Grainy results from Canoscan FS4000 film scanner

[mark]

Hi all,

I bought a Canon Canoscan FS4000U film scanner at the weekend so I
could get through all my APS rolls and I'm finding the quality of the
scanned images to be dismal. I've scanned the images at 4000dpi 42-bit
colour which copied put the output into Photoshop 6. The results are
very grainy. I've also tried 2000dpi at 24-bit colour but I get the
same problem.

As a test I brought one of the rolls into a film processing shop and
had them scan the images onto a cd. They aren't hi-res scans, file
size is between 650kbs and 1.2mbs, but the quality is so much cleaner
than the results from my FS4000. Therefore the problem can't be with
the roll, it has to be with either
1) me
or 2) the scanner

I then downloaded the trial version of Vuescan and the results are the
same.

What am I doing wrong? The quality is far to sub-standard to print
anything at the moment.

Please help.

Mark...

 

[Wilfred van der Vegte]

I bought a Canon Canoscan FS4000U film scanner at the weekend so I
could get through all my APS rolls and I'm finding the quality of the
scanned images to be dismal. I've scanned the images at 4000dpi 42-bit
colour which copied put the output into Photoshop 6. The results are
very grainy. I've also tried 2000dpi at 24-bit colour but I get the
same problem.

As a test I brought one of the rolls into a film processing shop and
had them scan the images onto a cd. They aren't hi-res scans, file
size is between 650kbs and 1.2mbs, but the quality is so much cleaner
than the results from my FS4000. Therefore the problem can't be with
the roll, it has to be with either
1) me
or 2) the scanner

I then downloaded the trial version of Vuescan and the results are the
same.

What am I doing wrong? The quality is far to sub-standard to print
anything at the moment.

Could it be grain aliasing?
http://www.photoscientia.co.uk/Grain.htm

 

[Bruce Graham]

Hi all,

I bought a Canon Canoscan FS4000U film scanner at the weekend so I
could get through all my APS rolls and I'm finding the quality of the
scanned images to be dismal. I've scanned the images at 4000dpi 42-bit
colour which copied put the output into Photoshop 6. The results are
very grainy. I've also tried 2000dpi at 24-bit colour but I get the
same problem.

As a test I brought one of the rolls into a film processing shop and
had them scan the images onto a cd. They aren't hi-res scans, file
size is between 650kbs and 1.2mbs, but the quality is so much cleaner
than the results from my FS4000. Therefore the problem can't be with
the roll, it has to be with either
1) me
or 2) the scanner

I then downloaded the trial version of Vuescan and the results are the
same.

What am I doing wrong? The quality is far to sub-standard to print
anything at the moment.

Please help.

Mark...

what film? for example for me Fuji 100 scans much easier than Fuji 200.
With Fuji 200 I sometimes needed to defocus the scanner a bit (user
control in both the Canon and Vuescan software) to solve what I think
is the the grain aliasing "speckle" noted by other poster. I haven't
done any APS in a while though.

does the scanner work OK on other 35mm films??

 

[joe]

Could it be grain aliasing?
http://www.photoscientia.co.uk/Grain.htm

Okay, a couple questions before we get too deep on this one.

1) Have you ever scanned film before and do you know what you are
looking for?
2) Have you ever scanned slides before (and know that slides scan
different)?
3) Have you ever used one one the noise reduction tools like "Neat
Image" before?
4) You know that you might have to select what dynamic range you wnat
from film (unlike slides)?

I once thought film was inferior (and very grainy) and useless until I
figured out that its just different and that it was critical to select
the range that is importnat to you. Film just contains a lot of
info. Question #1 above is very important one.

Joe

 

[mark]

Hi there Joe,

Thanks for the response.

1) Nope, have never scanned film before. From all the reviews I was
hoping the whole 'chuck a roll of film into the APS holder and let it
go' would do the trick. Colour matching or Dust & Scratch settings
don't make any difference. But the key for me was looking at the
results from a photo shop who scanned the same roll. The difference
was astounding.
2) Haven't scanned slides as of yet... that's the next step once I've
scanned my 60 or so rolls of APS film.
3) Haven't used a toll like "Neat Image". Is this a post scanning
process whereby I open my image using said package and let it do it's
magic?
4)Dynamic range??? How do I do this? Does that come with the FilmGet
software?

