Hi,

I believe that scanner calibration is used to compensate the sensors'
output so that the output from each sensor is constant for a constant
light source. So, if I calibrate my scanner using Vuescan and then scan
a blank slide I would expect all pixels in the raw file to be identical
and very close to white (say RGB 252, 252, 252). However, I just can't
get results like this.

Generally I find the average pixel is (247, 239, 245) and each sample
varying by up to plus or minus 4. The variations are random and I can't
see any lines in the image from the raw file. But the variations and
the colour imbalance do seem excessive to me.

The intra-colour variations (+-4) look like noise in the scanner to me.
Does anybody think this is reasonable or excessive or have any other
comments ?

The inter-colour variations are consistent (red higher than blue higher
than green) when I compare the averages from different scans but the
colour balance of pixels in a scan does vary.

I thought that the calibration would remove this colour cast. Is the
imbalance I am seeing the way things are meant to be ?

Finally, does anybody have any details about the Vuescan calibration
file (96,000 dwords) ? Perhaps I can fiddle with things here and
evaluate the results.

Thanks,

Steven

SJS <(E-Mail Removed)> wrote:

> I believe that scanner calibration is used to compensate the sensors'
> output so that the output from each sensor is constant for a constant
> light source. So, if I calibrate my scanner using Vuescan and then scan
> a blank slide I would expect all pixels in the raw file to be identical
> and very close to white (say RGB 252, 252, 252). However, I just can't
> get results like this.
>
> Generally I find the average pixel is (247, 239, 245) and each sample
> varying by up to plus or minus 4. The variations are random and I can't
> see any lines in the image from the raw file. But the variations and
> the colour imbalance do seem excessive to me.

This is the film base color. Slide film base is not exactly clear but
slightly colored. Thats why vuescan needs to determine film base color
if you use media type slide.

The +/- 4 variaton is a mixture of scnner noise and uneven film base.
Not all color might have been washed out during processing. You can
reduce scanner noise by averaging multiple scans. Set Number of passes
to 2, 4, 8 or 16. Each doubling should reduce noise by the same amount.

--
Erik Krause
Digital contrast problems: http://www.erik-krause.de/contrast

On Wed, 9 Jun 2004 00:36:23 +0200, Erik Krause <(E-Mail Removed)>
wrote:

>> Generally I find the average pixel is (247, 239, 245) and each sample
>> varying by up to plus or minus 4. The variations are random and I can't
>> see any lines in the image from the raw file. But the variations and
>> the colour imbalance do seem excessive to me.
>
>This is the film base color. Slide film base is not exactly clear but
>slightly colored. Thats why vuescan needs to determine film base color
>if you use media type slide.

Hi Erik,

I am using media type image and there is no film (the slide holder is
empty). Surely this is as uncoloured as I can get.

>The +/- 4 variaton is a mixture of scnner noise and uneven film base.
>Not all color might have been washed out during processing. You can
>reduce scanner noise by averaging multiple scans. Set Number of passes
>to 2, 4, 8 or 16. Each doubling should reduce noise by the same amount.

As there is no film base I can only assume the variation is due to
noise. But, is +-4 really reasonable in a scanner which should have a
density range of 3 and precision to match ?

Still, in the case above (calibrate and then scan blank slide) don't you
think I should get a fairly constant sample reading (e.g. +-1) and also
a better colour balance ?

-- Steven

Steven,

the transp IT8 target you're using 'could' have been drifting a little
bit from what the reference/descriptionfiles tells in numbers for
targetvalues. Aging of the target or a massproduction of targets will
always show "offs". They're not measured individually or they are VERY
expensive.

Scanning the lightsource itself – and not through the clear filmbase –
is giving you light levels above what is profiled. The filmbase will
always steal light and is profiled as white a bit lower than the empty
slideholder. Variations can be found there maybe. If you're scanning a
clear slide it could be tinted as someone else mentioned here. Tinted
in another way than the target. If the readings are the same on
filmbase as is in empty slideholder, I would expect the drifting
ref-file>IT8 target beeing a thing here.


A part from that, when building *scanner*-profiles with Profilemaker
Pro or Monaco Platinum I do see differences inbetween these softwares
themselves and I do also see that non is really totaly perfect in
greyramp when reading numbers. How could it be possible that a simple
utility like Vuescan IT8 or the freestanding brother of it, would
deliver better than these industrystandard apps? Well, it could be if
you're a lucky guy but you don't seem to be that for the moment. We
would have to read the IT8 target with a Spectro and compare the
measurefile in Profilemaker with the ICC profile from Vuescan, to get
delta-E values. To see how much Vuescan fails when reading the IT8 and
by that understand what our eyes will se or not of it based on these
delta values.

Also, let Vuescan deliver a scan from the profiled settings **in the
profiled space** (if possible) and in Photoshop make conversions from
that into your workingspace with different renderings (perc vsv relC).
That could make a difference of what you start working with for the
editingpart...and I think that is what counts. We do avoid editing in
the scannerpace as is. This space is not greyballanced like a
workingspace like Adobe RGB or Colormatch RGB.

