Steganography in video

[tormento]

I have read somewhere that majors send their prerelease video to dubbing studios with a steganography embedded to prevent leaks or, at least, punish the guilty one.

Has somebody some experience with it? I'd like to embed some sort of it, possibly resistant to reencoding, during my hevc processing.

[benwaggoner]

Traditional steganography presumed lossless encoding (like GIF or PNG), where data could be stuck in the least significant bits and masked with general dithering noise. Since that is the kind of low amplitude high frequency data that DCT-like (JPEG through VVC) compression used in all modern codecs tends to filter out, different approaches are required. This is generally called forensic watermarking, and there are a number of vendors offering solutions. I'm not aware of any open source or roll-your-own algorithms available for doing that, however. It's a really challenging and complex field, and gets all the more gnarly the deeper one dives.

Dubbing can use this sort of forensic watermark, but also generally gets burned-in text watermarks as well.

[Blue_MiSfit]

This is a very patent encumbered field that I am very familiar with :)

Some big names in this world:

Nagra
Verimatrix
Irdeto
ContentArmor

[tormento]

This is a very patent encumbered field that I am very familiar with :)
Nothing on counsumer level?

[benwaggoner]

Nothing on counsumer level?
Oh my goodness, no. No no no.

Practical implementations are along the lines of "pay six figures and assign a half dozen engineers to work on integration."

Marking some content so you know you made it, sure, as long as people aren't highly motivate to copy it while removing your mark. But marking some content so you can figure out which device or account that version of the content came from, in a way that doesn't impact quality and that is robust against motivated attempts to remove that mark. That is a whole lot. And it is in fundamental conflict with core tenants of compression, which is to remove psychovisually irrelevant details. It's making something people don't see but that is retained after an encoding process that is designed to remove things people can't see. Even if the content is scaled down and recompressed.

[MoSal]

I have read somewhere that majors send their prerelease video to dubbing studios with a steganography embedded to prevent leaks or, at least, punish the guilty one.

And all it would take is two parties to collude and compare versions for this magnificent expensive patent-encumbered technology to be beaten.

A single party getting their hands on more than one version would work two of course. No collusion necessary in that case!

[benwaggoner]

And all it would take is two parties to collude and compare versions for this magnificent expensive patent-encumbered technology to be beaten.
Collusion attacks seem to be a lot rarer than I'd imagined. The only real documented case I'm aware of was a large Middle East government breaking forensic watermarking from live satellite feeds. They were still caught using other means.

A single party getting their hands on more than one version would work two of course. No collusion necessary in that case!
In most FW models, the same title on the same device would always have the same watermarking sequence, so it's not as easy as just capturing it twice. Although a reader learned in the art has already figured out three ways around that.