Tips would very much be appreciated.

Thank you.

Mark...

 

[mark]

what film? for example for me Fuji 100 scans much easier than Fuji 200.
With Fuji 200 I sometimes needed to defocus the scanner a bit (user
control in both the Canon and Vuescan software) to solve what I think
is the the grain aliasing "speckle" noted by other poster. I haven't
done any APS in a while though.

does the scanner work OK on other 35mm films??

Haven't tried 35mm films yet but maybe should give them a go. As for
film type I think it is a Fuji 200 roll that my test was made using.
I've found that while Kodak still has the same problem, it's not
anyway as bad.
But sa for defocuing... how do I do this? There's just a tiny window
in FilmGet which feels like closing your eyes and pointing randomly.
Is there any easy way to do this?

 

[mark]

Could it be grain aliasing?
http://www.photoscientia.co.uk/Grain.htm

Thanks for the link. Will try it out. It'll take ages though using
FilmGet. Fiddle and test scan, fiddle and test scan, etc... Just find
it strange coz I can't imagine the processing shop using this
technique to scan my photos and they were much better quality at a
lower resolution.

Think I'm gonna have to chuck a sample comparison up on some random
geocities website to give ye all a better idea of where I'm coming
from.

 

[Kennedy McEwen]

Hi all,

I bought a Canon Canoscan FS4000U film scanner at the weekend so I
could get through all my APS rolls and I'm finding the quality of the
scanned images to be dismal. I've scanned the images at 4000dpi 42-bit
colour which copied put the output into Photoshop 6. The results are
very grainy. I've also tried 2000dpi at 24-bit colour but I get the
same problem.

As a test I brought one of the rolls into a film processing shop and
had them scan the images onto a cd. They aren't hi-res scans, file
size is between 650kbs and 1.2mbs, but the quality is so much cleaner
than the results from my FS4000. Therefore the problem can't be with
the roll, it has to be with either
1) me
or 2) the scanner

I then downloaded the trial version of Vuescan and the results are the
same.

What am I doing wrong?

Probably nothing at all - see below.

The quality is far to sub-standard to print
anything at the moment.

Are you sure?

Please help.

One problem for folks new to scanning is to resist the temptation to
just zoom in to the image to see the finest detail that the scanner can
pick up . If it is a good scanner that can be one heck of a shock
because most of what is there to see is just junk! Film grain, dye
clouds and some soft fuzzy image components - and you have just paid a
lot of dosh to get it!

On a poorer scanner, with lower resolution, a lot of that junk is just
averaged across the bigger pixels in the scanner - so you never see it
and the image just looks nice and smooth and grain free. That is what
you are seeing on the 650Kb & 1.2Mb scans from the processing shop. Nice
clean low resolution images that just can't pull any of the real detail
(and the junk) out of the image on the film.

However, when you print your high resolution scans, all that junk that
you see at pixel level is also reproduced much smaller on the page and
your eye just averages it out instead - in a better way than the cheap
scanner's pixels can. Hence my question about whether you are sure your
results are not fit for printing - try them, and compare them to the
same prints from the photo store CD. In all likelihood you will be
surprised - the resolution of the scanner will be much more obvious than
the junk, and the images are probably much better than the CD prints!
;-)

You can make your good scanner seem like a cheaper scanner, producing
nice clean low resolution images, but not just by decreasing the
resolution - all that does is make the problem relatively worse (same
magnitude of junk, but less image detail). This is almost certainly why
you did not see any change when scanning at 2000ppi. Try scanning at
4000ppi and then blurring the image before resampling to 2000ppi. In
Photoshop, bilinear and bicubic interpolation already include the
necessary level of blur as part of the resampling process, so don't add
any more, but other imaging suites vary - you might need to experiment a
little, depending on the application you are using. What you are trying
to do is reconstitute every pixel in the 2000ppi output from an average
of 2x2 groups of pixels in the 4000ppi input. Some imaging packages
have user defined filters which will permit you generate exactly this
kernel.

