This page shows a summary of the thresholding that would apply to each of the three
possible choices of input images to the thresholding step.

(1) uses a single dark frame,
(2) uses the average of 4 dark frames, and
(3) uses the absolute difference between the averaged dark frame
and a median-blurred copy of the average.

As we move from choices (1) to (3), we employ more processing steps, each designed to
reduce noise components in the dark frame. Averaging dark frames removes sources of random noise
(noise that is different from one dark frame to the next). The absolute differencing step
removes larger scale background components.

Brightness histograms are shown for each of the three input images. I used the
Transformation -> Stack Images dialog to allow me to show all three histograms at the same time.
The shift in the maximum of the distribution indicates we have removed some noise
as we process the dark frame(s). The threshold, for comparison, has been set to include
the 50 brightest pixels in each image. If we divide the threshold value by the
brightness value at the peak of the histogram, we obtain a measure of the
signal to noise ratio (where here our signal is hot pixels).

The extra processing has improved the signal to noise ratio (as defined here). But the
final thresholding operation ultimately provides us with pretty much the same colored image
of our hot pixels, no matter which route we take. If the simpler route works for your camera,
then take it. If you are interested in image calibration, are starting with a
lower quality dark frame, or want to also map out dead pixels, the more complete
processing route may work better.