Vuescan + LS40 : grain reduction compared to Nikon Scan GEM

[Dominique Deleris]

Hello,

I've always scanned my negs with my LS-40 using Vuescan, and did
not even try Nikon scan so far. I was concerned about my images
being a little "grainy".

I've now discovered ;-) the "Grain reduction" feature of Vuescan,
and compared it to the Nikon scan Digital GEM...

I am disappointed to discover that Vuescan's grain reduction:
1. is far from GEM in terms of results
2. does not even seem to have any effect on my scans

For point 2, I see no difference between a scan without grain
reduction, or with "Grain reduction" set to medium...

This is unfortunate for 2 reasons:
1. I scan under Linux, therefore with Vuescan
2. Vuescan is much faster that Nikon scan in all other aspects,
and I find it more intuitive.

Ed, can you comment on that?

Kind regards,

 

[ThomasH]

Hello,

I've always scanned my negs with my LS-40 using Vuescan, and did
not even try Nikon scan so far. I was concerned about my images
being a little "grainy".

I've now discovered ;-) the "Grain reduction" feature of Vuescan,
and compared it to the Nikon scan Digital GEM...

I am disappointed to discover that Vuescan's grain reduction:
1. is far from GEM in terms of results
2. does not even seem to have any effect on my scans

For point 2, I see no difference between a scan without grain
reduction, or with "Grain reduction" set to medium...

This is a known issue. As it seems, it is not an easy to match
performance of GEM. Personally I always make "raw files" processed
by ICE and GEM and than I "scan" them with Vuescan to achieve
good color fidelity.

For me of all ASF products GEM has a unique standing.
ICE has a few comparable software solutions, including also
Ed's IR cleanup in Vuescan, which albeit enormously slow,
also seem to make a nice job with Kodachromes, while ICE is
failing completely. Canon made their FARE system. ROC has
also a few nicely performing competitors, including Vuescan
fade and color restorations.

But I found nothing performing even close to magnificent
smoothing results of GEM, with its relatively nice preservation
of edges and sharpness. Vuescan's grain reduction seem to work
like one of the averaging filters, while GEM works like a pattern
analyzer, according to the description.

Thomas

 

[Bruce Graham]

snip>>
But I found nothing performing even close to magnificent
smoothing results of GEM, with its relatively nice preservation
of edges and sharpness. Vuescan's grain reduction seem to work
like one of the averaging filters, while GEM works like a pattern
analyzer, according to the description.

Thomas

I haven't used GEM because I have a canon FS4000 scanner and I use
NeatImage when I need to reduce grain. Does GEM do a similar job to
NeatImage?

Bruce Graham

 

[Bart van der Wolf]

I haven't used GEM because I have a canon FS4000 scanner and I use
NeatImage when I need to reduce grain. Does GEM do a similar job to
NeatImage?

I find NeatImage vastly superior to the GEM trial version I downloaded from
ASF.

Bart

 

[ThomasH]

I find NeatImage vastly superior to the GEM trial version I downloaded from
ASF.

I find NeatImage a fully different kind of product.

All examples which I saw so far look... filtered.
By myself I failed to achieve any respectable results.
My conclusion: NeatImage is a freeware because nobody
wants it as a product, rightfully so.

Thomas

 

[Andrey Tarasevich]

I find NeatImage a fully different kind of product.

All examples which I saw so far look... filtered.
By myself I failed to achieve any respectable results.
My conclusion: NeatImage is a freeware because nobody
wants it as a product, rightfully so.
...

NeatImage is not a freeware. It is $60 for Pro edition. And from what I
can see, the product is extremely successful form commercial point of view.

From my experirnce, when properly used NeatImage's produces results that
are incomparably superior to anything GEM can offer at this time.

 

[Bruce Graham]

I find NeatImage a fully different kind of product.

All examples which I saw so far look... filtered.
By myself I failed to achieve any respectable results.
My conclusion: NeatImage is a freeware because nobody
wants it as a product, rightfully so.

Thomas

I paid $30? for the home edition of NeatImage which allows writing tiff
files (the free version is restricted to jpeg output). There was a
learning curve and it is time consuming but I have been very happy with
most of the results I have achieved. The trick is to carefully profile
the noise and then don't get too agressive with the filters. Sometimes
need to treat different parts of the image differently too (layers).

I didn't realise that GEM was available standalone. Given the polarised
opinions from respected posters above, I think I need to download and
experiment and find out which works best for me!

Thanks everybody.

Bruce Graham

 

[ThomasH]

A reminder: the primary question was about Grain Removal
of Vuescan versus GEM, not about NeatImage, but lets
conclude about this anyway

I paid $30? for the home edition of NeatImage which allows writing tiff

Thanks for the hint, yes, indeed it is not a freeware now,
I stand corrected.

