I don't think it would properly recover an image with a dark field with the exception of a single pixel with a value of 10^7 above the dark noise.

Then again it doesn't matter, because that's a contrived example (the photography equivalent of a "microbenchmark").