Facial recognition software is widespread and has been for a while now. Although the software is known to be somewhat finicky (there was one unfortunate and highly publicized incident in which Google Photos tagged a young African American woman as a gorilla), they are getting more and more accurate and tend to be very helpful and widespread. You may be wondering exactly how engineers have created software that can recognize something as subtle and complicated as a unique human’s face. Here’s an article to tackle the fundamentals of the issue and help explain how we got where we are today.
Image may be NSFW.
Clik here to view.
If you’re wondering where the money came from, you won’t be surprised to hear that facial recognition software has both a purely economic market and serves government interests. There have been attempts to use cameras outfitted with the software to serve law enforcement purposes, though the attempt in 2001 by the Tampa Police Department to install such cameras in the Ybor City nightlife district was definitely a failure; people just made a game of putting on masks and making obscene gestures at the camera, to the point that the whole project was scrapped by 2003.
Efforts were made to incorporate the software into the security system at Boston’s Logan Airport as well, though it only worked with a 61.4 percent accuracy rate that eventually led airport officials to try other methods of enhancing security.
So what about the systems that do work? Facebook seems to have developed a pretty accurate system of automatically tagging friends in photos you just uploaded.
Most operational systems function by identifying nodal points in a face. These points are distinguishable landmarks that occur in approximately 80 different places and include the distance between the eyes, the width of the nose, the depth of the eye sockets, the shape of the cheekbones and the length of the jawline. These nodal points are measured with the sue of a numerical code called faceprint.
It wasn’t so hard for engineers to make programs that identified facial features among pictures that were all taken the same way in a controlled environment that had, for example, the same shot taken from the same angle with the same lighting. But in real life, photos of people vary widely in setting which means wide variances in face angle, brightness, contrast, etc. In order to make up for this, software writers had to come up with something called 3D facial recognition.
Image may be NSFW.
Clik here to view.
3D facial recognition uses a 3D model of a face to more accurately store distinctive features. The software looks at where rigid tissue and bone is most apparent, such as curves in the eye socket, which varies less over time and depending on facial angle. With this kind of modeling, a face can be recognized in its profile form after being input in a straight-ahead shot. It takes the alignment of the face into consideration when it is given its first example of the face to be recognized, allowing for more accurate results.
Surface texture analysis can also allow for faces to be recognized, though this is more common in security contexts than for Facebook.
