On 23/03/2013 08:54, Alfred Molon wrote:
In article , Martin Brown says...
If you can characterise the exact point spread function *and* it in
invariant across the field of view then astronomical deconvolution codes
will do a better job but at a cost of some artefacts.

Has anybody done this for a not so sharp lens or a slightly defocused
image?

Pretty much routine these days for certain types of forensics where the
original photograph is important critical evidence and blurred.

A very old example: http://www.maxent.co.uk/example_1.htm
(done in the 1980's)

Computational cost is about 200x normal linear inverse methods so it is
still time consuming even today. But it can be done.

You can get roughly 3x the raw image resolution on a good day with a
trailing wind on the highlights and with sufficient signal to noise. The
catch is that resolution then depends on local signal to noise.

--
Regards,
Martin Brown