Originally posted by jstevewhite .....
I'm curious, though, how a software package can do DFS, since it can't create a dark frame exposure post-hoc.
Assuming the sensor noise is random, not particular to any pixel or group of pixels, and follows some sort of mathematical distribution (normal, skewed to one side, whatever), has some sort of known relationship with exposure time (doesn't even have to be linear) , its possible to simulate the noise mathematically and create an equivalent DFS.
There are a wide range of electronic design situations where engineers have to simulate the effect of noise, either produced by the circuit itself or forced from outside, even before a single piece of hardware has been build. Software-tunable "noise generators" are now an essential Computer-Aided design tool.
The idea of using random noise to cancel out other random noise is not new, but only recently have processors been fast enough to put it into practice.
I think they have just adapted bits of such technology into Post-Processing programs.
To avoid complex number crunching, digital cameras take an actual shot with the shutter closed. Very old-fashioned if you think how powerful digital processors have become these days. Mathematically generated DFS is definitely possible, depends on cost and complexity for the in-camera hardware, really.