AI Can Recognize Your Face Even If You’re Pixelated

I would think that a 64 by 64 grid could be sufficient for pretty high level of confidence matching. And matching doesn't need to be perfect it often is far from it it needs to be good enough to provide a basis for elimination.

It might even work with considerably lower resolutions because with video pixels aren't static and they change so you get considerably more data than what you would get out of a single frame, once you account for things like lighting and video compression you can get quite a bit of data out of it.

This is not that dissimilar from analysing aerial or geospatial photography you can have features represented by a single pixel sometimes and you need to identify them with high enough confidence so you develop models based on how the pixels would "change" over time, or how would the effect nearby pixels at the end this what allows you to distinguish between a tank and an SUV or between a man holding a gun and a man holding a cat, these aren't perfect by any stretch of the imagination but we have put decades of work into this.

I would assume that you can both extrapolate directly and indirectly apply a lot of those methods onto facial recognition of pixelated images.