YouTube announces face blurring feature

Good idea -- YouTube announces a way to blur faces in videos. I hope they add a feature that puts Guy Fawkes masks on people, too.

Whether you want to share sensitive protest footage without exposing the faces of the activists involved, or share the winning point in your 8-year-old’s basketball game without broadcasting the children’s faces to the world, our face blurring technology is a first step towards providing visual anonymity for video on YouTube.

YouTube is proud to be a destination where people worldwide come to share their stories, including activists. Along with efforts like the Human Rights Channel and Citizentube that curate these voices, we hope that the new technologies we’re rolling out will facilitate the sharing of even more stories on our platform.

Face blurring: when footage requires anonymity



  1. This is technologically really interesting, and certainly has many applications to protect the identities of people who need protection.

    But on another level there is something awful about this. Can I look forward to videos with huge portions of the picture blurred to hide crowds of people? Why is “broadcasting the children’s faces to the world” so taboo?

    1. Presumably it’s their happy eight-year-old faces, not their impressive passing skills and mid-court game, that would be of interest in that hypothetical video. Another nadir of the security theater state has been reached.

    2. Ummm… there are parents with really good reasons not to broadcast their children’s faces.  Adoptive parents of children from damaged homes, for example – often the worst thing that could possibly happen to such children is for their original family to find them.

      On a more general level, you can make an ethical argument that it’s unwise to do anything irrevocable to someone else without informed adult consent.  And putting footage online is irrevocable.

  2. It’s probably not a good idea to upload sensitive, unblurred video to Google servers. Sure you’re given the option to “delete” the original after you’ve blurred those faces, but can we be certain that a simple subpoena isn’t able to quickly “undelete” that video?

    1. Nope. And even so, the meteorite trench of data you leave wherever you go online can also be used to your disadvantage, as all the ad networks have shown.

  3. I’ve seen people able to undo the “twirl” process in Photoshop.

    I wonder if it’d be possible to do the same with these blurs.  If people would upload images with faces blurred out for privacy/safety, but someone figures out the algorithm and can figure out who’s underneath.

    1.  The way that people undo the twirls is by twirling the opposite direction. No pixel values are changed, they’re just moved.

      Blurring on the other hand changes values in the image, so you can’t work backwards. It’s like how you can’t find more detail by zooming in on a digital image.

      1. But if the blur changes the pixels in a specific and predicatble way, can’t you theoretically find a way to undo those changes?  Kind of like breaking a code by figuring out the cipher.

        1. It can be specific and predictable without being reversible.

          Blurring results in a loss of finer details. Unlike what CSI would have you believe, you can’t just “enhance” those details back into existence.

          1. “It can be specific and predictable without being reversible.”

            Yeah, I figure.  I guess I’m just wondering if it is in this specific case.  I’d assume Google’s engineers are smart enough to think of this, but to paraphrase Bruce Schnier, anyone can make a system they themselves cannot break.

  4. Now if only we can cover up all the comments with trollfaces. “Herp derp” plugin just isn’t cutting it anymore.

  5. [Guess it wasn’t me after all, this should be a reply to Uthor]

    A blurred (or mosaicked) image carries less information than the original, there are a lot of potential sharp images that produce the same blurred output.

    I suppose moving images complicate matters a bit. You might be able to wring some extra information from the image by analysing the way a region of the image shifts colour when the face being blurred shifts a few pixels. With real-life movement also involving possible changes in pitch, roll, yaw or distance of the face, I don’t see this working very well.

    What you’re suggesting does (sometimes) work for things like credit card numbers that aren’t sufficiently blurred, because the underlying source material is limited and known. Just because a blurry six doesn’t look like a six any more, doesn’t mean a smart person can’t figure out that blurring an eight would give a different output.

  6. Many assume blurring cannot be easily reversed. If it’s designed using a one-way function like hashing, then it wouldn’t be practical. However, if it’s designed like encryption and the key is maintained, then it could be reversed. I guess what I’m saying is that you should assume they have the key. If something’s not trustworthy, then why trust it?!

    1. You could probably design a blur scheme that encodes some of the original data on a backchannel.  But it would be enormously hard to do, and to what end? 

      By their nature, blur algorithms tend to be one-way functions.  Image manipulation has common ground with encryption, but they’re not at all the same thing.

  7. (replying to Uthor): Not a problem in this case.  It’s actually really difficult to make a blur filter which is reversible; even the most naive approach results in permanent data loss.  Blurring, by it’s nature, deletes the detail that was originally there.

    1. yes. assuming that the authorities can’t get hold of the original, which you just uploaded to the youtube server.

  8. Good on Google for acknowledging the legitimacy of activism (along with the identity of “activist”).

Comments are closed.