Don said:No, he's not. He's been very evasive and contradictory. He keeps
moving the goal posts and I keep submitting everything asked of me.
And every time he's cornered he just goes silent.
Don said:I sent it to "(e-mail address removed)" with the subject line "Analog Gain
saga...".
If you can't find it, it's the same temporary account I set up for the
images I posted:
tempdon100164833 (at) aol.com
Actually I've been using white highlights to detect where my scanner
fails and then adjust Analog Gain (AG) until I recover dynamic range
(see below for more).
I'm not entirely happy with this because I would prefer an objective
method. I still believe Nikon should have provided the values needed
for this adjustment but those in-the-know around these parts tell me
that's impossible.
I'm still not convinced and I'm now close to figuring out the formula
(as a function of exposure) to get at the ratio of distortion
empirically and then correct it i.e. in my case that means how much I
need to cut blue and boost red to correct Nikon's distortions.
The way I go about it is to examine the histogram of the area in
question. This exposes the inadequacies of the scanner as individual
channel don't reach 255 e.g. R=120, G=130, B=250. I then adjust AG to
stretch R & G until they also approach 250 or thereabouts. After that
I have enough dynamic range across the board to color balance and do
other editing without causing banding or other artifacts.
degrub said:Silverfast loads its own version of the Nikon scan libraries (maid
modules). You may have updated or gotten a slightly "fixed" version
versus what Nikon has released.
Frank
degrub said:Twain or PS plugin version, Mac or Win ? i can vouch for the Win PS
plugin version as you used to have to load NS before SF. If you did it
the other way, it would not work. i forget which version that changed
on. So i expect SF overwrites older maid modules if required. i have not
needed to test that however.
Frank
I never received it.
I've sent an e-mail with a VueScan serial number to this account.
You keep being asked for a raw scan and keep refusing to comply. Who's
evasive?
Somehow, this doesn't make much sense to me. At R=120 you lose about 2.4 bits of
accuracy compared to the full scale 255 (assuming gamma=2.2). This
should result in slightly more noise, but nothing major. For negatives, two
bits would be much more important.
I don't why you don't consider the analog gain to be objective.
I think you should
simply be able to auto-expose, preview (without ROC or GEM, I don't trust those),
compute the required gain in bits, double the amount, and set the analog gain to
that amount.
1) Delete vuescan.ini (this is important)
2) Run VueScan
3) Set "Output|Output raw file"
4) Press Preview button
5) Adjust cropping
6) Press Scan button
7) Put scan0001.tif somewhere I can download it from
Set "Input|Scan resolution" to a value that makes the raw scan
file less than 1 MByte.
The only deviations from the above:
- 450 dpi
- Output / Raw compression = On <===!!!!!!!!!
- Renamed the file to VS-RAW
http://members.aol.com/tempdon100164833/VS/VS-RAW.tif
No problem with this scan:
set input/scan mode: Transparency
color/ slide vendor: Kodak
slide brand: Kodachrome
color balance: White balance
Adjust the histogramm and brightness slightly and you have a perfect
scan without blue cast.
Erik Krause said:Not perfect but acceptable.
In this case the
film base color does the adjustment. Lock film base color after initial
preview and you see on the color tab, that it has values Red 0.142,
Green 0.268 and Blue 0.882. These are extremely different values as I
would expect (the other way round) if scanning color negative as slide.
So there *is* a heavy blue cast in the raw data.
If you look at the histogram of the raw image in photoshop there are no
pixels with values higher than 93 for Red and none higher than 163 for
Green channel. Blue goes up to 255. If you stretch the histogram in
order to get a balanced, bright image with the levels dialog it gets
posterized even though it is a 16 bit image. Too much data missing.
Now then, but boosting Red and Green selectively the histogram can be
stretched and data recovered so a meaningful adjustment is possible
without posterization - as I've demonstrated in the Nikon scan.
Don said:
Right, I tried it for about 1-2 minutes and came to a proper
result. You surely can find some better settings. But IMHO the result
of my quick scan was par or better than the scan Don has posted before
with Nikonscan.
This is nothing new. It's a well known and well documented shortcoming
of Nikon scanners. Newer ones have a special Kodachrome mode to try
and deal with this, the older ones don't.
Note that using a raw scan file and setting
"Input|Scan mode" to "Transparency" produces a
perfectly normal looking image of snow, once you
click with the right mouse button on a neutral
color.
However, you end up with quite bizarre colors when
you set "Input|Scan mode" to "LS-30/LS-2000". This
is undoubtedly what you've been seeing when you've
tried scanning this slide with VueScan.
The problem ou're seeing is that the 3x3 matrix
transform is ssuming better color balance between the
three LED brightnesses. Calibrating your particular
scanner with an IT8 calibration slide should give
you significantly better (almost perfect) scans.