Dan Margulis Applied Color Theory
16-Bit in LAB
Date: Tue, 09 Nov 2004 20:49:40 -0000
From: “pt210w”
Subject: 16 Bit images only into LAB color space
I attended a seminar at Graph Expo last month where we
were told only 16 Bit images should ever be brought into the LAB
color space for manipulation.
The presenter did not seem to have much information to
support the statement, but no one in the audience seemed to disagree
either.
Does anyone out there know what the peril could be to
working on 8 Bit images in the LAB color space?
Thanks in advance,
Pat Theobald
___________________________________________________________________________
Date: Tue, 09 Nov 2004 16:48:12 -0500
From: Jim Rich
Subject: Re: 16 Bit images only into LAB color space
On 11/9/04 3:49 PM, “pt210w” wrote:
I attended a seminar at Graph Expo last month
where we were
told only 16 Bit images should ever be brought
into the LAB color
space for manipulation.
The presenter did not seem to have much
information to support
the statement, but no one in the audience seemed
to disagree
either. Does anyone out there know what the
peril could be to working
on 8 Bit images in the LAB color space?
This is a controversial topic and some folks get all
wrapped around it.
So my .02 says this:
If you ask folks who are pro 16 bit and who possibly
have religion then you might hear you are doomed to work any other way.
The peril is supposed to be that you get
obvious banding due to image editing and then when you print you might see
it if you don1t use 16 bits. Oh yea and your histograms will have
gaps. But then again, we don't sell those histograms.
And people on the other side of the argument the 8 bit
lovers might say it doesnt matter.
The hard evidence (prints viewed by experts) I have
gathered as well as others on this list like Dan, indicates that
there is no discernable difference between 8 and 16 bit images that have
made the trip to LAB, heavily edited and then printed.
However, IMHO there is not reason to fight over this,
because of how Photoshop works today, I don't see a problem working
in either 8 or 16 bits modes in RGB or LAB or even cmyk.
Jim Rich
___________________________________________________________________________
Date: Tue, 09 Nov 2004 14:14:22 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
Here1s the deal. The wider the gamut of a color space,
the more steps between what is only being defined by 256 total steps. So
editing can cause banding and excessive data loss due to view bits
describing more data. You also have to get into LAB from something (RGB
usually) and while Photoshop does this with 20 bit precision, every time
you convert from a smaller to a wider gamut space in 8 bit, especially with
a color space gamut mismatch, you lose even more data. For example, if the
working space is Adobe RGB, which has 256 values available, converting to
8-bit Lab reduces these down to 234 values producing a loss of 22 levels.
Doing the same conversions from ProPhoto RGB reduces the values to only 225
producing a loss of 31 levels. Bruce Lindbloom a well respected color geek
and scientist has a very useful 3Levels Calculator2 that allows one to
enter values to determine the actual number of levels lost to quantization
(see the 3Calc page2 at http://www.brucelindbloom.com ).
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Tue, 09 Nov 2004 22:51:20 -0000
From: “Jen & Ron”
Subject: Re: 16 Bit images only into LAB color space
While CMYK and RGB may be related, Lab does not work
the same when working in curves in Photoshop. Therefore, the peril could be
that Lab is such a potentially powerful tool for color manipulation that
one could easily overdo things causing major problems. It is true while
doing moves in RGB/CMYK may still cause major problems if things are
overdone, it seems it’s allot easier to screw things up while in Lab.
One needs to look at their own situation, for myself - everything that
comes to me is 8bit, so 8bit is my world. Perhaps that is why no one
in the Expo group seemed to disagree. I would also add that doing some
manipulation in Lab is very beneficial to the images I work on, however I
have to admit I probably only take my images into Lab on rare occasions and
that is usually to perform some blending function with my original RGB/CMYK
image to make adjustments to a particular color channel. I would hate to
take a hacksaw to some of Dan Margulis’s chapters in his book just
because someone told me that 16 Bit images should ever be brought into the
Lab color space for manipulation. Especially if they can’t support it
very well.
Ron Scratch
RR Donnelley
___________________________________________________________________________
Date: Wed, 10 Nov 2004 03:29:26 -0000
From: “kuhammer2004”
Subject: Re: 16 Bit images only into LAB color space
A image-editor has to know his limitations (his skill
when using it) when it involves lab. While their is “color
theory” (mathematically)
from a scientific stand point. Their is also
“applied color theory” (hands-on)from the application level.
Sometimes one contradictes the other. Which one contradictes the
other?.........Trial and error.
John Opitz
___________________________________________________________________________
Date: Thu, 11 Nov 2004 07:56:41 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
Pat Theobald writes,
I attended a seminar at Graph Expo last month where we
were told only 16
Bit images should ever be brought into the LAB color
space for manipulation. The
presenter did not seem to have much information to
support the statement, but
no one in the audience seemed to disagree either.
