Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: After DRM, watermarking (Read 21139 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

After DRM, watermarking

Reply #25 –
Quote from: marcan on
Yes but with 50 copies, you can identify the ectropy, which is probably very close from the original and without a coherent watermark...
The entropy represents the watermark tho.


1. Your use of 'entropy' is not correct in this context, check the definition
of this word.

2. If you hope to rebuild an original file from the 50 versions, you're
pretty naive. A very obvious information which can be embedded
key-independently is the content provider from which you downloaded
your songs, and which will always be present. So you can manipulate
your 50 copies all you want, your combined version will still be identified
as coming from itunes or whatever. Further information can be embedded
and could survive the averaging (e.g. day of download), and can be used
to narrow down the search of the leakers.

3. Arguing on forums is fun and all, but if you're serious about learning
something about watermarking, you will get much more (correct) info in
the many research papers available on the internet.

After DRM, watermarking

Reply #26 –
Quote from: Woodinville on
Even if we have key-dependent methods, how would you prevent cross-interference at the modem level?

What you described is called a linear collusion attack, and there
are various ways to mitigate these. For a nice explanation you can
read this paper (where some old algorithms behave pretty badly) :
http://www.ipsi.fraunhofer.de/merit/forsch...fingerprint.pdf

Luckily, newer algorithms are more resistant  :
http://www.springerlink.com/index/0J68EU6FN588WBTY.pdf

Quote from: Woodinville on
Well, it seems to me that 'n' is the only question here. Enough sums and you'll have diluted each watermark signal, even if they are orthogonal.

Yeah, let's stop all research because it all comes down to that n value 
FYI the theoretical limit 'n' depends on the watermark size and is known
(http://www.cs.rutgers.edu/~jkilian/collected-papers/ErKiKu99.pdf),
but it is not the most immediate threat (for instance merging 50
copies is not a realistic attack scenario).

Quote from: Woodinville on
An interesting question is "will this affect how the music sounds".

Some papers mention that the average version has been reported to
sound better, but I wouldn't want to violate some TOS by giving links

After DRM, watermarking

Reply #27 –
Quote from: spath on
Yeah, let's stop all research because it all comes down to that n value 
FYI the theoretical limit 'n' depends on the watermark size and is known
(http://www.cs.rutgers.edu/~jkilian/collected-papers/ErKiKu99.pdf),
but it is not the most immediate threat (for instance merging 50
copies is not a realistic attack scenario).

Well, yes, it's an elementary exercise in modem design. 
Quote
Quote from: Woodinville on

An interesting question is "will this affect how the music sounds".

Some papers mention that the average version has been reported to
sound better, but I wouldn't want to violate some TOS by giving links


So, say, if the watermark was conveyed by phase modulation there's no chance here and there you'd get the phase modulation way, way out, and get some strange sounds?
-----
J. D. (jj) Johnston

After DRM, watermarking

Reply #28 –
Quote from: spath on
2. If you hope to rebuild an original file from the 50 versions, you're
pretty naive. A very obvious information which can be embedded
key-independently is the content provider from which you downloaded
your songs, and which will always be present. So you can manipulate
your 50 copies all you want, your combined version will still be identified
as coming from itunes or whatever. Further information can be embedded
and could survive the averaging (e.g. day of download), and can be used
to narrow down the search of the leakers.
Say you can make a statistical analyse of n different version in order to see what is different and what is not different. First you can see where the watermark is and whether his position is moving or not. If it's moving, you can easily create a version based on the parts that are the same. If it's not moving, you can identify where you need to concentrate your attack.
If you do this analyse on several releases, IMHO you will end up with a tool able to defeat a family of watermarks.

Quote from: spath on
3. Arguing on forums is fun and all, but if you're serious about learning
something about watermarking, you will get much more (correct) info in
the many research papers available on the internet.
Well, I have seen a lot of watermark developer claiming that their watermark is unbreakable but I didn't see a fair challenge on them… The only one I'm aware of was the SDMI contest and as the result they were all cracked pretty fast. But it was in 2001…
However I didn't any see any faire challenge since then…

Finally spath, you seem to know about the subject.
So what do you think about watermark?
Do you really think it's unbreakable?

After DRM, watermarking

Reply #29 –
How much info can be coded in simply 0-padding before the music content? Cant typical 1.gen lossy encoders be "reversed" to obtain approximately the sources first non-zero sample?

If not, could this serve the purpose of making "averaging" of several instances more difficult (how can one generally distinguish between low-level noise and low-level delay?).

Would listeners generally be able to hear a small pitch-shift ("FM-modulation") introduced to the signal, given that its symbolrate is ridiculously low? Given the audiophile interest in jitter, this threshold would be interesting nonetheless.


-k

After DRM, watermarking

Reply #30 –
Quote from: Woodinville on
So, say, if the watermark was conveyed by phase modulation there's no chance here and there you'd get the phase modulation way, way out, and get some strange sounds?

Hrm, I guess you didn't follow my second link.

"This article proposes a countermeasure against such collusion, providing imperceptible
phase shift in the individualization. Listening tests and computer simulations show that
the fingerprint can be detected correctly while the sound quality of the colluded signal is
effectively degraded."


Quote from: marcan on
Finally spath, you seem to know about the subject.
So what do you think about watermark?
Do you really think it's unbreakable?

I don't know, but the state of the art is definitely far ahead of what most
people on this board seem to believe it is at.

After DRM, watermarking

Reply #31 –
Quote from: spath on
"This article proposes a countermeasure against such collusion, providing imperceptible
phase shift in the individualization. Listening tests and computer simulations show that
the fingerprint can be detected correctly while the sound quality of the colluded signal is
effectively degraded."



