Face recognition, 2017: where is technology up to?

In 1997, the film Face Off, raised the first ideas about the possibility of unlikely enemies wanting to literally steal our faces from us. Twenty years later we often hear people talking about unlocking their smartphone with an image of their face; even at Pycon Otto there is talk of face analysis and the great potential of this tool.

Seven years have passed since Microsoft published the much discussed Kinect function, connected with recognition, which raised various issues about its involuntary ability to ignore black people and face recognition technologies are growing and improving every day.

The idea of face recognition by machines came about in the 1960s, when the US government funded software that could find the typical features of a face (eyes, nose, lips, …). In the 1970s there was another step forwards, with the creation of a model comprising twenty-one points that represents the mathematical generalization of a face. From here we jump to the end of the 1980s, when more defined concepts were introduced, which are still used in current tools. Research performed by Kirby and Sirovich at the time helped to define Eigenfaces, i.e. the unique features of faces in a very limited storage space, enclosed within a single matrix.

It was precisely due to Eigenfaces that the automation of face recognition grew rapidly leading to the creation of many dedicated software applications from 1991 onwards.

The first use of this type of technology was in 2001 during Superbowl, for identifying dangerous subjects and the result led to the recognition of nineteen of them. Obviously, as this was a test, the data was not exploited, but it clearly showed the great potential of the tool.

To date there have been many practical uses of face recognition. The Facebook research team developed one of the most highly performing software applications which is used to suggest tags in the photos that we upload. This tool was also involved in the dismay generated on social media by the announcement of  Facezam, a sort of Shazam for faces, able to recognize the identity of a person starting from a few photos of them and confirmation within the media available on Facebook. The app fortunately turned out to be a big advertising campaign, but existing applications can instead boast Helping Faceless, which aims to look for missing children in India and that in 2015 helped to find over 2100.

Microsoft has taken giant steps forward since the big Kinect slide, with Windows Hello, a Windows 10 function that allows access to users through biometric factors such as digital fingerprints and, obviously, face recognition.

The open source development world also has an eye on Face Recognition, with OpenCV, a known software application mainly used by anyone approaching the use of biometric factors in their products, like OpenBR, which is also able to perform profiling related to age and gender, and OpenFace, among the highest precision level libraries according to LFWtests.

We are still a long way from the day that John Travolta will have to defend himself again against the engaging expresiveness of Nicolas Cage, but it is certainly the right time to consider the idea of a sequel or of investing in these technologies, as Antreem did for my dissertation project: Methods and technologies for face recognition.