Par for the course. Someone should have asked the
speaker to demonstrate on a real image what the terrible consequences of
working in 8-bit would be, either in LAB or elsewhere. It would then have
been discovered that the speaker had no such image available. Research the
issue, and you’ll find a lot of “experts” saying that
correcting in 16-bit is absolutely critical, creates a night and day
difference, that anyone who doesn’t do it is an amateur, etc., etc.,
etc. What you won’t find are any side-by-side examples of how working
in 16-bit creates a better result at all—let alone a huge increase in
quality—as opposed to doing exactly the same thing in 8-bit.
Most of the 16-bit discussion pertains to working in
CMYK or RGB, but the proponents, none of whom actually work in LAB or
understand very much about mathematics, argue that there are mathematical
reasons why working in LAB is worse in 8-bit. In fact, it’s the
opposite—it’s technically (although as a practical matter,
inconsequential) more damaging to work in 8-bit RGB than 8-bit LAB.
Does anyone out there know what the peril could be to
working on 8 Bit
images in the LAB color space?
Thousands of people use LAB successfully every day,
including hundreds of my own students who report back to me with complaints
and problems as they arise. If anybody was actually encountering this
problem, we’d all know about it.
Dan Margulis
___________________________________________________________________________
Date: Thu, 11 Nov 2004 07:33:29 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
on 11/11/04 5:56 AM, Dan Margulis wrote:
Research the issue, and you’ll find a lot of
“experts” saying that correcting in 16-bit
is absolutely critical, creates a
night and day difference, that anyone who
doesn’t do it is an amateur, etc., etc.,
etc.
http:
//www.brucelindbloom.com/index.html?DanMargulis.html
Also be sure to check out Bruce1s background. And no
I1ve not seen any text that says 316-bit is absolutely critical, creates a
night and day difference, that anyone who doesn’t do it is an
amateur, etc., etc2. It1s simply a reflection of math and physics.
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Thu, 11 Nov 2004 11:09:26 -0500
From: Ric Cohn
Subject: Re: 16 Bit images only into LAB color space
On Nov 11, 2004, at 9:33 AM, Andrew Rodney wrote:
And no I’ve not seen any text
that says “16-bit is absolutely critical,
creates a night
and day difference, that anyone who doesn’t do
it is an amateur, etc.,
etc”.
Gee, I sure have. Jeff Schewe was saying exactly this
a few years ago. I don’t know his current views. I assume he’s
backed away from this somewhat as I don’t think he’s at all
stupid. However, I continue to see this claimed every day.
Personally, I still think 16 bit editing has a place,
but I think there are about a hundred other editing issues that will lead
to more visibly higher quality. For the vast majority of images I
don’t believe there will ever (with today’s or tomorrow’s
reproduction methods) be a viewable improvement from working in 16 bit, yet
I see very few images that IMO couldn’t be improved from where they
end up. In my experience the big difference between capture devices (scan
or digital) is the quality of the bits. A device that outputs bad data will
look just as bad in 16 bits as in 8 bits. This is a separate issue from
whether a device needs to be able to use more than 8 bits. Any decent
camera or scanner I’m aware of already uses 12 to 16 bits under the
hood. I would agree that if we are taking the Raw data and the Raw data
needs a large correction (which it generally does) that working in 16 bits
will be necessary.
Just my 2¢.
Ric Cohn
www.riccohn.com
___________________________________________________________________________
Date: Thu, 11 Nov 2004 11:47:12 -0500
From: Bill Morse
Subject: Re: 16 Bit images only into LAB color space
Ouch!! ;^)
Thanks for this reference, Andrew. I would love
to see Dan’s response.
Regards,
Bill Morse
Digital Eye Editions
343 Medford St., Studio 2A
Somerville, MA 02145
(617) 429-3298
PS- from a “16-bit Advocate”, but
hopefully not a closed-minded one.
___________________________________________________________________________
Date: Thu, 11 Nov 2004 09:37:29 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
on 11/11/04 9:09 AM, Ric Cohn wrote:
Gee, I sure have. Jeff Schewe was saying exactly this
a few years ago.
I don’t know his current views. I assume
he’s backed away from this
somewhat as I don’t think he’s at all
stupid. However, I continue to
see this claimed every day.
I know Jeff still is a big proponent of working in
high bit.
Personally, I still think 16 bit editing has a place,
but I think there
are about a hundred other editing issues that will
lead to more visibly
higher quality.
I agree. But that doesn’t dilute both the math
behind what happens in 8 bit, or the fact that we don’t have a
crystal ball that tells us what could happen in the editing process in the
future to our files. I for one have a RIP that fully supports 16 bit files
for output (ImagePrint) driving my Epson’s. So while I doubt ink on
paper processes will ever require a high bit file, today I have a printer
that clearly takes advantage of this data. And I have no idea how many
printers will come onto the scene in a year, let alone 10 years that will
allow this.