Maybe I gave up too quickly before I managed to use the tool
properly? Partly though because I do not see the need for it
in my particular setting: I am getting perfect results with
GEM as I scan. I do not need to work on images, analyze their
noise, tune filters etc etc. Result is as smooth as it gets
under most detailed visual scrutiny, images are tack sharp
on the screen and also after I print them on my Canon S9000.

files (the free version is restricted to jpeg output). There was a
learning curve and it is time consuming but I have been very happy with
most of the results I have achieved. The trick is to carefully profile
the noise and then don't get too agressive with the filters. Sometimes
need to treat different parts of the image differently too (layers).

I didn't realise that GEM was available standalone. Given the polarised
opinions from respected posters above, I think I need to download and
experiment and find out which works best for me!

Good choice. GEM is now available as a Photoshop plugin, not as
a standalone application. As far I recall from my communications
with ASF (before they became a part of Kodak,) GEM bundled with
scanners uses though somewhat hardware during scan and that's
why the plugin is not precisely the same software...


To summarize: the subject here was *OBTAINING SMOOTH SCANS*
from the scan process, not working extra on images to remove
grain or noise because the scanning software failed to do so!

Thomas

 

[Bart van der Wolf]

SNIP

To summarize: the subject here was *OBTAINING SMOOTH SCANS*
from the scan process, not working extra on images to remove
grain or noise because the scanning software failed to do so!

Looking at the subject line, I thought the subject was grain reduction?
A good scan with the LS40 will show the grain structure in film, thus the
need to post-process. Whether VueScan Grain reduction, GEM, NeatImage or
another tool is used seems unimportant to me, as long as the result is good.

Do I hear someone calling for a shoot-out on grain reduction?

Bart

 

[Greg Campbell]

Bart van der Wolf wrote:

Do I hear someone calling for a shoot-out on grain reduction?

For those who might not have seen this before...

http://www.michaelalmond.com/Articles/noise.htm
GEM is missing, but most other apps/actions are tested.

-Greg

 

[Dominique Deleris]

On Thu, 29 Jan 2004 14:16:40 +0100, "Bart van der Wolf" <[email protected]> said:

Bart> Looking at the subject line, I thought the subject was grain reduction?
Bart> A good scan with the LS40 will show the grain structure in film, thus the
Bart> need to post-process. Whether VueScan Grain reduction, GEM, NeatImage or
Bart> another tool is used seems unimportant to me, as long as the result is good.

Bart> Do I hear someone calling for a shoot-out on grain reduction?

Yes indeed, my question is all about grain reduction.

For me, Vuescan clearly outstands the native Nikon scanner
software in all aspects (speed in the first place, and then also
batch facilities, colors...) EXCEPT for the grain reduction
aspect... Digital GEM is far better than Vuescan's algorithm.

I was also wondering why I don't see any difference while setting
different levels of grain reduction in Vuescan: maybe I've missed
something?

I did not know at all about NeatImage. It seems pretty cool, and
turns out to run properly under Linux with Wine (at least version
2.6, I happen to have problems with the latest 3.0). It doesn't
bother me to post-process my image for grain. What I do want, is
to do all my scanning and editing business under Linux.

I must admint I whished Vuescan could do it all, but...

Come on Ed, maybe you can do something ;-)

 

[Erik Krause]

Hello, Dominique Deleris
you wrote...

I am disappointed to discover that Vuescan's grain reduction:
1. is far from GEM in terms of results
2. does not even seem to have any effect on my scans

What vuescan version do you use? There is still bug that was introduced
in version 7.6.71 preventing grain reduction to work if infrared clean
is on.

Grain reduction medium should be visible even in preview window.

 

[Dominique Deleris]

On Thu, 29 Jan 2004 22:41:00 +0100, Erik Krause <[email protected]> said:

Erik> What vuescan version do you use? There is still bug that was introduced
Erik> in version 7.6.71 preventing grain reduction to work if infrared clean
Erik> is on.

Erik> Grain reduction medium should be visible even in preview window.

Hmmm, from what I can see in the changelog, the problem you
mention should be corrected in 7.6.73. I am using 7.6.75 :-¦

 

[Kennedy McEwen]

Do I hear someone calling for a shoot-out on grain reduction?

Possibly. ;-)

There simply is no means of implementing true grain reduction without
losing some of the image content as well and, generally speaking, the
more grain you get rid of the more fine detail you lose as well. You
can reduce grain aliasing without loss using some of the techniques
which are discussed on this forum from time to time, but not the
residual true grain.

