Bert,
By coincidence I was reading this today.
Its a technical note by Defence R&D Canada. (pdf format)
Accuracy of Dark Frame Subtraction
Its a method developed to extract faint signals of satellite smears below the normal Signal to Noise ratio of a CCD (below 3db !!) - so its of interest to the AstroPhotographers here.
http://pubs.drdc.gc.ca/PDFS/unc67/p528754.pdf
Whilst not intended to be used to measure the CCD temperature it may be able to be applied to this purpose although you would need some reference data to start with - the issue is the correlation between dark frames subtracted from each other having a temperature differential and the resulting residual gradient.
They are talking about 0.1°C differentials here and the fact that they can better determine CCD temperature than the the temperature readouts of the sensors attached to the chilled CCD cameras due to lags between actual real CCD temp and the measured temp from a sensor located on the chiller or elsewhere.
From Page 14 Point 3.
"The real temperature indicator can be retrieved by calculating the mean dark frame intensity value."
For the astrophotographers its a method to determine if your darks are actually at the same temperature (because they say, they are not always, despite what the camera says, within a range that is able to to resolve signal below this lower threshold.
Have fun.
Cheers
Rally