There are users who need to edit a file and output it
once (to say a press). They really don’t need to be all that
concerned with high bit capabilities. There are others producing very
demanding output that do. It’s all about flexibility. To suggest,
based on one kind of output useage that high bit editing is smoke and
mirrors is pretty close minded considering how editing and output devices
have improved since desktop imaging started with Photoshop 1.0.7!
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Thu, 11 Nov 2004 19:14:34 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
Andrew Rodney writes,
And no I1ve not seen any text that says 316-bit is
absolutely critical,
creates a night and day difference, that anyone who
doesn’t do it is an amateur, etc., etc2.
As is so often the case when one of Andrew’s pet
color concepts goes down the tubes, his memory becomes unreliable.
In the current edition of my book, I quote at some
length a 2001 thread in AdobeForums where a number of “experts”
were on the defensive, because several participants were asking them why,
if 16-bit corrections were so dramatically better, they couldn’t
supply any images that showed this. The experts said they were too busy to
do so, but that we should take their word for it, and tried to change the
subject into a discussion of whether I was a wicked and evil person.
Apparently feeling that if they only sounded decisive enough the skeptics
might go away, they said the following things, all direct quotes, separated
by ellipses, many different speakers, I forget which one is Andrew.
"16 bit capability is critical during all aspects
of tone compression....The difference CAN be seen in the final output very
easily. Most definitely on the printed page, especially when using
high-quality halftoning and even more so to a film recorder....It’s
very easy to see that substantial color & tone editing will eventually
result in data loss and banding....If it means the difference between
taking a 16-bit image capture and editing that to the final image and
taking that same image in only 8-bit and editing that to the final image
then there is a difference like between the day and the night...Yes, if a
histogram full of holes has no impact on final output, then throw away the
graphs and just get on with the print run. However, all of us have Real
World Output showing the superiority of superior data acquisition....My
advice? Take the information you’ve read here to the bank. Stop
doubting and start applying what you’ve learned here....If you really
start out with a RAW image in high-bit form and a raw image downsampled to
8 bits, the difference really is night and day....It’s totally
obvious to anyone who looks that it’s very advantageous to do the big
moves on high-bit data."
The following year there was considerable discussion
here, in which Andrew participated vociferously, of the merits of the
following statement by one of the above experts, who presented it in
writing to a class at a major trade show:
"I’m not sure just how to stress the
importance of doing color and tone corrections in 16 bits in Photoshop. One
can point to the obvious advantages...I suppose the difference could be
characterized as professional vs. recreational use of Photoshop"
That thread can be found at
http:
//www.ledet.com/margulis/ACT_postings/ColorCorrection/ACT-16-bit-2002.htm
Now, certainly, none of these experts would be likely
to say any such thing today (although, AFAIK, none of them have just bitten
the bullet and said that they were wrong, as I hope the rest of us would if
we were in their shoes). It is, however, always useful to look back on what
they *did* say at the time, lest revisionist historians suggest that they
didn’t.
Dan Margulis
___________________________________________________________________________
Date: Thu, 11 Nov 2004 23:02:01 -0600
From: “john opitz”
Subject: Re:16 Bit images only into LAB color space
From: Bill Morse
Ouch!! ;^)
Thanks for this reference, Andrew. I would love
to see Dan’s response.
This is an old reference. This is not anything new.
Not speaking for The Man, but if you have The Book..... in Chapter 15,
starting on page 310 he explains the 8 bit vs. 16 bit. The images and the
curves The Man applied are on the cd. Also you can find it here...
http:
//www.ledet.com/margulis/PP7_Ch15_Resolution.pdf
Refer to page 16 of this pdf.
John Opitz
___________________________________________________________________________
Date: Thu, 11 Nov 2004 23:17:43 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
Bill Morse writes,
Thanks for this reference, Andrew. I would love
to see Dan’s response.
I find the whole topic so distasteful that I
ordinarily pay no attention when it arises, but since it has come up I will
make one post, but no more.
I have no objection to people who disagree with my
opinions, even when they aren’t particularly civil about it;
that’s how things get sorted out eventually.
The site in question is a different animal. It
involves a frustrated individual deliberately posting information that he
knows to be false. Plus, Andrew Rodney, who knows full well that the page
is a troll, repeatedly posts links to it.
Bruce Lindbloom’s whole contention is that
before I tested the 16-bit images, I converted them to 8-bit and then
reconverted them to 16-bit before applying curves. If that were true, it
would be like a test comparing how well two washing machines work on the
same kind of load, except that one of them had a bucket of mud poured in it
first.
Of course, it’s ridiculous. What he is seizing
on and deliberately misinterpreting is a phrase in my original description
of how I would proceed. I said I would run a series of drastic corrections
on two copies of each of many, many files. One set of corrections would be
done entirely in 16-bit and the conversion to 8-bit would come only at the
end; then, on a copy of the file, I would convert to 8-bit immediately and
load the identical actions, whereupon I would compare the two.