I have looked at NeatImage as an alternative to GEM and spent a lot of
time trying to optimise it. Compared to GEM, NeatImage implements a
much more complex algorithm which is matched to the spatial spectrum of
the grain in the image itself. On the other hand, GEM uses a filter
with one of 4 filter settings. NeatImage gets a *much* cleaner image
than GEM, reducing visible grain almost entirely - but takes much longer
to implement. However, NeatImage also loses *much* more fine detail
from the scan than GEM does and can result in a very strange plastic
look to the result. That may be fine in some cases where smooth
surfaces or sky dominates, but it is particularly bad on areas which
have naturally fine structures, such as trees and fine foliage etc. I
have a lot of images of waterfalls and, to be honest, it is impossible
to get a realistic looking result from NeatImage, especially when you
run a side by side comparison with the unprocessed image. By contrast,
although residual grain is present in the GEM scans, they look a lot
more realistic.

 

[Kennedy McEwen]

Bart van der Wolf wrote:



For those who might not have seen this before...

http://www.michaelalmond.com/Articles/noise.htm
GEM is missing, but most other apps/actions are tested.

There's a good reason why GEM is missing from this test - none of the
test images are scanned film!

These are all digicam output and consequently contain noise, but not
grain - the two are *not* the same - whilst scanned film images contain
both noise *and* grain (and often a significant amount of aliased grain
too!).

 

[Greg Campbell]

There's a good reason why GEM is missing from this test - none of the
test images are scanned film!

I know.

Oh, DOH! I'm having a conceptual leap!
I thought GEM was a stand alone or plug-in app.

These are all digicam output and consequently contain noise, but not
grain - the two are *not* the same - whilst scanned film images contain
both noise *and* grain (and often a significant amount of aliased grain
too!).

IMO, sensor noise and grain are kissing cousins; the comparison is at
least somewhat relevant for scanner users. FWIW, NI works well as well
on my FS4K film scans as it does in the review.

- Greg

 

[Bart van der Wolf]

Possibly. ;-)

There simply is no means of implementing true grain reduction without
losing some of the image content as well and, generally speaking, the
more grain you get rid of the more fine detail you lose as well. You
can reduce grain aliasing without loss using some of the techniques
which are discussed on this forum from time to time, but not the
residual true grain.

Depending on the scan resolution, that's right, because non-aliased grain IS
the detail in film.

SNIP

However, NeatImage also loses *much* more fine detail
from the scan than GEM does and can result in a very strange plastic
look to the result.

I usually hold back on NeatImage's grain reduction. Restraint is the key,
and too much reduction will make the image look like plastic. Some
graininess will also reduce the chance on posterization in postprocessing
and output.

That may be fine in some cases where smooth
surfaces or sky dominates, but it is particularly bad on areas which
have naturally fine structures, such as trees and fine foliage etc.

The IMHO best way is to use layers. Mask the areas in the "Neat" layer that
allow noise to be mistaken for detail, allow to clean-up the areas that have
little natural detail (but don't over-do it). Also, masks lets one partially
redo stuff at will.

Bart

 

[Kennedy McEwen]

IMO, sensor noise and grain are kissing cousins; the comparison is at
least somewhat relevant for scanner users.

Only inasmuch as they are unwanted artefacts on the recorded image. The
significant difference, certainly as far as any removal algorithm is
concerned, is that grain - or rather dye clouds - are sampled by the
sensor and can therefore extend over several pixels whereas the noise on
a pixel is completely independent of the adjacent pixels. Indeed, it is
a requirement of grain aliasing reduction that a dye cloud subtends
several pixels. As such, the spatial spectra of film grain and noise
are completely different.

FWIW, NI works well as well on my FS4K film scans as it does in the
review.

On some images that is true, but on many that I have tried it on a
direct comparison with the original reveals significant distortion, as I
mentioned in my other post on this thread.

 

[Kennedy McEwen]

Depending on the scan resolution, that's right, because non-aliased grain IS
the detail in film.

Ultimately that becomes the case but I don't think we are actually at
that point for fine grained film, such as Velvia or Reala, with any of
the desktop scanners on the market today.

SNIP

I usually hold back on NeatImage's grain reduction. Restraint is the key,
and too much reduction will make the image look like plastic.

Indeed, but to get acceptable results I find that the grain reduction
has to be reduced in NeatImage to a level which is comparable with GEM.
Since GEM is much faster than NeatImage, and integral to the scanner
driver thus requiring no user intervention other than its selection, it
is my preferred option.

 

[Erik Krause]

Hello, Dominique Deleris
you wrote...

Erik> Grain reduction medium should be visible even in preview window.

Hmmm, from what I can see in the changelog, the problem you
mention should be corrected in 7.6.73. I am using 7.6.75 :-¦

It should yes, and it was in test version Ed provided after I reported
the bug to him I found in 7.6.71. Apparently this fix was kicked out by
accident. I tested 7.6.73 and 7.6.75.

Try it: switch off infrared clean and you'll see how grain reduction
works...