Main content starts here.

Face-off: Facebook Photos and the Limits of Facial Detection

October 06, 2016

Many security firms and companies rely upon facial detection as a core component of their biometrics security solution. After all, if there's any feature we use on a daily basis to recognize one another—to distinguish colleague from competitor, friend from foe—it's the human face. A recently publicized security demonstration, however, has questioned confidence in such a security measure. Security researchers demonstrated that they were able to simply use photographs accessible via social media to hack popular facial detection authentication systems. This article takes a look at this example of the limits of facial recognition and describes the process of staying one step ahead of hackers through intelligent security research.

Security paradigms like biometrics, of which facial detection is an example, are consistently caught up in a game of cat-and-mouse. The mouse represents the hacker that seems to remain one step ahead of the security solution. In this case the cat represents the facial detection solution, which is in constant pursuit of the security-breaching mouse. (For more on this cat-and-mouse paradigm, see our new article on AI for Cyber Security .) So the task of security researchers then is to find potential weaknesses in security technologies before hackers have the chance to discover them first.

For facial detection, as a biometrics security technology, this dynamic is certainly applicable. For example, when it was discovered that a hacker could use still photographs to fool facial detection systems, security systems implemented ways of determining whether the image was moving. They looked for features such as blinking or moving, so-called "liveness cues," or ways of telling whether the image represented a real human being.1 After implementing such checks for "liveness," facial detection solutions were, once again, ahead of the game. Though not for long.

Recently, a group of computer scientists from the University of North Carolina (UNC) at Chapel Hill managed to trick prominent facial detection systems by mimicking liveness features with 3D animations.2 The researchers began by creating a 3D composite image of the face in question, which they would then animate and present to the facial detection algorithm. Remarkably, they were able to construct such 3D models by using images of the person in question found on Facebook.

This is alarming because many of us don't think twice before uploading personal images on social media sites that feature our face. The researchers could trick four out of five facial detection systems 55-58 percent of the time by using only a few publicly available Facebook images.3 Indeed, they noted that the Facebook photos were often limited in number or contained bad lighting. Not a problem, they write. "We leverage robust, publicly available 3D face reconstruction methods from the field of computer vision, and adapt these techniques to fit our needs. Once a credible synthetic model of a user is obtained, we then employ entry-level virtual reality displays to defeat the state of the art in liveness detection."4

Of course, this doesn't spell the end of facial detection technology, nor the need to refrain from posting personal Facebook photos. Rather, facial detection solutions will be required, as always, to stay ahead of the curve by improving robustness accordingly, and by improving their ability to distinguish between real and fake, 3D composite and human subjects. Furthermore, hybrid identification technology combining two or more modes of biometric authentication methodologies (i.e., facial detection and fingerprint, facial and finger veins) is being researched as a potentially effective solution to counter spoofing. 5 But let the work of the UNC computer scientists be seen as another lesson in how security must continually draw on research to do what it does best: keep us secure.

1 WIRED, Hackers Trick Facial-Recognition Logins With Photos From Facebook (What Else?)
2,4 USENIX, Virtual U: Defeating Face Liveness Detection by Building Virtual Models from Your Public Photos
5 NEC, Hybrid Finger Identification

Did you enjoy the contents?

Your comments and suggestions are welcome

  • *Please write messages in English. Thank you for your cooperation.
  • *Response to comments will not be made.
  • *All personal information will be handled according to our privacy policy.
Main content ends here.
Top of this page