Ah,
deconvolutions, just the sort of thing that I loved to get my teeth into when I was in school..
We used to use fourier methods to deconvolve the spectrum of copper kalpha and kbeta xray radiation from the back-scatter signature from a rotating lithium crystal. I wish I had a computer as powerful as my mac back then.
Anyway, when two functions are combined the result is a convolution.
If one had the function of say, an AA filter, defined as a spatial function, then is is possible, to the extent that the defined function for the AA filter is accurate, and that information in the convolved function is not lost, to actually remove it.
We have the situation now, where some AA filters produce effects more than just low pass, due to the fact that they are not just perfect low-pass filters, they have other optical properties.
The theory is that those other optical properties can be removed from the image.
Actually, in fourier space, it is not all that difficult, and there are a number of algorithms coming to the fore many using other techniques. So far, none are perfect, but that is due to the loss of information in a low pass that filters at a higher frequency than twice nyquist, and that the function describing the other optical properties are not perfectly defined for every camera.
-bob