ðòïåëôù 


  áòèé÷ 


Apache-Talk @lexa.ru 

Inet-Admins @info.east.ru 

Filmscanners @halftone.co.uk 

Security-alerts @yandex-team.ru 

nginx-ru @sysoev.ru 

  óôáôøé 


  ðåòóïîáìøîïå 


  ðòïçòáííù 



ðéûéôå
ðéóøíá












     áòèé÷ :: Filmscanners
Filmscanners mailing list archive (filmscanners@halftone.co.uk)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: filmscanners: RE: filmscanners: VueScan clipping & flat images





> -----Original Message-----
> From: owner-filmscanners@halftone.co.uk
> [mailto:owner-filmscanners@halftone.co.uk]On Behalf Of Rob Geraghty
> Sent: 23 November 2001 04:09
> To: filmscanners@halftone.co.uk
> Subject: filmscanners: RE: filmscanners: VueScan clipping & flat images
>
>
> Jawed wrote:
> > I would agree with this.  The intention is quite
> > clearly to make the data fill the range of possible
> > values.  For reasons analogous to the use of 16-bit
> > scans (really 10, 12 or 14 bits, generally): to
> > maximise tonal smoothness and provide resilience
> > under further editing.
>
> Maybe I'm misreading this, but something I've wondered is whether
> any scanners
> actually remap the data.  It's hard to describe without a diagram
> - suppose
> you have a monochrome scanner with an 8 bit A/D.  The A/D can only produce
> numbers from 0 to 255.  If the actual image information is all
> between (say)
> 10 and 245 then you've effectively lost 8% of the resolution possible from
> 8 bits.  But if you remap that range so that 10->0 and 245->255, you're
> "stretching" the values in between.  I presume this "stretching" is what
> adjusting levels actually does.  But presumably the fewer bits from the
> A/D, and the smaller the input range, the poorer the result would be.

I dare say you're correct.  If the scanner is, internally, 8-bit, then some
data values have gone "unused" in your example.  BUT, because of exposure
control in scanners, the scanner and its controlling software in your
workstation *may* manipulate the exposure (brightness of the light or
duration of exposure per scan step) and/or manipulate the gain of the
electronics (like the volume control on your hifi).  So the end result is,
frankly "there's no predicting"!!!!  What happens when the image "naturally"
contains more than 8 stops of dynamic range (how does the 8-bit scanner and
associated software cope)?

>
> Does setting the black and white points cause the data to be remapped?

Probably depends on the scanner and associated software.  Some may alter the
exposure and gain, like I've described above.  When exposure and/or gain
don't respond to your changes to "levels" then, with a notionally 8-bit
scanner, you will be "losing data" (tonal smoothness).

The gamma function that is used to convert from scanner RGB space to a
normal graphics datafile space (say, sRGB which is defined with an in-built
gamma factor of 2.2) has the side effect of "squeezing" the data in a
non-linear way.  The end result is that shadow detail has coarse tonality,
and highlight detail has very fine tonality.  So, when your scanner/software
outputs a gamma-corrected file, in doing so it effectively has to throw away
some "tonal smoothness" in the shadows in a trade for "tonal smoothness" in
the highlights.  This trade was chosen back in the mists of time when TV
systems where being designed.  Human perception to shadow detail is not
great so why waste data (bandwidth, or radio frequency space) in high
quality coding?  Keep the quality high for the bits humans see best, but
trade against the part we see least distinguishably.  The designers of
graphics software realised that the same trade can be made to squeeze more
"perceived" quality out of 8-bits than might otherwise have been achieved.

RAW files don't necessarily have gamma applied to them as far as I can tell.
Trouble is, to view the image represented by such a RAW file, you have to
apply the gamma function...

8-bit data is basically working at the limits here - IF you consider the
whole *system* from image in film, though light passing through film from a
lamp onto a CCD, through gain stages, through analogue to digital
conversion, through negative film de-masking/inversion, through device
profiling and finally to user-initiated Levels manipulations.  It works, but
I'd suggest it's only by the skin of its teeth.  Not being able to control
the exposure of an individual scan will prolly reduce the chances of getting
the best scan, by modern standards.

