The basic technical approach of the current DRM systems is sound, which is why they're a problem. High quality implementations would restrict people.
In terms of pattern recognition, you're not exactly doing facial recognitiion. The hard part would be putting a signal in there which was:
- distributed through the whole picture (so you can't just mask off the logo in the corner)
- sufficiently unobvious to humans threshold so that quality doesn't go down too far. Dunno about this one, apart from the fact that we have a lot of bandwidth here to hide a signal
- could survive basic disruption attempts by 'hackers' (colour filters etc)
It is a fairly tall order, but so is convincing the entire PC parts industry to put crypto in all their components against the interests of their customers.