For each image, image registration involves 2 steps: finding the astrometric transform to the reference grid, and estimating the value of the image on this reference grid. The astrometric transform itself involves several independant steps. First a catalogue of the stars in the image is extracted using a very fast algorithm. This catalogue is then matched to the reference catalogue. A first association between the catalogues is made by looking for the offset. Once the association is done, an astrometric transform (a 2 dimensional polynomial of order 1) is fitted to the data. The method iterates, and performs rejection of the bad associations, until a proper astrometric transform of order N has been found.  This astrometric transform defines new grid points which lies between the points of the original image sampling grid. Thus an interpolation of the values in the original grid is required to estimate values of the image on the new reference grid. This interpolation is performed using bicubic splines functions.
 
 

The image registration procedure is fairly automated, all you have to do is to choose an image which will define your reference system, and to specify what is the order of the astrometric transform you want to use. For the reference image take a good image , with decent seeing and signal to noise.  Put all these parameters in the process_config file. You can then run the script by typing:

Once the image registration procedure is completed, all images have been re-sampled to the same reference grid. These interpolated images have the following names: interp_"original image name".  Of course, one image will be missing, the image you choose as a reference image. Just do a copy: cp "reference image" interp_"reference image".