You could imagine a web app that serves subtly different versions of a page to each user. For example, it might encode the identity of the user in the images it serves. This would be a per-user watermark. On a private site, the consequence would be that if any image were leaked, it would be possible to identify the leaker from the watermark.

More sinisterly, the watermarks could be added using steganography, so the variants appear the same to the naked eye. Users would not be aware of the watermarks, and it could only be detected if two users collaborated to compare downloads (their files would have different hash sums).

Is there a name for this idea? Has anything been written about its feasibility or implications? Are there any examples of software that does it?

Edit: My example discussed images, but text documents could also be watermarked, perhaps by inserting invisible zero-width characters. With that method, the watermark would be lost if the document were printed and scanned. However, another steganographic algorithm might hide information in spelling mistakes, or by replacing words with synonyms. That kind of watermark would survive being printed and scanned, or even being transcribed by hand.

  • If I were a government hosting secret documents (say diplomatic cables) for 'internal use only', I would want to watermark them in this way. It wouldn't prevent leaks, but it would make it easy to identify the leaker. Jul 21, 2014 at 18:08

2 Answers 2


Good watermarks are meant to be invisible to the naked eye anyway, or "mostly". The point of the watermark is that it does not disturb too much the data contents, but the presence of the mark can still be detected. In many bank notes, the watermark is made visible to the naked eye because its presence must be detectable by a vendor who will use, indeed, his naked eyes. However, it is better if the mark can be somehow "hidden" so that it does not degrade the quality of that which is marked, especially when the marked element has some aesthetic value. In that sense, good watermarking, especially for pictures and videos in the digital age, should be "steganographic".

In fact, if it is interesting for anybody to remove the mark (e.g. to distribute copies which cannot be traced to him), then the mark ought to rely on steganographic techniques so that its presence is not detected.

Unfortunately (or not, depending on point of view), it is nigh impossible to do user-specific watermarking that cannot be removed. Indeed, if two users purchase the access right to the contents, a simple comparison (byte by byte, pixel by pixel, frame by frame...) will reveal where the user-specific watermark lurks, and allow removal. Watermarks have been studied a lot in order to enforce DRM; the rough summary is that it does not work. The hope was to be able to add a watermark that users cannot remove. Nobody found a method for that. At best, you may add some data whose presence the user does not even suspect, but the digital comparison between versions for two users destroyed that hope.

It used to work in the analog age; for instance, video-tapes containing copies of new movies, sent to critics one week before the theatrical release, were heavily tagged (in the analog signal) with an identifier for the target critic.

The only successful application that I can think of is the story of a teacher who voluntarily pushed (obviously) fake information in Wikipedia so as to trace within his pupils who were stupid enough to copy&paste from Wikipedia without even trying to understand what they were doing.

(The teacher and the powers-that-be at Wikipedia then engaged into acrimonious bickering and all failed to draw the proper conclusions. The teacher wanted to prove to his pupils that "the Internet" is unreliable; the Wikipedia people vindicated the teacher for "vandalism" but really for the capital crime of suggesting that Wikipedia is not the most reliable of sources. The only correct inference was that of the pupils: if you want to cheat, do it well. Don't forget that teachers are nasty cunning buggers.)


Individual watermarking by a webserver application can be done in a way that is invisible to the eye of the user downloading it.

If individual watermarking is applied, it will lead to different files having different hash-signatures.

But it can be done in a way, that makes an image unusable (meaning havily reduced in quality) if the file is downloaded two or more times and afterwords the different parts within this individually marked copies are deleted (i.e. by substraction) or overwriten.

Anyway - if only the server holds the original carrier and this original can be used laterly to extract the clients identifier, i guess it is possible to do such a watermark.

This would involve prior image-analysis, gaussfiltering/resharpening to produce a series of controlled artefacts, controlled pixelrelocation and random-noising beside the steganographic approach. Furthermore an error-correction algorithm to prevent fading away the user-identifier would be needed. That will lead to a point, where the same visual content is altered to large scale and is electronically unique for any user to an extend that the visual information is destroyed, if differences between copies are removed.

There might be an limit of the number of usable individually watermarked copies - some hundrets or thousands but not millions, i guess.

Anyway - it will allways be possible producing a copy without the electronically watermarks by screen-capturing, taking a photo of the screen or an imageprint or electronical image format conversion. Since an image will be resampled by the graphiccard, printer and by conversion software changing the (color information) format (i.e. from png (RGBA) to jpeg (YCrCb)) the watermark might be totally lost or largely degraded. This applies also to any electronic text-file: If something can be displayed by the user it can be converted&copied by the user. Though taking away the visual/textual information is always possible but this will perhaps come at the cost of content quality degration if we talk of high quality originals.

On the other hand, the lately stated methods will lead to copies where the source of the copy cannot be proved any more - which might be important to the client either (i.e. leaking information to the press) and therefore is omitted on client side. In this cases watermarking might be an option.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .