It was mentioned at another forum the possible use of watermarks for protecting meshes. This is a copy of my analysis about this subject: There is a big problem to where put the fingerprints, let us analyze the case of an *.obj file. You can divide the obj file in - The text - Vertices information - Texture vertices information - Normals information - Faces information The text (any extra thing and comments) and the normals information is not used by Poser and importing/exporting destroys all this information, so you can't put the fingerprint there. Remains - Vertices information - Texture vertices information - Faces information You can't put the fingerprints in the face information because it will create extra or wrong faces destroying your model. Remains - Vertices information - Texture vertices information If your remap the model all original texture information is lost, so you can't put the fingerprints there. So, only remains the vertex information for the watermark. If you put the watermark as extra vertices don't used by the model the watermark will be destroyed by any mesh cleaner software, so only you can put the fingerprint in the used vertices information. The watermark has to be able to survive to any order change of the vertex list. So the only way is to put the fingerprint is as a slight deformation of the vertex coordinates in a 3d space. But any morph can destroy completely this watermark because it cause severve deformation to the spatial information. You can't put the watermark in a specific location because if you cut those part of the mesh the watermark is destroyed. The only way is to put the watermark is in all the mesh, so if the you take a part of the mesh, unless is too small, the watermark is still present in that part, the same way as is done with pictures, but as in case of images watermark are easely destroyed and the mesh is subject to more severe manipulations than the image. Another big problem is than an image for example of 800x600 has 480000 points and a typical mesh can have less of 50000 vertices or even 3000 vertices, this number is very small for watermarks! An image of 200x200 has 40000 points and you can't put watermarks in small pictures, the normal jpeg compression process destroys the watermark! So, for small images with watermark you have to use bmp of tiff and never jpeg. Conclusion: The mesh watermark can be only be usefull for identifying models that have not been manipulated, watermarks have very limited application, the rest is illusion or propaganda of vendors. Remember that I kave not mentioned the use of softwares designed for removing watermarks, so if with normal manipulation the result is bad for fingerprints, using these softwares the result is tragic.

I remember Ockham mentioning an interesting approach to this problem last year in the same thread where PhoenixRising (Anton) approached this subject. It would be interesting to know whether any vendors here actually use any kinds of mesh-protection techniques and have they run tests as to whether they work.

you can 'watermark' meshes, and even morphs if you are patient. you can move vertices around to actually spell your initials if you so wish, or whatever you like. you can do it in as inconspicuous a place as you like - like between the toes for example. but while there are many ways to mark images and meshes, are they worth it? not really. its not like most modellers and texturers do not keep hundreds of megs of WIP images/meshes that the infringer cannot reproduce (especially unmerged materials, NURBS polysurfaces, etc). proving that a mesh or a texture is yours isnt much of an issue - the issue is having action taken on your evidence. as for recognising the mesh? hah! any modeller or texturer who has spent days slaving over a mesh/texture can recognise it in a heartbeat, even from a partial glimpse. i recognised the last 3 infringements of my products from the thumbnail in the marketplace.

IIRC, Anton's technique was to put a kind of spider-web of distinctive polygon links in one section. Distinctive enough so it could be recognized, but not so obvious (e.g. letters of the alphabet, words) that somebody with a vertex modeller would notice it and remove the extra polygon links. But if the rest of the mesh (99% or more) was identical to a pre-existing model, that would be enough proof of violation, so adding special polygons isn't necessary.

I haven't been able to successfully embed a watermark into a texture that would be recognized in a render, BTW. Anyone here familiar with exactly WHY Casual P4 and Nude P4 have identical meshes except for vertex order? Could there be a clue here?

"If you put the watermark as extra vertices don't used by the model the watermark will be destroyed by any mesh cleaner software." What exactly does this mean. I am busy (with Ockham) trying to find a way to include extra vertice information in a model that can be manipulated by Python, but what will clean that information? Poser? UVMapper? I need to know, before I do all the work and then realize that the extra vertices will be "cleaned" whenever someone uses the model in a particular way.

