In this entry: http://iloapp.thejll.com/blog/earthshine?Home&post=374 we considered a blind deconvolution method.
The author, Michael Hirsch has kindly been in touch with us and pointed out that the isoplanatic patch is small so using the OBD on instantaneous images from a typical stack, will yield speckle-like PSFs that are then smeared by the method. A smeared PSF will cause the concurrently deconvolved image to have too sharp edges! This is what we see.
We should instead either
a) use subimages the size of the isoplanatic patch and receive the PSF for that image and a deconvolved version of that image, or
b) generate a series of stack-average images and apply the OBD to that, receiving the time-average PSF and the deconvolved image.
Of course, we thought we were doing b), but should instead perform the time averaging before OBDing.
Doing a) would not solve our halo-problem as the isoplanatic patch is small and we are looking for info on 1x1 degree PSFs.
Testing to follow ...
Note that night JD 2456076 is the one with most (15) 100-image stacks in all filters. You could get 30 in a row if you combined IRCUT and VE1 ...