[FFmpeg-user] Watermarking vs. Lossy compression (Was: coding video for some old Sony)

[Moritz Barsnick]

On Tue, Aug 13, 2019 at 15:59:39 +0200, Henning Thielemann wrote:
> > "Real" watermarking should be resistent to this, upto the point that
> > you need to "ruin" the video (or audio, for that matter) to get rid of
> > it.
>
> Off topic: I always wondered how "real watermarking" is able to survive
> even lossy compression, but is still invisible. Wouldn't this mean that
> the lossy compression is not good enough? A compressor should be able to
> eliminate the invisible artifacts for more compression gain, shouldn't it?

Interesting point. You'll actually have to dive into the research
papers on the net. Obviously, those companies using watermarking have
put some thought and money into this. ;-) I guess basically you find
the weaknesses of the known algorithms/encoders and circumvent their
strength.

For example I quickly found a paper concentrating on "surviving" JPEG
compression, summarizing "The result shown [sic!?] that the embedded
watermark is robust to JPEG compression up to image quality 60 (~91%
compressed)." Obviously, this is still/single image and not motion
video in this case, but you get the point. In streams with a timescale
(i.e. audio and video), you can probably do stuff like e.g. a modulated
wobble in volume or brightness, which may be something the compression
algorithms don't consider to "optimize" away. OTOH, you can probably
craft an encoder which is specifically able to remove such watermark
effects, if it knows about them. A lot of waste of engineering energy,
better invested into e.g. better security or saving the planet, if you
ask me. ;-)

Cheers,
Moritz