On 07/18/2018 06:22 PM, clepsydrae wrote:
Hmm. Make copies of images, convert them to B&W (1 bit color), pull
into Hugin and cpfind/optimize/etc on those. Save project, replace
B&W versions with color, stitch?
Thanks David -- I did try two versions of that method: one where I did a
curves adjustment that brought out the stars and crushed everything else
to black, and one that just boosted everything (including the noise
floor). Neither worked (cpfind found no points at all, which was even
worse than before). I tried true 1bit as you suggest and hugin said it
didn't support 1bit and suggested greyscale, so I converted the 1bit
images to 8bit greyscale. None of the cpfind methods I tried with that
found any CPs.
Also, I don't recall if you mentioned this earlier. Do you do noise
reduction on your images before pulling them into Hugin?
No -- since the reason for the stack is that I intend to do a gmic
median_image for the purpose of noise reduction I didn't want to do any
per-image NR. Would it help for CP-creation?
Possibly. Others can explain about how cpfind works, but I think it
works on areas rather than specific points. Otherwise, how is it
supposed to determine that Bright Spot A in upper left isn't the same as
Bright Spot B in lower right?
My experiment described
just above seemed to imply it would not (since the first curve
adjustment effectively removed a lot of noise.) I.e. it seems like the
only way cpfind is finding anything at all is by matching subtle
variations in the darker areas.
Don't know. I haven't worked with starfield images like yours. Perhaps
setting a few manual control points in Hugin would help get the process
started?
How would -starfield mode define what a "star" is? A threshold of
brightness, size, what?
I confess cluelessness, but it seems like cpfind is not designed to look
for little bright points in a dark field. but rather to match more
textural image areas.
Well, it does pick control points on edges of things or where edges
intersect, at least on my ordinary photos.
Even on the stack that worked, half of the control
points it found were in the ~black areas (though I assume it was still
using the nearby stars to locate the CP) and there were plenty of CPs
that were obviously wrong, where "obviously" is defined as a human
looking at stars. :-) Meaning the CP is clearly not just using bright,
contrasty points to make decisions. (Which makes sense, since it's not
an astronomical imaging tool.)
It seems like a "dumb" version of cpfind could be told to just find
stars, defined as less than X pixels in diameter and with a certain
degree of contrast to the background (or even just a brightness
threshold as you describe).
That would make sense to me!
I'm sure the author(s) of cpfind don't want to create a special
algorithm for every type of image in the world, but it seems like star
field alignment is common enough application that it might make sense.
And since it seems so hard to make cpfind do it (see above thread), and
since it seems like it "should" be among the most simple kinds of
alignment to do, that maybe it's a good suggestion? Or maybe it just
belongs in a different CP tool altogether.
Something worth mentioning to them. They may know of others using cpfind
in similar situations as yours, people who could give you some ideas.
Somebody has to prepare all those astronomical photos for publication!
(Or maybe someone knows some magic I can pass to cpfind on the command
line to make it work!)
Don't know. IIRC, you were having it run with the --fullscale option?
Does it do any better without that?
--
David W. Jones
[email protected]
wandering the landscape of god
http://dancingtreefrog.com
--
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
---
You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/hugin-ptx/2627d3b9-9404-ac8e-fb83-9813ef7ca377%40gmail.com.
For more options, visit https://groups.google.com/d/optout.
