Vignetting correction fields have been made using the following recipe:
• Extract ~100 8x8 averaged plate images (2880x2880 image size) for the J band (f[123]??_avg.fits).
• Divide each image by the mode of the central 1/9th image (2 deg x 2 deg) I call the new images the cmode images since I use mode which is less likely to shift (compared to mean or median) given that the central 1/9th image is bound to have a few bright objects.
• Use imarith to stack the ~100 images to obtain their median. Leave out 10 faintest and 10 brightest pixels in each stack. The median image is the vignetting correction field. Throwing away 10% on either side should take care of changes in batches of plate leading to some odd members.
• While doing the above step, also generate the sigma images. They are good to have when you need them.
• The images need not be normalized at this stage since the callibration will come from CCD fields. Also, the central region for all three images have comparable mean/median and it will not really affect matters.
• Should one interpolate the correction at a point? An indication to the contrary.

The images are 33 MB each and are 2880x2880 and made from 8x8 average images:
J
F
N

The stats on the images is as follows:

```#               IMAGE      NPIX      MEAN    STDDEV       MIN       MAX
f_median_10.fits   8294400    0.8642    0.2974        0.     1.586
f_median_sig_10.fits   8294400   0.04982    0.1059        0.     1.695
j_median_10.fits   8294400    0.8626    0.3111        0.     1.015
j_median_sig_10.fits   8294400   0.05004    0.1088        0.     1.072
n_median_10.fits   8294400     0.793    0.3276        0.     1.023
n_median_sig_10.fits   8294400   0.07981    0.1083        0.    0.9137
```

The values at typical locations on these radially symmetric images are:

```  x    y    J	  F	N	d(J-F)	d(F-N)

100  100 0.789 0.754 0.638	0.085	0.312
150  150 0.829 0.790 0.760	0.090	0.072
260  260 0.885 0.846 0.785	0.084	0.140
500  500 0.972 0.958 0.877	0.027	0.165
700  700 0.997 0.988 0.933	0.017	0.107
1000 1000 1.000 0.998 0.979	0.004	0.036
1200 1200 1.002 1.004 0.999    -0.004	0.009
1440 1440 1.010 1.011 1.009    -0.002	0.004
```
The d() are the color changes at that point. Thus the g-r always changes by <0.1mag while the r-i can change by >0.3 mags. Thus when corrcetion is applied, all stars move up in the g-r Vs. r-i plot. I have below plots from certain blocks of the matched catalog m454.
• For the edge I have used (1,7), (7,1), (13,7) and (7,13) combined.
• For the corner I have used (1,1), (1,13), (13,1) and (13,13).
• For the center I have used (7-8, 7-8).
• I combine them into what I call "all".
Two rows for each type contain the uncorrected and corrected colors. Each row has three columns where column 1 has points for stars, column 2 has numbers in bins and column 3 has %numbers in bins. The correced corner should look like the corrected center and that does seem to be happeneing if you look at the last column.