Actually, I don't think you quite followed what I was suggesting, which is that by gross summation and averaging of many, many watermarked signals, it MUST be possible to disappear the watermark.

Either that, or Shannon was wrong.

The question, and bear in mind that it is in fact the paper I cited that proposes perceptually inaudible phase shift, is "can this be done without at least occasionally causing something to sound odd".

If you're arguing that collusion can not disappear a watermark, I want to know how you get around the issue that you're degrading each watermark by sqrt(n) (presuming orthogonality) for 'n' differently watermarked signals.
-----
J. D. (jj) Johnston

After DRM, watermarking

Reply #32 –
Quote from: Woodinville on
Suppose 50 people who all bought differently watermarked copies all get together and calculate 1/50 (sum of each sample) as a file?


It would be very surprising if a water mark could survive an attack like that.  (Might need more than 50 copies though....  or maybe not)  Bottom line: this would deter casual song swappers.  Professional bootleggers might simply be annoyed.
 

After DRM, watermarking

Reply #33 –
Quote from: Woodinville on
Actually, I don't think you quite followed what I was suggesting, which is that by
gross summation and averaging of many, many watermarked signals, it MUST
be possible to disappear the watermark.

If you don't understand what I say and at the same time refuse to read
all the papers I link to, there's no point continuing this discussion. The
answer is in the links.

After DRM, watermarking

Reply #34 –
Quote from: spath on
Quote from: Woodinville on

Actually, I don't think you quite followed what I was suggesting, which is that by
gross summation and averaging of many, many watermarked signals, it MUST
be possible to disappear the watermark.

If you don't understand what I say and at the same time refuse to read
all the papers I link to, there's no point continuing this discussion. The
answer is in the links.



Without having to read the links, one can directly expand your assertion into the idea that the Shannon Capacity Theorem is wrong.
-----
J. D. (jj) Johnston

After DRM, watermarking

Reply #35 –
Averaging 50 songs together will fail if as someone here suggested, there is some part of the watermark that is identical between songs, such as a distributor identifier. A watermark would be stronger if there was a "stationary part" and a "key dependent part".  But I'd like to not argue this much further until I have time to educate myself properly and actually know how these darned things work.

After DRM, watermarking

Reply #36 –
Quote from: NeonRain on
Averaging 50 songs together will fail if as someone here suggested, there is some part of the watermark that is identical between songs, such as a distributor identifier. A watermark would be stronger if there was a "stationary part" and a "key dependent part".  But I'd like to not argue this much further until I have time to educate myself properly and actually know how these darned things work.



Of course, if the watermark on all 50 is the same, nothing will happen.  (all of it, part of it, whatever, but if the part that is the same is part of an ECC, at some point everything will go down the tubes in many cases, but that's a more complex example than I think you're considering)


This does nothing to identify the 50 people who colluded, unless, by some miracle of chance, those 50 people are the only people to have gotten that watermark.

The assumptions I am operating under are

1) Independent, pseudo-orthogonal watermarks on each and every song.
2) An attempt to recover 1 or more of the watermarks.

The person telling me that I don't understand has an interesting conundrum here. Let us take 50 different watermarks made his way, and then try to recover them. According to his logic, we must be able to recover all 50 from the combined file.

I find that puzzling, to say the least.
-----
J. D. (jj) Johnston

After DRM, watermarking

Reply #37 –
Quote from: Woodinville on
Quote from: NeonRain on

Averaging 50 songs together will fail if as someone here suggested, there is some part of the watermark that is identical between songs, such as a distributor identifier. A watermark would be stronger if there was a "stationary part" and a "key dependent part".  But I'd like to not argue this much further until I have time to educate myself properly and actually know how these darned things work.

Of course, if the watermark on all 50 is the same, nothing will happen.  (all of it, part of it, whatever, but if the part that is the same is part of an ECC, at some point everything will go down the tubes in many cases, but that's a more complex example than I think you're considering)

This does nothing to identify the 50 people who colluded, unless, by some miracle of chance, those 50 people are the only people to have gotten that watermark.

The thread is about tracking single persons, or group of persons, who have bought tracks and then illegally spread them to other people. That "identical part" can only, as the given examples suggest, tell from where and  on which date(s) they were bought.

For the watermarks to be legally binding, your personal watermark has to be in it. The watermarks have all to survive the different applicable methods, or make the resulting output from those methods unlistenable. And it has to be proven that it's a watermark having survived eventual transcoding, and not some "random data" thereby introduced.

However, usually all these discussions about DRM / encryption / Watermarking are moot. Pirates don't try to crack copy protection schemes the way the studios want. Just look at DVD-Audio and Blu-ray/HD-DVD. They would probably just buy the tracks using stolen credit card information and not care about any watermarking.

After DRM, watermarking

Reply #38 –
Quote from: tommypeters on
However, usually all these discussions about DRM / encryption / Watermarking are moot. Pirates don't try to crack copy protection schemes the way the studios want. Just look at DVD-Audio and Blu-ray/HD-DVD. They would probably just buy the tracks using stolen credit card information and not care about any watermarking.


Agreed entirely.  People usually forget about "going around" schemes.
-----
J. D. (jj) Johnston

After DRM, watermarking

Reply #39 –
Would a pirate buy the same music 50 times anyway?

After DRM, watermarking

Reply #41 –
The problem for the pirate here is to do it on a large scale as it's done nowadays. So you can use a stolen credit card for few releases but on a regular basis?
However it's probably true that pirates will try to find a way that the industry didn't think about.