Quote:
Originally posted by roadtrip
You could write a program to "brand" the image,
as long as the webmaster didn't fuck with the orginal.
Take the 8th bit of the first 38 pixel's RGB values
and use them for the data storage. This will only
put the colour off on the first 38 pixels very slightly.
(38 for the scheme discribed, we could do better if
it was an ID referanced to a database.) A simple
program could check right on the submittion verification
as they do a thumb is really link to a pic check.
I have a program that brands an ID into all our
orginal content "somewhere" in the middle and in
a "pattern" that jpeg doesn't like to further compress.
98% of the time when I pull a pic of ours off the net
that has been borrowed, I can still run my verifier
across it and pull out the ID. It's only when the theif
messes with the brightness and contrast greatly that
I lose the ID tagging.
|
Very cool. There are a lot of watermarks that use the low order bits of the blue byte as the eye doesn't see much there but jpeg compression ruins most of those. Having a pattern in the middle that jpeg doesn't like to compress is great as it ought to be there even after a severe crop.
Once again, playing devil's advocate. The problem with a standard pattern would be that a thief (not your run of the mill thief, were talking a thief with a background in statistics here) could grab a lot of images that they knew were branded and search for patterns that were consistent across all of the images. This pattern would presumably be the identifying brand. Once the brand (if it is a constant) is isolated it would be a simple matter for the thief to write a script or program to remove/modify the pattern in the images. Chances are this caliber of thief won't be happy keeping such a program to himself and will attempt to distribute it on the net with a point and click interface for dummies.
What if the CRD ID or a unique password was used in the algorithm that defined the embeded pattern? That would make it a lot more difficult to gather reliable statistical data. The downside is that it would make it a lot more difficult for the automated process to check for the existence of this pattern.
Quote:
|
Very nice ideas.....but the problem is to get everybody to work with a system like this.....I really donīt see that happen.
|
Exactly. The infrastructure costs of expecting every TGP/adult search engine to change to a new system would be huge and unrealistic. However, if there was a way of reliably branding the content I don't believe that it would require the active participation of the TGPs/ search engines. A standard LWP bot could actually surf the net looking for potential violations.
I know, I know "the internet is huge, even google can't spider the entire internet with their 12000+ computers". The internet is huge but most of it is not very busy. Traffic on the internet, especially adult traffic looking for free xxx, seems to funnel into a relatively small collection of web sites (when I say small I mean maybe 1-100 thousand?? just guessing here). The 'copyright vortex effect'
. Sites that didn't get a lot of traffic really wouldn't be worth spidering but sites that got a lot of traffic would be. Joe Nobody steals some pics and makes an i-love-jenna website but he doesn't submit to the high traffic TGPs/lists and only gets a little bit of traffic. The bot never knows about Joe Nobody. But Fred Riches steals some pics and submits them all around town hoping to make a million overnight. The bot would probably find Fred.
roadtrip, if you're still reading this thread, do you mind me asking how robust your brand is to scaling, shearing and rotation? Ultimately if the image has to be significantly degraded to get rid of the brand then the image would no longer be usefull and everybody is happy except the thief.
Digimarc has some watermarking and digital id technologies but I don't believe that they withstand rotation and scaling very well.
cheers