If you think carefully about these steps, you can see that the point in the
chain at which the gamma function comes into play can have a significant
effect on the overall "neutrality" of the system (freedome from introducing
arbitrary errors).  It should come at the device-profiling stage.  If you're
really lucky your scanner allows your Levels manipulations to be computed at
the same time as the device-profiling.  OR, you have a scanner that allows
you to perform primitive Levels manipulations by changing the exposure
manually (i.e. rather than letting it decide).  Changing exposure will
normally only change the black point/white point in lock-step.  It won't
change the contrast.  (Erm, unless the scanner has an S-shaped contrast
curve that is fixed in relation to exposure.  Technically, CCDs have
S-shaped contrast curves, but the scanner should be able to un-S that...)

(Though it is worth noting your results from the following experiment in
Photoshop: open a high contrast, finely toned image.  8 or 16 bit I don't
think it matters.  Now, bring up the Posterize dialog
(Image|Adjust|Posterize).  Change the Number of Levels to 128 (7-bits).
Play with the Preview check box.  Can you see the difference?  Do the same
with Number of Levels set to 64 (6-bits).  Play with the Preview check box
again.  Can you see the difference?  Hint: zoom in, and choose an area with
dark and very bright areas to examine.

I can only just see the difference with 128 levels (I have to search around
the image to find somewhere that will show a difference).  But, I reckon
that if I alternately opened two images, one the 8-bit original, say, and
the other a 6-bit posterization, I would not be able to see the difference.

So this experiment shows that on my monitor 6-bits are all that are needed
to satisfy my perception.

Naturally, your mileage might vary.  It would be interesting if peeps here
can give this a try and report what they perceive...

So, in my environment an 8-bit scanner, as long as it could maintain the 6
most significant bits faithfully, would provide me with scans that I
wouldn't be able to distinguish from a scanner with more bits at its
disposal.  The trouble is, as soon as I make levels changes, or the film is
poorly exposed and I need to manipuate the scanned image to rescue it, then
the 6-bits, even manipulated in a 16-bit file, are gonna come and bite me!)

> I presume this remapping is always post A/D.  Are there any scanners which
> attempt to adjust the input levels to maximise the range of values from
> the A/D?

The LS40 and LS4000 (used with Nikon Scan) do.  It's what happens when the
auto-exposure kicks in (I believe) which changes the brightness of the
"lamp" (there's logic for why I could be wrong - I'll let somebody else
argue the point).  They also allow the user to play with the brightness of
the lamp.  I guess some other scanners do, too.  Vuescan, with its "long
exposure pass" provides a kind of exposure manipulation which seems to work
pretty well (depending on scanner, and not all scanners support this
option).  Exposure, on its own doesn't necessarily affect the range of
values.  But any kind of S-shaped contrast curve inside the scanner or the
software means that exposure changes *will* cause different ranges to be
used.

Any compromises that you can see in exposure/contrast control, during
scanning, are made up for by modern scanners' low noise and high-bit
scanning (typically 12-14 bits).  If the scanner works internally at 12
bits, say, and Nikon Scan works with 12-bit accuracy, too, the qualitative
impact of a (real-world) shortfall in range will be, I'm sure, damned near
impossible to perceive.  If perception operates within 7-bits on a monitor
and the scanner/software is operating with a further 3-bits of headroom (say
2 are lost to system implementation compromises, I'm being cruel I hope),
then something *really gross* needs to happen to your data before you'll
perceive it.

I dare say for maximum tonal smoothness, the image should be exposed as
bright as can be stood.  This will push the data towards 255.  Because of
gamma's tonal raggedness at the 0 end, if the scan output has to use less
than full range, then it should be biased towards the top, not the bottom.
After manipulation (e.g. colour cast removal) you can then re-level the
image so that it looks "normal".  But this is all academic if you ask me,
because 8-bit images derived from 12-plus bit scanning systems stand-up to
all sorts of wear and tear without showing that the 8-bit data format was a
limitation.

