Average and median stacking.

[Jairo]

I know the median stacking is more resistant to spurious data like cosmic rays, but by the very definition of median, will it only produce integer or integer-and-a-half pixel values, right? Does this mean that the average stacking preserves much more data for a smoother histogram when adjusting levels for faint objects?

Thanks.

[ngc3314]

For unprocessed data, yes. If you've done a flat-field correction and the result is saved as floating-point, not necessarily integer values (but the digitization issue still applies to those values). If the gain of a CCD is enough to sample the noise (whatever combination of dark current, readout, and Poisson noise applies to an image), the problem is reduced because each integer samples an intensity range finer than real information in any single image. If there is a large number of images to stack, you can also improve the result by combining subsets with one and the whole group of subsets with the other.

A still better solution, not implemented in all software, is taking the mean with clipping beyond n standard deviations from the mean at each pixel location, often with an offset to follow changing sky brightness - for example, Iris and Mira have this option. This retains the precision of a mean and does a better targeted rejection of wild pixel values than a median.