The correction factor computed from each image is set to zero at
the positions of CR hits.
The contribution to the correction factor which is then missing
due to those zeros, , is filled with contributions
from all
images, where each image contributes proportional to
its local weight. If for example the PSF is undersampled, a
hole
in image 1 will closely resemble an otherwise flat
image that is zero at the affected pixels. In that case, the
missing contribution,
, will be filled only by
the other images which are not affected. On the other hand, for a well
sampled PSF, only little information is locally lost, but many
pixels are affected. Therefore, the weight of the not affected
images is only slightly increased relative to the affected
images, but over a large number of pixels. In other words, CR
hits do not only destroy information about the flux at the
position of the hit, but also some information for neighboring
pixels within the PSFs. This is accurately taken care of by this
procedure.