Using this approach you will get half the noise at the 2000ppi output
than you will get from a direct 2000ppi scan - a level probably more
comparable to the highest resolution CD scans. Similarly if you want a
1000ppi image (still large by monitor display sizes) then you should
average every 4x4 pixels before resizing.

 

[Bruce Graham]

Haven't tried 35mm films yet but maybe should give them a go. As for
film type I think it is a Fuji 200 roll that my test was made using.
I've found that while Kodak still has the same problem, it's not
anyway as bad.
But sa for defocuing... how do I do this? There's just a tiny window
in FilmGet which feels like closing your eyes and pointing randomly.
Is there any easy way to do this?

As Joe has pointed out, if you are new to this game, there are a lot of
things that can go wrong! However, given that you are scanning Fuji 200,
I would try the manual defocus as it has solved problems for me with the
same scanner and same film.

I assume you are using Filmget and not Vuescan.

scan a frame on auto, then from the Filmget menu:

Settings | Set Focus | uncheck auto focus, shift the slider a big amount
(from the position that the scanner has just autofocussed).

Rescan a frame and see the effect (turn off cleaning to speed things up
here)

Compare the two frames side by side at 100%. If you have lost a lot of
detail but smoothed out the scan, then position the slider halfway back
to the original position. Find a position where your noise is reduced
with an acceptable sharpness. It is not very critical and this
adjustment can be used in future. (Just remember what offset you want
from the autofocus position - I have not used Filget for a while so I
can't remember the settings I last used for this)

I like Vuescan for APS because you can scan the whole roll at high
resolution without running out of memory in PhotoShop. However, if you
have a lot of memory, you should get through a big chunk of a roll with
Filmget. (I only have 512MByte and I can do about 10?? frames at full
res.) I would stay with FilmGet until you get a feel for the whole
scanning process as VueScan has its own extended learning curve.

You can VERY nice results with this scanner, so keep trying. Let us know
how it works.

(Have you seen wayne Fulton's site www.scantips.com??)

Bruce

 

[user]

Try scanning at
4000ppi and then blurring the image before resampling to 2000ppi. In
Photoshop, bilinear and bicubic interpolation already include the
necessary level of blur as part of the resampling process, so don't add
any more, but other imaging suites vary - you might need to experiment a
little, depending on the application you are using. What you are trying
to do is reconstitute every pixel in the 2000ppi output from an average
of 2x2 groups of pixels in the 4000ppi input. Some imaging packages
have user defined filters which will permit you generate exactly this
kernel.

Using this approach you will get half the noise at the 2000ppi output
than you will get from a direct 2000ppi scan - a level probably more
comparable to the highest resolution CD scans. Similarly if you want a
1000ppi image (still large by monitor display sizes) then you should
average every 4x4 pixels before resizing.

Will the noise be reduced by half if the noise is not completely random?
I've downsized images by a factor of two using PS 6 with bicubic
interpolation and observed that the noise's standard deviation decreased
to 68% of its original value. Does PS 6 blur before resampling?
<http://www.unc.edu/~pedit/noise3/noise3f.html>

Thanks,
Joe

 

[joe]

Mark:

I would definitely support you scanning a slide for comparative
purposes because they are definitely different (slide vs film). One
of the most concise explanations of dynamic range differences between
the two is in the Vuescan user guide. Please look at the following
link:

http://www.hamrick.com/vuescan/html/vuesc10.htm#topic9

As far as Neat Image is concerned, I find that this is an invaluable
tool for noise removal and should be done as close to scanning as
possible (specifically before anything else is done, like selecting
white and black points to set the range).

This point is actually key, how do you set white and black points and
are you doing that? Lastly, I would recommend that you use Vuescan
over Filmget. Better, faster, etc.

Hope this helps. Be prepared for a big learning curve on this stuff.
Theres a lot to learn.

Joe

 

[Kennedy McEwen]

Will the noise be reduced by half if the noise is not completely
random?

Certainly not - systematic noise can not be averaged out this way.
Similarly low spatial frequency noise cannot be reduced this way either.