Makes sense so far, but I added: if it turns out that
the version done in 16-bit is significantly better, then, and only then,
I’d run a third test—converting the 8-bit file back to 16-bit
before correcting it, to see if there wasn’t something about the
calculation method that was producing the better result, rather than just
the extra data.
I’ve studied what happens when you do that, and
it’s pretty interesting. However, it never was any part of my testing
of the images, for the simple reason that the corrections done in 16-bit
all the way were never any better than in 8-bit, regardless of how drastic
they were.
Bruce asked a lot of questions about this at the time,
including correspondence with me off-list. I was aware then that he was
trolling, because he was professing to be neutral on the subject, whereas
on the ColorSync list he had recently called 16-bit correction “a
must” in certain circumstances. So, when I saw he was trying to poke
a hole in the methodology, I made sure that he grasped completely what the
purpose of the hypothetical 16>8>16 conversion was.
There is no possibility that he misunderstood this;
indeed, he could not possibly be so stupid as to think that anyone would
perform a test this way. Nevertheless, he chose to post what he did,
knowing that it was false.
While the content of the page collapses when this is
understood, there is one other side note. Bruce accuses me of keeping the
results to myself and appointing myself the sole judge, which he knows
perfectly well is not true. I printed 10 pages of side-by-side samples in
my current book, many at extreme magnifications. The versions are not
identified until a box later in the chapter, and readers are invited to
judge for themselves which is which, plus, I indicate how I voted when I
first saw the proofs and did not know which was which. The original 16-bit
files, and the corrections I applied to them, are all on the book’s
CD and anybody who likes can either verify what I did or do their own
experiments.
What prompted the accusation was an unfortunate
incident involving my first round of tests. The test files were provided by
a list member whom I did not know at the time. After I performed the tests,
I was flabbergasted to learn that he would not give me permission to
exhibit the images and results publicly, which I thought was the whole
point of the exercise. He did confirm, several times, on this list and at
least two others, that he and I were not personally acquainted; that he
believed I did not realize he was not giving permission; and that he had
seen my results and that I was describing them accurately. When Bruce
challenged him, he offered to give him pieces of the pictures (and my
results) that would have been sufficient for Bruce to verify the findings,
but Bruce refused to take them.
Because of this incident, I changed the rules. I said
I would not do any testing on 16-bit files without an advance agreement
enabling me to publish them and to release the raw data to illustrate
whatever was found. Ric Cohn and a couple of other people kindly agreed to
allow their images to be used in this fashion. Consequently the files are
available for anybody on the CD.
The upshot of this: you guessed it. With the rules now
having been altered to accommodate Bruce, Bruce wrote that the test was
invalid, because the rules had been changed.
In summary: the issue with the page is not its
dishonesty as much as its irrelevance. Bruce is the one advocating the
inconvenient workflow, just as it would be if he said that there’s a
big quality difference if one only wears garlic around the neck while
booting Photoshop up. Nobody can disprove that, because if your tests show
that there’s no difference, probably the garlic was too old or there
wasn’t enough of it.
Similarly, nobody can disprove that 16-bit correction
may be better under some circumstances, because nobody can test every
conceivable image with every conceivable tool. However, the only two people
who have ever run extensive tests—Jim Rich and I—tortured the
images almost beyond belief and still were not able to identify any areas
in which 16-bit correction did better with a color photograph.
It’s painfully apparent at this point that Bruce
and the others have never actually done side-by-side tests to support their
theories. Otherwise, we’d have seen examples long ago. For anybody
wishing to successfully advocate the 16-bit workflow, the recipe is quite
easy:
1) Here is a photographic image;
2) Here is what I did to it in 16-bit;
3) If you convert it to 8-bit first and
then repeat what I did, it looks a lot worse.
Without that demonstration, he has put up a blank
page, IMHO.
Dan Margulis
___________________________________________________________________________
Date: Thu, 11 Nov 2004 23:04:06 -0700
From: Ron Kelly
Subject: Re: 16 Bit images only into LAB color space
I must admit that I haven’t done extensive
personal testing on the question of 8 bit vs.16 bit editing.
Why not? Well, I’m getting along just fine in 8
bit. I have no problems that I know of that 16 bit editing would fix, and I
know of several problems I would have using 16 bit.
What problems are those? Much bigger file sizes,
limited editing tools, more time watching the beach ball revolve, clogged
hard drives and a lot more CDs and DVDs to burn and catalog, to name a few.
What problems would 16 bit editing solve? Well,
I suppose, I wouldn’t get as much banding in a heavily edited
file. But wait, I don’t get banding now, and I routinely do some
pretty serious curving. I can only think of one image where banding was a
problem in output. I must admit that I didn’t think to try 16 bit
editing; I just selected the sky and added a bit of noise.
