to cause maximum quantization error on recompression at a chosen target JPEG quality factor
Neat, but pretty worthless in the domain explained though isn't it? Exposing re-compressing proxies? So if your float for re-compression matches their float for re-compression, you'll see a slightly blurry word!
Like someone commented on the site itself, I also can see a faint outline of the hidden message in the original (left) image. It changes in clarity if I change the angle I'm viewing my LCD screen at - most visible when I view it from below (which darkens the screen).
There was an article (of all places) on /b/ on explicitly how to manipulate compression filters to display 2 completely different images within the error domains that make the word "void" express. 1 or 2 people were actually publishing code in /b/'s conversation explaining how to repeat and why this happened.
However, they were using this algorithm to hide animated child porn and other just plain nasty stuff, but they were discussing it all right. If prodded, I could provide links. But I wont be held responsible for the disgust resulting.
(I hate the "markdown" used on HN, btw. Took me three iterations to figure out that I can't just write (asterisk)prod(asterisk) without prepending space.)
Context - in a lot of clients, you can type "/me did an action" and it will render as "pavel_lishin did an action". Other chat clients, though, just display the action surrounded by a pair of asterisks.
Hence (asterisk)prod(asterisk) would be me prodding him.
My impression was that it had to do with the gamma correction of PNG images. The 4chan thumbnail generator seems to ignore the gamma correction while the browser does not.
Not to the best of my knowledge; certainly, no such thing is running on my system.
The point is, most people would have absolutely no idea of how to even begin answering that question. If you teach them that 'VOID is Bad' and then your legitimate images have 'VOID' on them, it's going to cause problems and you're not going to be able to educate them well enough, fast enough.
Neat, but pretty worthless in the domain explained though isn't it? Exposing re-compressing proxies? So if your float for re-compression matches their float for re-compression, you'll see a slightly blurry word!