Show profile
Link to this post
Subject: Invisible watermarking for images
Recently on Slashdot there was a discussion about a new watermark technique for protecting images. Basically, a modified checkerboard is applied to an image that messes up attempts to re-save the image. This would be used to make edits more obvious, but wouldn't necessarily prevent copying.
A blog post by one of the authors uses this example:
![[Image: http://nfgworld.com/grafx/WaterHash-1.jpg]](http://nfgworld.com/grafx/WaterHash-1.jpg "http://nfgworld.com/grafx/WaterHash-1.jpg")
It's not a particularly clever solution, basically they modify a pattern of pixels so that the colour is changed by a tiny amount. This subtle adjustment is not visible to the naked eye, but when the image is re-saved as a JPG it wreaks havoc with the compression system and becomes obvious.
They us a Google cache version of the image to demonstrate the effect:
![[Image: http://nfgworld.com/grafx/WaterHash-2.jpg]](http://nfgworld.com/grafx/WaterHash-2.jpg "http://nfgworld.com/grafx/WaterHash-2.jpg")
This is what the pattern looks like when you change the modified pixels so they're more easily seen. The checkerboard would normally have only two colours, but in this example had five: three additional colours were used:
![[Image: http://nfgworld.com/grafx/WaterHash-1.png]](http://nfgworld.com/grafx/WaterHash-1.png "http://nfgworld.com/grafx/WaterHash-1.png")
The problem with this technique is that it relies too heavily on very specific circumstances. I couldn't duplicate the effect with any images or paint programs.
Using either Paintshop Pro or MS Paint I was able to get a garbage pattern:
![[Image: http://nfgworld.com/grafx/WaterHash-2.png]](http://nfgworld.com/grafx/WaterHash-2.png "http://nfgworld.com/grafx/WaterHash-2.png")
Consider the next example using the FastStone Viewer to create the JPG file:
![[Image: http://nfgworld.com/grafx/WaterHash-1-FS.jpg]](http://nfgworld.com/grafx/WaterHash-1-FS.jpg "http://nfgworld.com/grafx/WaterHash-1-FS.jpg")
Now that might count as a success when you're trying to preserve unmodified checkerboards or screen doors, but the effect isn't at all consistent. Nothing I could do would get it to look the same as Google's cache. Using the Opera Turbo recompressor didn't work any better. Sure, the effect is sorta noticable, but how useful is it in the real world?
I couldn't make it work at all, using the hash pattern as an overlay on photographs or any other kind of art. Maybe I'm doing it wrong, but anecdotal evidence suggests I'm not alone in this.
So how good is a technique that only works on a very specific set of systems? Not very, IMO.
I should point out that, after reading the entire paper, it seems that the method of creating these images depends very much on applying a hash/mask that's tailored to the image. I did not do this. I need someone with a maths degree, stat. =D
Update: I think it's somewhat telling that the original authors didn't do this either. Their conclusion:
So yeah, the Checkerboard and Screen Door Protection System is in full effect, but more work is needed to apply the system to, you know, useful stuff.
Yawn.
A blog post by one of the authors uses this example:
It's not a particularly clever solution, basically they modify a pattern of pixels so that the colour is changed by a tiny amount. This subtle adjustment is not visible to the naked eye, but when the image is re-saved as a JPG it wreaks havoc with the compression system and becomes obvious.
They us a Google cache version of the image to demonstrate the effect:
This is what the pattern looks like when you change the modified pixels so they're more easily seen. The checkerboard would normally have only two colours, but in this example had five: three additional colours were used:
The problem with this technique is that it relies too heavily on very specific circumstances. I couldn't duplicate the effect with any images or paint programs.
Using either Paintshop Pro or MS Paint I was able to get a garbage pattern:
Consider the next example using the FastStone Viewer to create the JPG file:
Now that might count as a success when you're trying to preserve unmodified checkerboards or screen doors, but the effect isn't at all consistent. Nothing I could do would get it to look the same as Google's cache. Using the Opera Turbo recompressor didn't work any better. Sure, the effect is sorta noticable, but how useful is it in the real world?
I couldn't make it work at all, using the hash pattern as an overlay on photographs or any other kind of art. Maybe I'm doing it wrong, but anecdotal evidence suggests I'm not alone in this.
So how good is a technique that only works on a very specific set of systems? Not very, IMO.
I should point out that, after reading the entire paper, it seems that the method of creating these images depends very much on applying a hash/mask that's tailored to the image. I did not do this. I need someone with a maths degree, stat. =D
Update: I think it's somewhat telling that the original authors didn't do this either. Their conclusion:
The technique now needs to be extended to handle arbitrary photographs, not just uniform regions. More work is also needed to develop an untargeted JPEG marker that is not tied to a particular quantization table Q0, but results in a readable message with high probability over a range of quality factors.
So yeah, the Checkerboard and Screen Door Protection System is in full effect, but more work is needed to apply the system to, you know, useful stuff.
Yawn.
BLEARGH
This post was edited 2 times, last on 2011-02-09, 14:35 by