Diffrent CMS engines are known to deliver differently...so letting
Photoshops engine do the conversion may deliver something else than
what Vuescan-conversions gives you on the fly.


nikita

On 9 Jun 2004 02:35:19 -0700, (E-Mail Removed) (nikita) wrote:

>Scanning the lightsource itself – and not through the clear filmbase –
>is giving you light levels above what is profiled. The filmbase will
>always steal light and is profiled as white a bit lower than the empty
>slideholder. Variations can be found there maybe. If you're scanning a
>clear slide it could be tinted as someone else mentioned here. Tinted
>in another way than the target. If the readings are the same on
>filmbase as is in empty slideholder, I would expect the drifting
>ref-file>IT8 target beeing a thing here.

Hi nikita,

I am scanning the lightsource itself (is this the ultimate in boring and
absurd ?) so your comment about the light level being above reasonable
may be relevant. But, isn't determining the white level one of the
steps of calibration ? I am assuming it is and therefore scanning the
lightsource after a calibrate should return samples all close the
255/256. I was expecting say 254/256 +-1 but I am getting a colour
imbalance and larger variations within each colour.

I am trying to determine if my scanner is unreasonably noisy or if I am
being unreasonably fussy. Yesterday I took the scanner back to Canon so
they can test it. Now I have to wait for two weeks so I hope they can
find something wrong.

-- Steven

Steven,

you're not actually *calibrating* the scanner. The profile is not
changing the bahaviour of the scanner iteself – it is DESCRIBING it.
If you feed the profilerapplication with the IT8 target it reads the
filmbase as the lightest patch to describe. Anything above that isn't
described in the profile. That info is used as a sourcebase for the
conversion into the workingspace/monitorprofile. Before conversion
into the workingspace, while still just assigned with the
scannerprofile, that is all it has to rest in. I don't think this is
much relevant anyway, but it might take away some doubts from your
mind.

Noise is mostly a problem in the shadows when scanning slides and in
the highlights of negatives. I think that if it was that noisy you
woulden't want to use it based on what it would deliver to your
eyes..... You *should see* if there is a lot of noise in the higher
end of the greyramp patches. Multisampling should kill most of it as
Erik said.

As for scanning lightsources, that might not be that absurd as half of
the population of earth were scanning the sun a few days ago and found
Venus passing...you never know what will pass the Canon lightsource
during these two weeks without having access to it. That would kill me
if I were you

nikita

nikita <(E-Mail Removed)> wrote:

> you're not actually *calibrating* the scanner. The profile is not
> changing the bahaviour of the scanner iteself

I think you mixed calibrating and profiling. As far as I understood
Steven did not use a IT8 target but simply the vuescan menu command
'Calibrate'...

--
Erik Krause
Digital contrast problems: http://www.erik-krause.de/contrast

SJS <(E-Mail Removed)> wrote:

> I am scanning the lightsource itself (is this the ultimate in boring and
> absurd ?) so your comment about the light level being above reasonable
> may be relevant. But, isn't determining the white level one of the
> steps of calibration ? I am assuming it is and therefore scanning the
> lightsource after a calibrate should return samples all close the
> 255/256. I was expecting say 254/256 +-1 but I am getting a colour
> imbalance and larger variations within each colour.

You will have to use Color Balance None at least if you want to judge
this. If you use any other color balance vuescan would set white and/or
black point in order to stretch the histogram to the relevant data.

Make a test if you have good nervs: Scan the empty film holder, Crop
well inside the white area and set color balance to Neutral and black
and white point both to 1%...

--
Erik Krause
Digital contrast problems: http://www.erik-krause.de/contrast

On Fri, 11 Jun 2004 01:59:15 +0200, Erik Krause <(E-Mail Removed)>
wrote:

>You will have to use Color Balance None at least if you want to judge
>this. If you use any other color balance vuescan would set white and/or
>black point in order to stretch the histogram to the relevant data.
>
>Make a test if you have good nervs: Scan the empty film holder, Crop
>well inside the white area and set color balance to Neutral and black
>and white point both to 1%...

Hi Erik,

Yes, I am using color balance none and image mode and brightness 1 and
also I am looking at the raw file. So I expect to see the raw output of
the scanner presumably corrected by the calibration data. I don't know
exactly what calibration does but I assume it corrects errors due to
differences in the sensors. I also assume that this is done in the
scanner itself rather than in any PC software (e.g. Vuescan). I know
that Vuescan builds a calibration file and assume the scanner provides
this after a calibration (perhaps a read after the calibrate). I guess
that Vuescan can load this file back into the scanner and this happens
for each Vuescan run.

When I do one of these blank slide scans, I calibrate first, ensure
exposure is 1 and mode is image, set the crop (say 2000 * 3000), and
then scan to raw file. So I should get a nice even white raw file but
instead I get a file with variations and a colour imbalance.

Perhaps this is all a bit acedemic now because I have taken my scanner
back to Canon for a test.

-- Steven

On 10 Jun 2004 05:43:54 -0700, (E-Mail Removed) (nikita) wrote:

>you're not actually *calibrating* the scanner.

Hi nikita,

I think I am calibrating the scanner. I haven't even got as far as
worrying about profiles, etc. I am just trying to get a clean, reliable
output from the device.

I was trying to determine if I was being too fussy or if the scanner is
defective. Now I will wait for Canon to provide a judgement.

If you can provide any details about FS4000 calibration with Vuescan in
particular I would be very interested. I would like to know exactly
what calibration achieves and whether the resultant corrections are done
in the scanner itself or not.

Thanks,

Steven