differences in pixel value after rearranging steps

[falkben]

After rearranging the steps to move the corrections before summing (see: #61), we observe changes in pixel value in the uncertainties.

The following from separate conversation (quote: @mdecleir, w/ minor formatting):

For all frames (except the uncertainties), the differences in the pixel values are small. For the primary frame (which is the most important), differences are below 1-2%, which is within calibration uncertainties (3%).

For the uncertainties on the coincidence loss correction, larger differences are visible (several factors), but mostly in 1 direction: the new uncertainties are larger than the old ones. One possible explanation I can think of is that the uncertainties are larger because we now do the coiloss correction on every individual frame (i.e. a few hundreds of times), compared to doing it on the total frame (i.e. just once). We are thus calculating the uncertainties for the individual frames, and are then propagating them (through quadratic summing) when summing the images. If the uncertainty on performing this correction step is X% per frame, the total uncertainty will thus be the square root of the quadratic sum of all these uncertainties, while if we only perform this step once on the total frame, the uncertainty will be X% and thus smaller. This is, of course, assuming that the uncertainty is the same for every frame, which is not true, but conceptually, I could see why performing a correction step 100 times increases the uncertainty vs performing it only 1 time. Another possible explanation is that this method to calculate the uncertainty on the coincidence loss correction is simply not appropriate for individual frames, which have much lower SNRs and lots of pixels with zeroes. The same concern could be raised about the coincidence loss correction itself, and this is something that should be investigated in the future.