"Anyone here familiar with exactly WHY Casual P4 and Nude P4 have identical meshes except for vertex order?" Probably for the same reason Victoria 1 and pre-enhanced Supermodel Vickie have identical mesh, (except for shape), but different vertex order - poor technique! ;-) (I do it better, now, though!)

Numanoid: I can't answer which software removes unused vertices. I don't know all the functions of 3dsMax, Maya or Lightwave and the tons of existent plugins. I have only 3dsMax version 3.0, but a mesh cleaner is a very important tool, specialy if you import 3ds models, so it must exist somewhere.
For this reason I have written a software that is a mesh cleaner, today it's functions are:

• Weld identical vertices
• Delete unused vertices
• Split faces with more than four sides
• Remove or correct wrong faces (faces with only two sides or with duplicated vertices)
  Some day I shall add the fuctions
• Remove unused texture vertices (UVMapper does it)
• Remove duplicated faces
• Remove double sided faces
• Weld vertices that are almost the same but not equal.

Message edited on: 09/15/2004 00:50

Attached Link: http://www.renderosity.com/messages.ez?ForumID=12402&Form.ShowMessage=1927718

Thanks kawecki. You seem to know what you are doing, so will you have a look at this thread and see if you can help me. I am a bit stuck on this. I learn fast, but I can't find the information to learn from.

Attached Link: http://research.microsoft.com/~hoppe/

Well, some very smart people at Microsoft, IBM and other places seem to think it's a worthwhile pursuit. See Hoppe's paper on "Robust mesh watermarking." I don't even begin to understand the high level mathematics involved in "Multiresolution Wavelet Decomposition," "Mesh Spectral Analysis," etc. but apparently these guys do and apparently they think it's doable. Now as to how robust "robust" can be, once you start hacking an arm and a leg off here and making things unrecognizable, all bets are off. ANy watermarking scheme can be defeated if you alter the content sufficiently. I think the point is that if you have the skill and the time to make Vicky Un-Vicky, beyond DAZ's ability to prove she's theirs, why not make your own to begin with? Actually, of course, they only have to keep you in court beyond your ability to pay a good lawyer I imagine. No lock is going to stop every thief, but that's not what locks are for. Masochists and math majors can find a bibliography on 3D watermarks at: http://www.kki.yamanashi.ac.jp/~ohbuchi/research/3dwmbib.html

Thanks for the links. People loves protecting things, you can spend one month making a protection scheme for a software and someone cracks it in 15 minutes, better spend your month doing something useful!

Hehe I agree :D

Let's see 'em crack my scheme which uses a radioactive isotope to tag each polygon - ha!

lmckenzie: Actually, if you read Hoppe's paper, cropping (hacking off a limb) is not a substantial enough attack on a mesh to fail recognition with their technique. They chop a model of a chocolate bunny to demonstrate this point. Adding 45% noise (which makes a mesh look like a raisin), downsampling, smoothing et al still leave behind the residual watermark. Even adding a second watermark retains the original. While the word "fractal" doesn't appear in the article, it's plain the transformation is intrinsic enough to survive the most likely attacks available to 3D mesh editing programs (which by and large do not apply complex statistical analysis techniques in their transformation algorithms). What I find fascinating is that this research began in the early 90s and this paper was presented in 1999. And still no indication of patents prohibiting CL or DAZ from implementing it. Read Hoppe's SIGGRAPH presentation; I can't follow the math but I can understand the concept and why it's robust. Basically the signature is embedded all over the mesh at multiple scales. Registration and recognition ignore reordered vertices, so both P4 casual and P4 nude heads would be recognized.

