Re: [APML]: Image Registration

["Philip Perkins" <philip@astrocruise.com>]

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

> 1. Someone said that Jonathan Sacks discovered a problem with the
> registration engine in the current release version of PW.  I believe he
> correted it in one of the later "astro" non-public releases, maybe the one
> with the ADD function as an addition to the composite methods. Philip may
> know the answer to this.

Yes, I heard this too.  I am pretty sure that this was fixed at the same 
time as the 48 bit version was introduced, if not before.  I will ask 
Jonathon Sachs.

I do not seem to have the image registration problems that some others have
reported.  I am pretty meticulous about registration.  Firstly I only scan
at 2700 dpi. Secondly, I only ever do 4 point alignment, to remove any
possible doubt.  After very carefully aligning to sub-pixel accuracy within
each star, I then do a Subtract as the very first step. I magnify the
Subtracted image to 100% or more and inspect it for alignment errors.
Generally I will accept it if I see only slight alignment errors in no more
than about 20% of the image, which must be at the edges.  By slight
alignment errors, I mean a very slight overlap visible as a very slender,
dim crescent on one side of a star.  In fact this sort of misalignment is
so slight that it is hardly visible once the images are stacked.  I will 
accept no alignment errors in the 60% central region.

There are several things that can cause non-correctable alignment errors. 
Jerry mentioned most of them, such as film warp in one image and not the
other, non-linear scanning, field rotation in one image and not the other. 
They all amount to non-linear scaling errors.  If one is faced with a
scaling error that is non linear then I think it's a a very tough one for
which I do not have a solution.  Herein lies the dilemma for Chuck, I
think.  My only suggestion is the not very helpful one of trying to remove
the problem at source.

I am achieving much higher registration accuracy with PW than I did with
PS, largely because PW automatically scales and rotates, and I found these
things quite hard to do in PS.  In fact it seems the only thing PW cannot 
handle is non-linear scaling errors.

Regards
--Philip

Philip Perkins  --  philip@astrocruise.com
51 27'N 1 36'W  --  Wiltshire, UK
43 54'N 5 32'E  --  Luberon, France
Astrocruise     --  http://www.astrocruise.com