[A complimentary Cc of this posting was sent to
Bart van der Wolf
<bvdwolf@no.spam>], who wrote in article <433c9a36$0$11078$e4fe514c@news.xs4all.nl>:

This is an interesting exersize.  However, note that IMO application
of Richardson - Lucy to "ordinary" photographic images is misplaced.

AFAIU, Richardson - Lucy "improves" the image basing on two assumptions:

 a) What you have is a convolution of the initial signal with known
    PSF with some noise;

 b) The initial signal is non-negative.

The usefulness of RL in astrophotography is, IMO, mostly due to the
fact that for this application part "b" carries as much information
content as the part "a" (at least in images where the improvements
given by RL are most striking).  Since the information in "b" is "very
nonlinear" in nature, very resource-consuming algorithms are needed to
use this information.  [*]

However, "traditional" photography does not have the 100000:1 contrast
ratio of the night sky; neither has it most of the image "manifestly
black" (unless you consider some styles of night photography - some
night cityscapes should be very similar to astro).  For such images
(where the information content of "b" is close to 0) simple
deconvolution may give *exactly the same result* as application of RL.
Thus using RL is not going to improve image a bit comparing to the
simple deconvolution.

The simple litmus test for usefulness of RL: if the image produced by
RL has no (or almost no) "absolutely black" parts (those with 0 at one
of the channels), then it did not use the information in "b", and it
did not use the information about the distibution of noise in the
signal.  Thus its results coincide with pure deconvolution.

Likewise, if pure deconvolution did not produce negative signal, then
its results will coincide with those of RL.

[*] Although I very much suspect that I would be able to deduce a
   better algorithm than just the plain iteration method
   traditionally used to solve RL equations...

Hope this helps,
Ilya

P.S.  And, of course, the question remains on how to obtain the PSF of
     the lens.  But this question is common to RL and "pure
     deconvolution".

     Note that in

ilyaz.org/software/tmp/KM_A200-resolution-chart-ACRraw-quadratic-58percent-quartic-60percent.jpg

     I obtained the PSF to deconvolve via spectral analysis of edges
     of big black sloped squares.  You won't get such nice reference
     points on the "real" images, like one of yours.

     And on my image is it is not "full" deconvolution; I had chosen
     coefficients so that the throughput MTF is practicaly flat up to
     50% of Nyquist:

1   """"""""""""""""xxxx____'''''''''''''''''''''''''''''''''''''''|
   |                       ""xx__                                 |
   |                             ""xx_                            |
   |                                  "xx_                        |
   |                                      "_                      |
   |                                        "x                    |
   |                                          "x                  |
   |                                            "_                |
   |                                              x               |
   |                                               "_ |           |
   |                                                 _|           |
   |                                                  x           |
   |                                                  |x          |
   |                                                  | x         |
   |                                                  |  x        |
   |                                                  |   "       |
   |                                                  |    "      |
   |                                                  |     "     |
   |                                                  |      "_   |
   |                                                  |        _  |
   ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,|,,,,,,,,,x,,
-.05|..................................................|.........."_
   0                                                           3000

     The scale is in single lines per picture height, and the
     vertical line is the Nyquist frequency.

P.P.S.  Looking at the above graph, you can immediately see that the
    AAF for this camera has its first zero about 1.17 of Nyquist.

There are a bunch of problems that make sampling images more difficult
than sampling 1D electrical signals (like audio), and thus everything's
a compromise:

In audio, you can build brick-wall analog filters that operate in the
time domain.  There is no equivalent spatial-domain brick-wall filter
for images that I know of.

You can massively oversample audio, then do the brick-wall filter
cheaply digitally.  Oversampling images is not practical because of
loss of sensitivity (light-collecting area), fabrication problems, and
the amount of data that would result.

You can use many-tap filters in audio because the data rate is modest.
Large 2D filters are often not practical in imaging, so most filters
are small (4x4).

Perhaps, but lens blur spot size (and profile) is still heavily
dependent on the aperture of the lens.  Fixing the lens aperture fixes
that, but also removes depth-of-field control and gives only one
shutter speed choice for any given lighting conditions (unless you add
variable ND filtering).

    Dave