Hans-Georg Michna
Guest
|
 Posted:
Mon Nov 14, 2005 3:05 am Post subject:
Re: What is better Nikon Super Coolscan 4000 or Coolscan V E |
|
|
Lorenzo,
thanks a lot for the quick user guide! That's very helpful.
Downloading ALE.
Hans-Georg
On Sun, 13 Nov 2005 16:42:27 +0100, Lorenzo J. Lucchini wrote:
| Quote: | Just specify:
ALE image_1 image_2 ... image_n output_image
and you'll be able to merge frames in a multi-pass multi-scan way (but
*with* the realignment).
This way, the output image will not have more resolution than the input
images, just (hopefully) less noise. If you want to try what we're talking
about in this thread, you can do
ALE --scale=x image_1 image_2 ... image_n output_image
where x is the scale factor (2 for example).
In both cases, you'll be using defaults (specifically, you'll default to
option "--q0") that try to make the process fast (sort of), but sacrifice
quality.
Add options "--qn", "--q1" or "--q2" (sorted by quality, from lowest to
highest) to change that. I think that, for experimenting with resolution
improving, one should use either "--q1" or "--q2", while "--qn" could be
adequate for normal multi-pass multi-scanning (the "n" in "--qn" stands for
"noise").
Also, by default, ALE only tries applying "Euclidean" (translation,
rotation) transformations to re-align images. You can specify
--projective
to get all the transformations ALE is capable of, or
--translation
if you are in a hurry.
|
--
No mail, please. |
|
Kennedy McEwen
Guest
|
| Posted:
Mon Nov 14, 2005 6:46 am Post subject:
Re: What is better Nikon Super Coolscan 4000 or Coolscan V E |
|
|
In article <ITIdf.1066$P16.108@tornado.fastwebnet.it>, Lorenzo J.
Lucchini <ljlbox@tiscali.it> writes
| Quote: | Kennedy McEwen wrote:
The question is whether it varies significantly enough to prevent
consistent alignment to half a pixel accuracy or so. That isn't so
demanding as your overarching statement, and many scanners are indeed
capable of this. Even for those where frame to frame alignment might be
out by more than a pixel, making dumb multipass multiscanning (as in
Vuescan) worthless, the variation across the field on each frame can
easily be less than half a pixel.
But if, say, the first scan line is "correct", and the last scan line is off
by half a pixel (or even more!), we can still do a geometrical
transformation to correct for this, can't we (ALE does this)?.
That depends on whether the transition from "correct" at the start of |
the scan to "incorrect" at the end of the scan is progressive or in one
or more discontinuous (whether monotonic or not) steps.
| Quote: | What I'm dubious about is what happens when *each* scan line is off by a
significant amount.
|
If each line is off by a significant amount then you won't get the
quoted resolution in the first place, let alone any enhancement of it.
| Quote: | At the end of the scan, the first scan line and the
last scan line could even both be in perfect alignment, if the previous
errors averaged out. But in this case, I can't think of a way ALE or any
other program could reasonably reconstruct alignment...
Quite - if all you have is random offsets between samples then you |
really can't do much about it. However, where I disagree is the
occurrence of such conditions in commercial scanners - I spent some time
today looking for just this type of effect on my Epson flatbed (having
established that it isn't significant in the Nikon) and it also steps
consistently with much better than half pixel precision on each step. It
is relatively trivial to establish this just by presenting the scanner
with a high enough resolution test image that induces aliasing.
Relatively small phase shifts of the sampling points, of the type you
are concerned about, produce very large phase errors of the aliased
signal. Simple arithmetic scales the alias phase error back to the
original.
| Quote: | Let's just say that the alignment of consecutive scans on my Nikon
LS-4000 is certainly better than half a 4000ppi pixel across the frame,
although the frame to frame alignment itself certainly is a lot worse.
There ay well be long term variations, say if one scan was performed
several hours after the other, but consecutive scanning is pretty good.
But even if the alignment is better than half a pixel across the frame, what
is the average misalignment between a pixel (or scan line) and the next?
Isn't this the most important factor?
|
It is an important factor - whether it is the most important depends on
the magnitude. As described above though, you can work all of this out
by inducing aliasing at the appropriate scale and observing the error.
If the line to line variation is significant then you won't be able to
induce aliasing in the first place. From the tests I have run on the
scanners I have at hand, line to line error isn't as significant as
general drift or step changes (the latter presumably due to the build up
and release of stiction at some point in the scan process).
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying) |
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in
