All imaging frames recorded by the camera include noise. When you add or subtract images, the noise always adds. Subtracting a single dark frame from an image will remove large pixel-to-pixel variations in the average accumulation of dark current. It will also increase the random noise in the image by 41%. If instead you averaged sixteen dark frames together prior to subtraction, the noise will only be increased by 10%.
The standard combine method is to average the frames. This produces the best results for purely random Gaussian noise. Unfortunately if there is an ”outlier” pixel on one frame (e.g. cosmic ray hit or a star in a twilight flat) then it will be included in the average.
Median combine is much more effective at suppressing outlier pixels. Unfortunately, median combine increases the noise level 25% compared to averaging. When median combining flat-field frames, renormalization is also required. This ensures that each frame is at the same average brightness (MaxIm DL does this automatically).
An alternative to median is to use Sigma Clipping or Standard Deviation Masking; these techniques throw out outlier pixels and then average the remaining. They are in effect a compromise between median and average, combining the noise reduction advantages of average with the outlier pixel rejection of median combine. The user can also select renormalization options for these methods.