Assuming an aligned set of images this finds the median value throughout the
set for each pixel, creating a new image from those median values. This can
be a better way of removing noise than simply averaging. For non-astro photos
it can remove moving objects, such as people, from a scene. This is a lengthy
process if the images will not all fit into memory. In such cases the images
are imagined cut into a grid so the whole sequence is loaded repeatedly in
order to process each rectangular section of the grid. So it is best to have
enough memory to hold all the images if possible.