Maybe I’m missing something, but I don’t
know what I’m supposed to get out of 16 bits except maybe a bit of
status with the color geeks. Anyone? Richer colors? Sharper results? More
shadow detail?
And what if you don’t make heavy edits on a
file? Is there any theoretical advantage to 16 bits here? In other words,
does the average image have anything to gain from 16 bit, or is it just the
very rare ones?
The only examples I’ve seen that purported to
show the benefits of 16 bit editing were terrible photographs to begin
with. The author (Bruce Fraser) was trying to make a silk purse out of a
pig’s ear, and it didn’t work in my opinion. If 16 bit is only
effective on this type of image, then I think I’ll do my best not to
start with a piece of crap to begin with.
My comment relating to this thread is that it seems
like another case of people reacting to what they see in the process, and
not the result of that process. Many others on this list have observed that
we don’t sell histograms to which I concur.
What I have proven to myself in my own personal work
is that I can get excellent results in 8 bits. Certainly, my files are
usually pretty good to start with; for example they’re not scans of
negatives that are two-stops underexposed. Your situation may be different,
I realize. Just don’t expect to start from this position and get a
result that’s as good as could be done properly, using only 8 bits
but with a good original.
Those of you who are using 16 bit editing, have you
proven it to yourself that it’s necessary? Vital? Beneficial?
Noticeable? Detectable? All the time? Sometimes? Ever?
If you have, but you go a long way down that
progression, perhaps you should reconsider your workflow.
And if you’re up at the front of the
progression, congratulations! You’ve just won the 8 vs.16 bit
challenge. Now please post your images for all to see.
Finally, ask yourself this question: is it not true
that if two techniques yield the same result, the better method is the
simpler, faster, the one that is a much lighter weight to bear?
Perhaps some of you are taken in by the old logic that
you used to use in photography: the bigger the camera, the better the
pictures. That doesn’t hold true here; this isn’t 4x5 vs. 35mm;
more bits aren’t necessarily better.
Cheers,
Ron Kelly
___________________________________________________________________________
Date: Fri, 12 Nov 2004 05:00:34 -0800
(PST)
From: Mike Bevans
Subject: Re: 16 Bit images only into LAB color space
I will tell you something
16 bit IS good for... charging customers by
file size. Everyone from color management
consulants, to scanning services, to camera
salesman, to photographers uses this trick.
I agree with Dan, this is a tiresome debate. I
also agree with Ron. I personallly think that only dogs can hear in 16 bit.
On a similar note, I work with the BetterLight camera,
shooting 8 bit files for fine art reproduction. No one
has ever had a problem with 8 bit files. I have encountered printers
who tell me my files are too lo-rez for their printers, since I scan
at 300ppi. The files I am producing are over 4000x5000 pixels.
I know from my own experience that you can print
photographic images banner-sized from these files.
Still, I encounter printers who want to
up-rez the files to match their print dimensions. This
makes for softer images and cumbersome file sizes. Has anyone
else had this problem sending work to print?
-Mike Bevans
___________________________________________________________________________
Date: Fri, 12 Nov 2004 07:06:19 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
on 11/11/04 5:14 PM, Dan Margulis wrote:
As is so often the case when one of Andrew’s pet
color concepts goes down the
tubes, his memory becomes unreliable.
Post a quote and a reference please.
Apparently feeling that if they only sounded decisive
enough the skeptics
might go away, they said the following things, all
direct quotes, separated
by ellipses, many different speakers, I forget which
one is Andrew.
If you1re going to put quotes here you better get your
facts straight bud. You may not remember? Then either find the exact post
and a reference or pull my name from this bogus list of quotes. Be accurate
Dan.
I1m still waiting for you to respond to Bruce
Lindblooms page.
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Fri, 12 Nov 2004 07:12:19 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
Bill Morse writes,
Thanks for this reference, Andrew. I would love
to see Dan’s response.
Well Dan wrote a lot but didn1t address anything Bruce
challenged him on. Bottom line is on this list, Dan1s right. There1s no
need to argue science. I actually considered cc1ing this to Bruce but in
the end, it1s a waste of bandwidth. Read Bruce1s page, do your own tests.
Consider the math (that1s all computers understand) and consider what will
happen to the file in the future and what devices it might be sent to. Then
consider the downsides of 16-bit editing and do what you feel is right.
Gee, I wonder why the vast majority of input (even
consumer devices) produce more than 8 bit per color????
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Fri, 12 Nov 2004 10:11:33 -0500
From: Fred D Yocum
Subject: Re: 16 Bit images only into LAB color space
The message I received seemed to do a fairly good job
of addressing the issues raised on the Mr. Lindbloom’s page.
