Is there a technique for taking photographs which will defeat facial recognition?

Question

Is there an approach to photography that involves preventing facial recognition?

I'm looking for a term, or some professional way of saying it, because I'm looking for a professional photographer who can capture photos of people without them being used for facial recognition. In addition, I know a few friends who also want professional photos that also can't be mined, but none of us know what this is called, or there's even a term for it (thus, what do we ask to a photographer so that they understand).

Answer 1

Face detection works by detecting elements in the photo which are common to faces (eyes, mouth). Recognition works by comparing measurements of relationships of these elements across a database of known faces. (Usually, a matrix of measurements in reconstructed 3D space.)

You can defeat detection by either a) obscuring so much of the face that the computer can't tell it's there, or b) putting many confusing face-like elements in the photo so the computer can't tell what's real. (Fake eyes everywhere!) The second approach may still not be good enough, though, and is certainly not future-proof.

Recognition algorithms are very good. You can defeat them by not having the face detected in the first place, but after that, your best best would be to distort the face so that the measurements aren't accurate. You may be able to simply use makeup to "mask out" places where measurements are taken, so for example it's unclear where the edge of your mouth is exactly, or the distance between your eyes. But these are uncertain, and again, not future-proof.

Unfortunately, in order for this to succeed, the photo will have to be distorted enough that it can't look like the subject. As a general rule, any data processing task which can be done by a human can be done by a computer (see Watson, the trivia-game playing computer, for example). That means that if you can recognize the subject, a computer will be able to too — maybe not today, but certainly tomorrow. So, for the purposes of having a professional portrait, the answer is probably no, unless you don't mind wearing masks.

There is one possibility, though. That is: keep it analog. Have the picture taken with film, and don't scan it. Then, there's no chance of it being analyzed short of someone else digitizing it, and if they're doing that without your knowledge, you probably have other issues to worry about.

(Or, you could use digital files and keep them under tight control, but that's hard to do: people like to post to Facebook and other social networking sites, and if you do that, too late.)

Answer 2

I am the founder of http://TagMeNot.info, a visual privacy initiative (@dontTag on twitter).

There are at least three different kinds of automatic "face recognition" techniques: 1) face detection: spotting a face in the picture 2) face matching: the face is recognized to be the same face as in another different picture 3) face identification: the face being recognized is being associated to a person whose identity is known.

And there are also some "visual obfuscation" techniques being developed to preserve the visual privacy and anonymity of the target, for instance passers-by or protesters in risky environments, where their identification may lead to imprisonment or worse. Consider that identification or re-identification is possible even without having previously enrolled the target but using FaceBook pictures or face recognition search engines. Quite scaring are the "re-identification" experiments done by Alessandro Acquisti from CMU using public Facebook pictures.

One of these techniques is the automatic blurring of faces being detected. This, for instance, is what Google Street view does. I know Witness.org was developing with The Guardian a smartphone app to take pictures and store them only after faces were automatically blurred, but the SecureSmartCam initiative unfortunately seems discontinued. However here's the code for Android https://github.com/guardianproject/SecureSmartCam

You may perhaps be interested in some more resources on this topic: A page from TagMeNot on face recognition and privacy issues: http://tagmenot.info/content/face-recognition-and-privacy A paper of mine on face recognition issues and possible solutions: http://cammozzo.com/Papers/Cammozzo-Ethicomp2011-FaceRecognition.pdf