[seasox]

That's an interesting question! Disclaimer: In my current research, I'm focusing on steganography on ML-based lossless channels. Not an export on image-based steganography.

As far as I know, image based steganography usually uses random noise added to a cover image and a preshared key to decide which noise is actually encoded information. Since printers and scanners are a lossy channel, the noise will deteriorate. So I think this is a hard problem -- this might be possible with some kind of error correcting code as used in QR codes and alike (or maybe even use QR codes as the cover channel)

[adg001]

Indeed! As far as I recall, the use of error correcting codes were explored in the context of digital watermarking at least since the ECRYPT - European Network of Excellence in Cryptology – in the early 2000s. However many notions of steganography have been broken ever since. Now we have Meteor which proposes a notion of steganography which is provably secure. And, although we know that what is provably secure is probably not, I was wondering whether this newer notion can incorporate some form of ECCs. Food for thought!