I've downsized images by a factor of two using PS 6 with bicubic
interpolation and observed that the noise's standard deviation
decreased to 68% of its original value.

Don't confuse the standard deviation, a measure of the distribution
across the image, with the noise, which would be the standard deviation
of scanning that same pixels many times. The standard deviation of the
image includes any image structure present, including very coarse
structure such as a gradual change of level across the field.

In your samples, you have a significant image structure which is much
coarser than a 2x2 pixel area in the original. For example, as well as
the overall grain, there is a clearly darker blob near the centre of the
image and fuzzier lighter blobs around the image. The image structure
will therefore dominate the standard deviation. As you will note, your
68% reduction of SD reduces to 55% and 53% for the x2-x4 and x4-x8
reductions respectively, since the structure is becoming less dominant.

Even so, any tilt in level across the image would still skew the
standard deviation reduction - although your examples looks pretty even.
Also, don't overlook the edge effects of the bilinear and bicubic
filters, which

Does PS 6 blur before resampling?

Unfortunately I jumped from PS5 directly to PS7, but I am pretty sure it
does. If it didn't then the standard deviations would remain almost the
same - as you would get if you resized them using the nearest neighbour
algorithm, which aliases the spatial noise that would otherwise be
filtered back into the pass band of the resampled data.

 

[Bart van der Wolf]

Also, don't overlook the edge effects of the bilinear and bicubic
filters, which

As a Photoshop alternative to binning or averaging multiple pixels, I use
Filter|Pixelate|Mosaic... , followed with an integer percentage (equivalent
to the amount of pixel binning) Resize.

Unfortunately I jumped from PS5 directly to PS7, but I am pretty sure
it does.

SNIP

To reduce aliasing , it is still beneficial to pre-blur with a small radius
(0.2-0.3 radius Gaussian per integer reduction factor), then sharpen. So
e.g. a 1:3 reduction (33.33%) would benefit from a 3*0.2=0.6 radius
pre-blur. This will sharpen very well, with little chance on clipping.

Bart

 

[user]

SNIP

To reduce aliasing , it is still beneficial to pre-blur with a small radius
(0.2-0.3 radius Gaussian per integer reduction factor), then sharpen. So
e.g. a 1:3 reduction (33.33%) would benefit from a 3*0.2=0.6 radius
pre-blur. This will sharpen very well, with little chance on clipping.

Bart

I tried this approach by applying a 0.6 radius blur to the original
200x200 image before reducing the image to 100x100. The results are at:

<http://www.unc.edu/~pedit/image1.gif>
<http://www.unc.edu/~pedit/image2.gif>

The standard deviations are:
200x200 original without blur 7.4436
200x200 original with 0.6 radius blur 4.3583
100x100 downsized original without blur 5.0493
100x100 downsized after blurring 3.9507

Blurring increased the spatial structure of the noise even though the
standard deviation was reduced from 7.4436 to 4.3583. The range over
which noise was correlated increased from about 3 pixels to about 4
pixels as a result of blurring. Pre-blurring lowered the standard
deviation of the noise in the downsized image, but increased the amount
of correlation for lags less than two pixels.

Joe

 

[Kennedy McEwen]

I tried this approach by applying a 0.6 radius blur to the original
200x200 image before reducing the image to 100x100. The results are at:

<http://www.unc.edu/~pedit/image1.gif>
<http://www.unc.edu/~pedit/image2.gif>

The standard deviations are:
200x200 original without blur 7.4436
200x200 original with 0.6 radius blur 4.3583
100x100 downsized original without blur 5.0493
100x100 downsized after blurring 3.9507

Blurring increased the spatial structure of the noise even though the
standard deviation was reduced from 7.4436 to 4.3583.

I doubt that very much. What it probably did was eliminate the finer
spatial structure, leaving only the coarser structure visible. The blur
filter is symmetric, it cannot increase the size of spatial structure.

The range over which noise was correlated increased from about 3
pixels to about 4 pixels as a result of blurring. Pre-blurring lowered
the standard deviation of the noise in the downsized image, but
increased the amount of correlation for lags less than two pixels.

Once again, don't confuse the standard deviation of your image with
noise - the two are quite demonstrably different.