There is not an easy answer for this problem. Deconvolutions are applied isotropically over the entire image. That means, the PSF doesn't change. One wat that may be used to deal with those images is to divide the image, and dynamically change the PSF over those small sections.
How to do that in PI? Well, that workaround would be painfully slow and tedious. We have plans this year to research on this problem, and write one or two tools to deal with it. My first guess is to try to find some procedure in the phase angle of the FFT, that may help. I already have simulated some optical aberrations for individual stars. Now I need to simulate an entire frame, and that's the tricky job.
Maybe I have to completelly change my focus, and go on other approaches.
Anyway, we'll be on this problem. Don't expect results too soon.
