What tools can be used in conjunction with dangerzone to get rid of watermarks?

[DeltaEpsilon19498]

Whistleblowers who want to retain their anonymity have to worry about watermarks, both visible and invisible, in the documents that they disclose to journalists. Dangerzone does not specifically remove watermarks. What tools and knowledge does the whistleblower need to have in order to remove these watermarks, visible and invisible? Alternatively, what tools do journalists need in order to remove the watermarks and protect the source?

[apyrgio]

That's a very cool topic actually, and I'd love if people have more data about the current state of watermarking. I can comment on a few things based on some research I did, but I really want to stand corrected.

Regarding physical watermarking, e.g., printer tracking dots, projects like this do exist, but I would not suggest them; at this point, the battle of removing them is lost.

(Added 2017) REMINDER: IT APPEARS LIKELY THAT ALL RECENT COMMERCIAL COLOR LASER PRINTERS PRINT SOME KIND OF FORENSIC TRACKING CODES, NOT NECESSARILY USING YELLOW DOTS. THIS IS TRUE WHETHER OR NOT THOSE CODES ARE VISIBLE TO THE EYE AND WHETHER OR NOT THE PRINTER MODELS ARE LISTED HERE. THIS ALSO INCLUDES THE PRINTERS THAT ARE LISTED HERE AS NOT PRODUCING YELLOW DOTS.

This list is no longer being updated.

-- from https://www.eff.org/pages/list-printers-which-do-or-do-not-display-tracking-dots

A lossy way I would maybe use to remove physical watermarks is to OCR the scans and copy the text to an empty document. In any case, I would consider the original document as not for publication, if journalists want to protect the source.

Regarding digital watermarking, things are even more tricky. Invisible Unicode chars, intentional typos, extraneous space characters, can all be used to watermark a document and associate it with a receiver. This is not a theoretical scenario, it has been used numerous times in the past (example). A very lossy way to counter watermarking here would be to translate the document from language A -> language B -> language A again, making sure to drop invisible characters and extra spaces along the road, but I wouldn't count on it 🫤 . In any case, if a whistleblower leaks a document that has been sent to them, I would consider it watermarked, and thus not for publication.

I had a talk on HotPETS 2024 where I mentioned the subject of watermarking, here are the slides for those who are interested.