In my experiments I've found, when scanning negatives, that Nikon Scan's
default exposure (Analog Gain setting 0 - this is a misnomer as it is
actually lamp intensity) provides a scan that works very well the vast
majority of the time (remember, my taste).  UNTIL, of course, the image is
badly exposed (too little/too much) or the image contains a huge dynamic
range.

If I manipulate the Analog Gain setting in Nikon Scan, I can gain highlight
detail, say (trying to rescue an over-exposed image).  But, depending on the
magnitude of the change, shadow detail may be thrown away.  Similarly, if
the image is under-exposed, an increase in Analog Gain can bring up shadow
detail, but if I try to go too far, highlight detail starts disappearing.

This appears purely to be a "compromise" within Nikon Scan (I'm quite happy
with it - plenty aren't!).  You can think of it as an S-shaped curve applied
to the scan data before the exposure measurement system takes its decision
and before the image is shown to you for Analog Gain or Levels
manipulations.

Vuescan, by comparison, can take the same image and obtain more detail in
both shadows and highlights.  Vuescan, as far as I understand, is more
conservative with its exposure choice ("analog gain") and has better
algorithms for maintaining detail in shadows and highlights concurrently.

This clearly demonstrates that there's more to a notional 8- or 12- bit
image than the scanner producing the scan.

>
> As far as I can tell, setting the black and white points in (say)
> Nikonscan
> is no different to doing the same thing to a "flat" file in Photoshop with
> Levels?

Yes.  With a minor caveat.  For some reason I haven't really got to the
bottom of, large manipulations inside Nikon Scan result in less "damage" (in
the form of exaggerated noise or posterisation) to the image than the "same"
manipulations in Photoshop.

Worse: in my experiments I have found that a gamma change of 1.2 in Nikon
Scan (3.x) is not the same as gamma 1.2 in Photoshop (6).  If you create a
single-point curve in Nikon Scan and Photoshop (e.g. 64->70) you'll notice
that the shape is different in the two applications.  Which is correct?
Dunno (though "correct" only really has meaning with reference to gamma).
Gamma (or a curve) is really tricky to be scientific with, when trying to
compare two pieces of software.  Ideally you need something dead simple like
a wedge (patches of grey), which when scanned come out as well-defined
datapoints on a histogram (admittedly each datapoint comes out a bit fuzzy).
One could then measure the difference between the two pieces of software.
Other suggestions?

(Maybe Nikon Scan's colour management is introducing significant-enough
errors into the image data, so that by the time Photoshop gets to it,
Photoshop can't help but create more damage than Nikon Scan would have
done).  I haven't experimented with CMS off.  Certainly there are reports on
this list that CMS has a detrimental affect on the quality of shadow detail
in Nikon Scan images, which if significant enough, would mean that Photoshop
was doomed to create more significant damage in any post-scan edits...)

> BTW as far as colour shifts are concerned, I found Picture Window
> Pro's ability to do levels in colour spaces other than RGB helped a lot.
>  RGB levels in PS tends to shift the colour balance, but HSL levels in PWP
> doesn't.

Well, you are forgetting that Nikon Scan (3.x) has the most unbelievably
psychedelic (and I mean that literally!) "Lightness, Chroma, Hue"  ***levels
and curves*** palette (my favourite bit being that you can tweak the
curviness of each datapoint you position on the Hue page - how the hell does
anyone get anything other than an instant high with this control?!!!).  The
lightness portion of this hippy delight is prolly just like the tool you've
found in PWP.  But also, LAB mode in Photoshop provides a Lightness channel
for the same style of manipulation (either with a curve or with levels).

Fun, aint it?

Jawed
(scanning away quite merrily as I write...)




 




Copyright © Lexa Software, 1996-2009.