To build this reference frame you need to stack a number of the best images. Usually this is the images with best seeing. Although other things needs also to considered, like: the sky background, the saturation level of the bright stars, the exposure time,...

Once you have chosen your set of frames, you can write this list of files to a file named: ref_list

This is already done for this example,  if you have a look at register2/ref_list, you will see that this file contains the name of the 10 best seeing frames.
 
 

Then the most simple procedure to build a composite reference is just to add the images you have selected in ref_list. To do that you have just to type:

./simple_ref.csh
 

The resulting image is: images2/ref.fits.

Of course the drawback of this very simple procedure is that if there defects in the individual of images (e.g. cosmics) they will accumulate in the stacking process, leading to an ugly reference images...To avoid this one must implement some method to reject the bad pixels. For instance one understand intuitively that taking something like the median of the pixel in the different images should be more satisfying. Although one must also consider that the images have different backgrounds and different seeing. Thus before we can compare the images some adjustment has to be made. One way to do that is to use the image subtraction method in order to transform each frame to the same seeing and same background level as a given image. This given image can be the best image, it can be selected by editing register2/process_config (keyword REF_STACK).  You can check process_config and see that nfile0.fits is selected, this is the best image. Once all the image are transformed in order to match the best image, the stack of pixel is filtered using a 3 sigma rejection from the median (it is as robust as the median but less noisy).You can run this procedure just by typing: (before you run the procedure save the previous image ref.fits by giving it another name).
 

./ref.csh

Compare the new ref.fits image to the previous one, the new image has better resolution. If some cosmics were present in the original images, they would disappear in this new reference.