Dan Margulis on 11/11/2004 23:17
“I printed 10 pages of side-by-side samples in
my current book, many at extreme
magnifications. The versions are not identified until
a box later in the chapter,
and readers are invited to judge for themselves which
is which, plus, I indicate
how I voted when I first saw the proofs and did not
know which was which. The
original 16-bit files, and the corrections I applied
to them, are all on the
book’s CD and anybody who likes can either
verify what I did or do their own
experiments.”
Computers have is math, humans have more flexible
forms of perception - what we perceive is not necessarily what is there or
not there. That goes for this little verbal fire storm as well as the bit
level of files. If you don’t like Dan’s results, perhaps you
can invent you own tests and publish your own results.
F D Yocum
Graphic Designer
___________________________________________________________________________
Date: Fri, 12 Nov 2004 08:12:21 -0700
From: Andrew Rodney
Subject: 16 Bit images only into LAB color space
Check out:
http:
//www.photoshoptechniques.com/forum/archive/index.php/t-336.html
This appears to be one of the first or original sets
of discussions (including the bit about recreational users). You1ll find no
posts by yours truly. You1ll also find some sensible advise from Bruce
Fraser such as:
"I have never, anywhere, claimed that 16 bits is
superior to 8 bits. What I’ve
said is that from my own experience, 16 bits gives a
great deal more editing
headroom and flexibility than 8 bits. If you
don’t need it, don’t use it. I’ve
always been pretty straightforward about the plusses
and minusses, and I’m not
trying to sell anyone on 16-bit workflows."
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Fri, 12 Nov 2004 17:20:59 +0100
From: Werner Tschan
Subject: Re: 16 Bit images only into LAB color space
Andrew Rodney writes
Gee, I wonder why the vast majority of input (even
consumer devices) produce
more than 8 bit per color????
The dynamics of modern digital cameras is such, that
you have a much wider contrast range than (most — and surely more
than any slide-) film could produce. When I process RAW files into
TIFF files I have the possibility to shift 2,5 Stops each way without
getting into serious problems (Canon EOS1Ds with C1 PRO from Phase One). In
my understanding it’s here where the more than 8 bits of the camera
come into play.
Could this be the answer to your question, or am I
getting things wrong?
Werner Tschan
___________________________________________________________________________
Date: Fri, 12 Nov 2004 09:46:13 -0700
From: Ron Kelly
Subject: Re: 16 Bit images only into LAB color space
Werner:
Try this test:
Lock your camera down on a tripod and photograph some
scene that has a full dynamic range, ie shadows with detail and highlights
with detail, maybe some speculars too.
Bracket your photographs. Choose one that appears to
be the best exposure for the scene, and choose one that is 2.5 stops darker
than that.
Open up the darker one, and manipulate it in software
to match the “best exposure” and compare the results. Use
on-screen and/or prints for your comparison, but be sure to use the
criteria that you most often have for “the final result.”
What does your comparison show? Are the results
identical?
Ron Kelly
___________________________________________________________________________
Date: Fri, 12 Nov 2004 18:49:52 -0700
From: Ron Kelly
Subject: 16 bit color
People:
I was just out jogging and I was mulling over the
whole 16 bit vs. 8 bit thing. As I rounded a curve by the river, I was
thrilled to see a gorgeous sunset.
Naturally, this got me to thinking: what color space
does God use? What bit depth?
After consulting my fellow jogger, who is very
strong in physics, we decided that God uses 64 bits. He said, (sic)
“There is just no way that you need any more than 64 bits to describe
ALL the colours, x-rays, and gluons or whatever. I don’t know exactly
how many wavelengths that is, but it’s A LOT.”
Which begs the obvious question: what if you used 128
bits for edits? I know, I know, there are no devices, software, or monitors
for this (yet!) BUT JUST THINK! Maybe someday you could have better color
than God! (VERY Big ; - ) )
I don’t really want to re-ignite the whole
thing, but I just couldn’t resist.
God Bless all of you who contribute; we may not
all have the same opinion but that doesn’t mean we can try to
understand each other, and have a little bit of fun too.
Sorry for wasting your bandwidth,
Ron Kelly
___________________________________________________________________________
Date: Sat, 13 Nov 2004 07:13:34 -0800
From: Mike Russell
Subject: Re: 16 bit color
Ron Kelly wrote:
Naturally, this got me to thinking: what color space
does God use?
What bit depth?
God uses double precision floating point. At the
end of the day, unlike His creations, God is big on pictures and scant on
words.
BTW, He is not big on upgrades and is still using an
early version of CP/M, running on an overclocked Sol, with hard sectored 8
inch floppies.
Mike Russell
www.curvemeister.com
www.geigy.2y.net
___________________________________________________________________________
Date: Sat, 13 Nov 2004 10:33:44 -0500
From: David Deaubrey Tighe
Subject: Re: 16 bit color
Makes me wonder what color looked like a few billion
years ago.
