[altrego99]

In both cases you'd need to know some properties of the distribution to get a random bitstream.

Which pixel would you be looking at? Below what threshold would you call it a 0 rather than 1? How long till you can confidently say next reading will be uncorrelated?

I don't know how the astronomical imaging solved these (and TBH cosmic ray scanning seems a bit overkill to me too).

[greglindahl]

USB webcams get hit by cosmic rays, too. This paper is interesting because it's not just looking at the low bits of a normal image, but rather at at exceptional events. It's a different sort of random.

[eru]

Actually, you can get pretty far with randomness extractors with scant assumptions on the properties of the bitstream.