[APML]: Image registration [Was: Latest Tony Hallas Images]

[Axel Mellinger <axm2@canopus.physik.uni-potsdam.de>]

The Astro-Photography Mailing List
------------------------------------

On Fri, 4 Sep 1998, Chuck Vaughn wrote:

> I guess you either like this or you don't. I prefer the more natural
> contrast range - although that's a laugh too because we stretch the
> contrast range like crazy on all astrophotos just to see anything.
> 
> >Example: multi-point image registration
> >
> >I saw Axel Mellinger's mathematical definition/implementation  in LINUX
> >many months ago (before the "craze" over the PW image registration
> >function) -- perfect example of how to do it right (no one gave it any
> >positive response -- ??).  
> >
> >http://canopus.physik.uni-potsdam.de/~axm/imgproc/node5.html
> 
> There's another problem he didn't address and it should be point (4).
> Curved space is being projected onto a flat piece of film which causes
> distortion from the center out. If two negatives are not closely aligned
> you can't get perfect alignment without non-linear rescaling one negative
> to the other. I've found this out when trying to alighn these tricolor
> negatives taken on different nights. This is the major reason why the I
> could not get good alignment on the Veil I just posted. Those three negatives
> did not cover the same field.

Chuck is making a good point - image registration of negatives that were
not taken on the same night with exactly the same camera setting is not
trivial and I didn't cover it in the mentioned article.
The problem of projecting the celestial sphere onto the film
plane came up recently when I created a mosaic of the southern Milky Way
(http://canopus.physik.uni-potsdam.de/~axm/images.html#mwpan). I have
worked out a procedure to do the registration, but did not have the time
to write up a detailed description. However, I am preparing a talk for a
meeting in Berlin early next month, and once I have all the text and
pictures together I will put them on my web site.

Writing a Photoshop plug-in is a good idea, but involves a significant
overhead to do all the GUI stuff. I did write some GUI-based applications
many years ago (in the age of the good old ATARI 1040), but have stayed
away from it since. However, if people are interested in my
stacking/registration programs, it should be fairly straightforward to
port them to DOS, so that anyone running Windows 3.1 or Windows9x could
use them.

As for the original posting by the anonymous "Space Boy", I find his
criticism way too harsh. After all, most of us do astrophotography in
order to obtain aesthetically pleasing pictures - not for scientific
results. And the least squares fitting procedures used for image
registration are definitely beyond high school level - I did not know
anything about them until my second year at university.
I just had fun applying that knowledge to astrophotography, and Linux is a
robust environment where I can get all the necessary tools (compilers,
image-processing software) for free.

Clear skies,


Axel


-- 
Axel Mellinger  <http://canopus.physik.uni-potsdam.de/~axm/astrophot.html>