Was Sol redder or more yellow? Were natural dust and
volcanic dust filters
warmer than our carbon filters? Moreover, what sort of
profile would it take
to match the cones of early hominids?
___________________________________________________________________________
Date: Sat, 13 Nov 2004 11:27:46 -0500
From: Jim Rich
Subject: Re: 16 bit color
Which begs the obvious question: what if you used 128
bits for edits? I
know, I know, there are no devices, software, or
monitors for this
(yet!) BUT JUST THINK! Maybe someday you could have
better color than
God! (VERY Big ; - ) )
Ron,
At first seeing your amusing comments
about God and higher bit depth I wrote what I had hoped to me would be an
entertaining response:
But what if you don1t believe in God. Then are you
stuck with those devilish 8 bits?
Then it became clearer that you are on to
something here that has been alluded to all along. This is a true
religious experience for a lot of folks who are on both sides of the issue.
The 8 v 16 bits argument is now is about his or her belief system and
for those people who know they have religion because they have been
to the mountain, there is a need to stand their ground no matter what the
evidence says. And that this argument is more about peoples core beliefs
(and egos) than it is about technology and facts.
Sounds just like the our last presidential election.
My .02.
Jim Rich
___________________________________________________________________________
Date: Sat, 13 Nov 2004 16:51:32 -0000
From: “Bob Frost”
Subject: Re: 16 bit color
There was none. Color only exists in the mind of an
suitably-equipped observer!
Bob Frost.
From: “David Deaubrey Tighe”
Makes me wonder what color looked like a few billion
years ago.
___________________________________________________________________________
Date: Sat, 13 Nov 2004 08:41:48 -0800
From: Mike Russell
Subject: Evolution and color spaces
David Deaubrey Tighe wrote:
Makes me wonder what color looked like a few billion
years ago.
Was Sol redder or more yellow? Were natural dust and
volcanic dust
filters warmer than our carbon filters? Moreover, what
sort of
profile would it take to match the cones of early
hominids?
There’s no need to guess about this, as this is
dictated by the molecular structure of the light senstitive dyes, and they
have a well-documented time line in our DNA, and that of other species.
Chevreaux documented the adaptive nature of color vision 100 years
ago - we use something more like individual curves, not profiles, for our
vision.
Looking at our retinas under a microscope, cones come
in three colors, cyan, magenta, and yellow. Rods, which are the more
primitive sensor, are senstive to monochrome only, and used for edge
detection at normal light levels. So, it may be argued, with
anatomical support, that we are CMYK creatures, with sharpening in the K
channel, not RGB ones.
There’s nothing magical about three cones.
Multispectral is the order of the day for many animals. Certain
birds, including Mallards have five different light sensitive pigments.
Add the rods, and you have hexachrome or hifi color.
It would not take much, just the presence of a fourth
type of cone, for a person to have multi-spectral vision - and such people
may be walking among us in small numbers. Perhaps a member of this
discussion group could design a billboard that would be readable only by
these people.
Mike Russell
www.curvemeister.com
www.geigy.2y.net
___________________________________________________________________________
Date: Sat, 13 Nov 2004 12:34:35 -0500
From: David Deaubrey Tighe
Subject: Re: Evolution and color spaces
Looking at our retinas under a microscope, cones come
in three colors, cyan,
magenta, and yellow. Rods, which are the more
primitive sensor, are
senstive to monochrome only, and used for edge
detection at normal light
levels. So, it may be argued, with anatomical
support, that we are CMYK
creatures, with sharpening in the K channel, not RGB
ones.
Mike,
Wow, interesting stuff.
Awhile back I was reading something about
color-blindness (specifically what is called red-green color blindness). I
recall the article said it should more accurately be called
“magenta-green” color blindness. This explanation made color
blindness sound like it was occurring in a LAB model.
I don’t know anything about the biology of
color, but it’s fascinating.
David
___________________________________________________________________________
Date: Sat, 13 Nov 2004 18:34:00 -0000
From: “Bob Frost”
Subject: Re: Evolution and color spaces
David,
An excellent book to read is ‘Vision and Art -
The Biology of Seeing’ by Margaret Livingstone (Harvard
neurobiologist).
Bob Frost.
___________________________________________________________________________
Date: Sat, 13 Nov 2004 14:04:53 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
Mike Bevans writes,
I have encountered printers who tell
me my files are too lo-rez for their printers,
since I scan at
300ppi. The files I am producing are over
4000x5000 pixels. I
know from my own experience
that you can print photographic
images banner-sized from these
files. Still, I encounter
printers who want to up-rez
the files to match their print
dimensions. This makes for softer
images and cumbersome file
sizes. Has anyone else had this problem sending work
to print?
Certainly. This goes hand in hand with Terry’s
post yesterday. There are some very knowledgeable printers who hang out on
this list and elsewhere who know most of what there is to know about
prepress. As I remarked yesterday, however, the chances of finding a
printer who knows what his own dot gain is or has any idea of what black
generation to use are not high. Similarly, few printers have any clue as to
what resolution is necessary.
