellent question to ask, because the answer is loads of things! Any spy-thriller or government expose documentary worth its salt has a scene of the crowd moving, a sea of anonymous people, all monitored and identified by some shadowy agency from a dark room filled with cigarette smoke. This isn’t a fantasy. Maybe the cigarettes in a government office building are, but the facial recognition cameras certainly aren’t. Their use has exploded in places like China, with a recent initiative by PRC police forces have seen augmented reality goggles linked in with their extensive facial recognition camera network so that street cops can have real time identification and information on anyone they see.

Of course we’ve already discussed how this technology is used in your phone’s camera, being used to automatically focus the camera and adjust light levels primarily, but what about other things? Possibly less sinister and tin-foil-hat-inducing? Perhaps we’ll pass over its use by a new generation of marketing executives to determine people’s race, ethnicity, gender, and other factors in order to give them more targeted ads. Best to gloss over its use for emotional inference as well, being used by politicians and the aforementioned marketing executives for surely dastardly purposes. Although to be completely fair this application can also be used to help those with autism, or even your average passerby. I’m sure we’ve all had a time where we could have our phones just tell us if the person we’re talking to is genuinely interested or perhaps falling asleep. Like you right now reading this article.

In short, face detection is used to enable further technologies, things that need to see and understand your face in order to figure something out generally. But what about future applications? We live in the encroaching cyberpunk future after all! Would be pretty bleak if we weren’t able to speculate about all the wondrous gadgets and gijamahoos being cooked up. Well, as anyone who’s purchased an HTC Vive or an Oculus Rift will tell you, “It’s not really there yet, but this is the future!” Augmented and Virtual Reality, or AR/VR. For pure virtual reality it doesn’t mean much, due to everything being a virtual environment anyways, but for augmented reality it has a lot of potential uses, particularly as AR becomes more common “irl” so to speak.

The possibilities really open up at that point. Using the aforementioned facial recognition and emotional inference capabilities to, say, easily know who everyone is at work or, like previously stated, help people with autism or social anxiety know what people around them are trying to convey non-vocally. You could have automatic links to your social media like linkedin or twitter pop up whenever somebody looks at you or even have some form of augmented reality fashion. If we have to live in a cyberpunk world might as well do it in style with an AR projected oni mask to show off with. These are just domestic consumer examples, the possibilities are just as expansive with applications in the defense or industrial world as well but we’ll let it rest before we loop back to Chinese cops having real time facial recognition on the street. Now you know about face detection. I hear the store has a sale on tinfoil.