No offense to anyone but as a practical matter... this is folly :-) !!!

unless you plan on spending all of your time
scouring 3D content websites and freebie sections
$$$buying$$ and downloading every mesh that "looks"
Like yours.
and then opening it in a modeling program to find your
hidden mark before you can pubicly make the accusation
that its your stolen work.

and then what?? ask the site host to remove that pillbox
hat model from his server??.

Good luck there :-)

legal action??? OK here in the U.S.
its going to cost you big time$$$$ to hire
an attorney to go after the mesh thieves
based on some "encrypted vertices ".
much more than you will ever recover should you actually
win your case.
and this will have no effect on P2P networks (KAZZA etc.)
unless you want to install them ,and their spyware and risk downloading
and inspecting EVERY file that claims to be a 3D model.

Message edited on: 09/15/2004 15:04

Thanks, duanemoody. It's been a year ao so since I first read it, during one of the great "mesh theft" furors. I didn't go back through it this time. Wolf359, you're right, the greatest difficulty lies in in the legal rather than the technical realm. Ultimately, this for the big content producers I think, and perhaps more at corporate theft than small fry. If you've got the bucks to do 'Toy Story,' etc., you can afford lawyers. Paramount, Disney, etc. can afford to go around looking for fan sites and T-shirts, they can certainly look for models if it came to that. For that matter, the RIAA certainly scared the heck out of some P2P folks as well. They won't catch everyone by any means but they'll probably catch some of the worst offenders, especially the ones who're making a profit. I'd be surprised if it's a matter of opening the mesh in a modeler to detect the watermark. I'd bet it's a software process, just like detecting a hidden 2D watermark in an image. T3 connection, Linux cluster, web and P2P spiders, keyword search, who knows what you can find?

I have a foolproof mesh water marking scheme. Make your mesh so ugly that no-one wants to steal it, lol. Then you can always hope that someone steals it and fixes it up for you.

True. I was commenting from the perspective of a renderosity merchant trying to catch mesh thieves. On the other hand Microsoft,despite its vast resources, routinely claims to be losing " billions" per year to software piracy in china so i dont see how watermarking your 3D models will be any real deterrent to dedicated thieves. when one considers the undeniable fact that digital software pirates will NEVER buy your product!!! but instead continue to find way to circumvent any protection scheme to get it FREE, I think my time is better spent aggresively marketing my products to those who DO actually pay for them.

China & places like it, where copyright has been pretty much a foreign concept are a big problem. They're being pressured by the US and other countries. As capitalism and private property grow there and the government wants a bigger share of the world economic pie, they'll be forced to reform. The counterfeit Nike's, Rolex's, Hollywood films and copies of Windows XP won't stop but Microsoft and others would be happy just to stop the big mafia style operations. Now, they're going to start marketing a cheap, slightly cut down version of XP in certain Asian countries as another tactic -every bit helps I suppose. For the individual, you're probably absolutely right, that you're better off devoting your time to other techniques. Rather than the emphasis on marketing though, I'd say perhaps invest in service and support for paying customers. Someone once before mentioned the idea os some kind of co-op or consortium. That might be the only practical way for small operations to have any clout against piracy. With enough members, you could at least pay a lawyer to send out cease and desist letters and share the tedious task of monitoring sites and tracking people down.

Given how things have gone here in the past, I'd think such a consortium might end up an association of members suing each other instead. And that makes SexualHarassmentPanda very sad.

Message edited on: 09/17/2004 04:19

"I'd think such a consortium might end up an association of members suing each other instead." Ha Ha, I'd almost bet on it! BTW, any chance of a "Faces of Passion" for Maya Doll?

After spending two years squeezing every possible drop of realism out of the default P4 figures and correcting the early mistakes in V1, it's wonderful being appreciated for the things that really matter. I'll look into it. After I'm done with The Girl.

Vox Populi Duane. It may not be the artist's favorite work or even necessarily his best, but it's the one that gets top billing in the obit. What could Mona Lisa's enigmatic smile be other than post-coital bliss?