4 ms·

It wouldn't really be a feasible solution for video, though. You'd have to re-encode the whole thing for every individual user, which would be incredibly ineffi

by Daiz 14y ago

It wouldn't really be a feasible solution for video, though. You'd have to re-encode the whole thing for every individual user, which would be incredibly inefficient. Even if you're showing the watermark in just a few places, it'd still require a damn lot of resources. There's also quite a few things out there for removing video watermarks[1], so unless the watermarks were incredibly obtrusive, they'd probably be quite easy to get rid of. And of course you couldn't really put the watermark outside the actual video stream either, since then the user could just strip it out. [1] Like various Avisynth plugins for example: http://avisynth.org/mediawiki/External_filters#Logo_Removal http://avisynth.org/mediawiki/External_filters#Logo_Removal

benologist 14y ago
Is it really true that they can't do it on the fly? Live streaming is creating the video on the fly completely isn't it?
- Daiz 14y ago
  In case of live streaming, the video is only encoded once and then distributed to a whole bunch of users. If you were doing unique watermarks, you'd have to encode an individual stream for every single user. Encoding is much more taxing than decoding. The end result would be that you'd need a metric crapton of additional processing power, and the quality would be worse since you'd have to use sufficiently fast encoding settings to be able to do it in realtime (as opposed to being able to use a lot more expensive options when the encoding is done just once and separately from the actual streaming process).
  - benologist 14y ago
    What about onlive, they stream(ed) cpu/gpu intensive games in real time on a one-to-one basis?
    - coldtea 14y ago
      No, they don't. You have the game assets downloaded already on your PC. They just stream control messages (move the camera there, that player hits the other player, etc). Real time gaming is not streaming video...
      - jbri 14y ago
        OnLive[1], which is what was being talked about, is literally streaming video to you of a game that's running in a datacenter somewhere. [1]: http://en.wikipedia.org/wiki/OnLive http://en.wikipedia.org/wiki/OnLive
- trotsky 14y ago
  no, they can do it. There is overlay support in current generation iptv support that can target scrolls and texture replacements at a resolution of less than a zipcode. Take that gear and apply a few static watermarks with timecodes and then swap the different streams at unique times for each viewer and you can end up with a stream thats uniquely identifiable. Motorola's (Google) new gear can do this, though the purpose is targeted advertising not watermarks. The truth is streaming DRM is about preventing your mother from giving a copy to the neighbors, almost all the content is already on P2P by the time it streams.
im3w1l 14y ago
You could split the stream into N chunks. Every time you watch the movie, M of them are selected and fudged just a little bit. Since M << N, this can be done in real time. The next user sees the same chunks you saw, except that some new chunks have been fudged also. Thus the movie slowly drifts away from the original. To counteract this, fudges are more likely to carry the chunk towards the original then away from it. This is what I came up with in 5 min, so there might be problems with the approach.
- derefr 14y ago
  Sounds like the steganographic equivalent of Y-chromosome genotyping.
wmf 14y ago
There's been some interesting research on compressed-domain forensic watermarking; check it out.
Silhouette 14y ago
You'd have to re-encode the whole thing for every individual user, which would be incredibly inefficient. It would be inefficient, but not necessarily prohibitively so if you're talking about the kind of business where piracy is a serious problem so there is real money to spend. There's also quite a few things out there for removing video watermarks If you're operating on the kind of level we're talking about, you don't just put in a visible watermark, you convert the entire file/stream to embed a unique forensic footprint tied to the account of whoever downloaded it. Such a footprint will stand up to just about any manipulation you can think of that doesn't reduce the quality of the data so much that the value is destroyed anyway, and you're probably not removing it unless you have a PhD or two in the field and you know someone's secret key. The limitation with any watermarking approach is that fundamentally it's still based on the honour system and obviously does nothing to prevent private copying between people who don't care about ripping someone off. It's not a technical limitation.

by Daiz 14y ago

benologist 14y ago

Is it really true that they can't do it on the fly? Live streaming is creating the video on the fly completely isn't it?

Daiz 14y ago
In case of live streaming, the video is only encoded once and then distributed to a whole bunch of users. If you were doing unique watermarks, you'd have to encode an individual stream for every single user. Encoding is much more taxing than decoding. The end result would be that you'd need a metric crapton of additional processing power, and the quality would be worse since you'd have to use sufficiently fast encoding settings to be able to do it in realtime (as opposed to being able to use a lot more expensive options when the encoding is done just once and separately from the actual streaming process).
- benologist 14y ago
  What about onlive, they stream(ed) cpu/gpu intensive games in real time on a one-to-one basis?
  - coldtea 14y ago
    No, they don't. You have the game assets downloaded already on your PC. They just stream control messages (move the camera there, that player hits the other player, etc). Real time gaming is not streaming video...
    - jbri 14y ago
      OnLive[1], which is what was being talked about, is literally streaming video to you of a game that's running in a datacenter somewhere. [1]: http://en.wikipedia.org/wiki/OnLive http://en.wikipedia.org/wiki/OnLive
trotsky 14y ago
no, they can do it. There is overlay support in current generation iptv support that can target scrolls and texture replacements at a resolution of less than a zipcode. Take that gear and apply a few static watermarks with timecodes and then swap the different streams at unique times for each viewer and you can end up with a stream thats uniquely identifiable. Motorola's (Google) new gear can do this, though the purpose is targeted advertising not watermarks. The truth is streaming DRM is about preventing your mother from giving a copy to the neighbors, almost all the content is already on P2P by the time it streams.

im3w1l 14y ago

You could split the stream into N chunks. Every time you watch the movie, M of them are selected and fudged just a little bit. Since M << N, this can be done in real time. The next user sees the same chunks you saw, except that some new chunks have been fudged also. Thus the movie slowly drifts away from the original. To counteract this, fudges are more likely to carry the chunk towards the original then away from it. This is what I came up with in 5 min, so there might be problems with the approach.

derefr 14y ago
Sounds like the steganographic equivalent of Y-chromosome genotyping.

wmf 14y ago

There's been some interesting research on compressed-domain forensic watermarking; check it out.

Silhouette 14y ago

You'd have to re-encode the whole thing for every individual user, which would be incredibly inefficient. It would be inefficient, but not necessarily prohibitively so if you're talking about the kind of business where piracy is a serious problem so there is real money to spend. There's also quite a few things out there for removing video watermarks If you're operating on the kind of level we're talking about, you don't just put in a visible watermark, you convert the entire file/stream to embed a unique forensic footprint tied to the account of whoever downloaded it. Such a footprint will stand up to just about any manipulation you can think of that doesn't reduce the quality of the data so much that the value is destroyed anyway, and you're probably not removing it unless you have a PhD or two in the field and you know someone's secret key. The limitation with any watermarking approach is that fundamentally it's still based on the honour system and obviously does nothing to prevent private copying between people who don't care about ripping someone off. It's not a technical limitation.