What you get, therefore, is some minimum-wage person
who runs the incoming job through FlightCheck or something similar and
demands that whatever the program thinks it requires be complied with.
Since this is the kind of treatment that *I* often get
from printers (who often get all huffy when I say that the resolution is
sufficient, and tell me I should go off and read a book called
“Professional Photoshop” before having the gall to challenge a
printer on what they do best), I can only imagine what a PITA it is for
other people.
Unfortunately, there’s nothing for it but to
improve one’s own knowledge—if you don’t know more about
image prep than the typical printer does, there’s a problem. In the
this-is-hard-to-believe department, I offer just three examples:
*******************************
1) My publisher calls me up asking how to deal with
another author’s book. The book is about Excel or some other
non-graphics software. There are no color images other than screen grabs,
and there are about 500 of them. It’s on press, but the printer says
it won’t print the book unless each and every one is upsized to 300
ppi. Author refuses. Publisher being berated by printer for providing
inadequate files. It actually takes a call from me explaining what the
purpose of resolution is, and why 300 ppi is three times higher than what
you need for a screen grab, before the printer will relent.
*******************************
2) I get a call in the middle of the night during the
REPRINT of one of my books—that is, after the book has already
printed successfully, using the same exact files. Following dialog ensues:
Printer’s CSR: Sir, I’m sorry to wake you,
but we’ve had to stop the pressrun because there’s a problem
with one of your files. It doesn’t have enough resolution and the
pressman is saying that it looks like [doo-doo] in print.
DM: What picture is it?
CSR: It’s the one on page 147, top left.
DM: I do love my book but not quite enough to have a
bleeping copy of it here with me in bed. What is the picture of, please?
CSR: I don’t know. The pressman didn’t
tell me.
DM: Well, do you have the bleeping dyluxes in front of
you?
CSR: Yes, sir.
DM: What does the caption for that picture read?
CSR (reading): “This image illustrates the
perils of printing with inadequate resolution.”
*******************************
3) The following is unsolicited correspondence:
Subj: Question concerning Levels.
Date: Friday, February 21, 2003 10:48:15 PM
To: dmargulis@aol.com
Hello Dan,
I just purchased your 4th edition of Professional
Photoshop. It is just as friendly and wonderful to read as the first one. I
work in a Screen print shop and rarely get to use a majority of your cool
moves. Earlier today, while studying your book and duplicating your moves,
the Production Manager tried to explain to me why the moves wouldn’t
work for high end printing. His evidence laid in the levels dialog box.
After the moves, the levels would show up spiky and with gaps. He insisted
the image would print like crap. He had me enlarge the image by one pixel
and upon checking the levels, they were now smoother. I looked at him like
he was crazy and muttered that I would look into it. My question to you, is
he crazy or does he have some sort of valid point?
*******************************
You can’t make this stuff up.
Dan Margulis
___________________________________________________________________________
Date: Sat, 13 Nov 2004 19:22:38 -0000
From: “Bob Frost”
Subject: Re: Evolution and color spaces
Mike,
Looking at our retinas under a microscope, cones come
in three colors, cyan,
magenta, and yellow. Rods, which are the more
primitive sensor, are
senstive to monochrome only, and used for edge
detection at normal light
levels. So, it may be argued, with anatomical
support, that we are CMYK
creatures, with sharpening in the K channel, not RGB
ones.
The actual peak response of the three types of cones
in our eyes is in the violet, green and yellow regions, so that perhaps
makes us VGY creatures!. The rods also have peak response in the green
region. Fish apparently do have cones that peak in the red region, but we
don’t.
But it is not so simple since each cone has a very
broad response to light. The longwavelength (or ‘Blue’) cones
peak in the magenta but respond over the range from UV to green. The middle
wavelength (or ‘Green’) cones peak in the green but respond
over the whole visible range of UV to red, as do the shortwavelength (or
‘Red’) cones which peak in the yellow but respond more strongly
in the red than do the middlewavelength cones. This means we can’t
tell anything about the color of the light from the response of a single
cone type. Only by comparing the ratios of the responses of all three cone
types can the brain work out what wavelengths of light have stimulated
them.
It would not take much, just the presence of a fourth
type of cone, for a
person to have multi-spectral vision - and such people
may be walking among
us in small numbers.
In a way, rods are our fourth type of color receptor!
Which is why colors seem different at dusk when the rods come into play and
our cones become less efficient. During the daytime when we are using
mainly our cones for color and luminance detection, if we look at a blue
bowl with red cherries in, it appears that the cherries are brighter than
the bowl. But at dusk as our rods play a greater part in detecting
luminance, the blue bowl becomes brighter than the red cherries, because
the rods are more sensitive to blue light than to red!!
Bob Frost.
___________________________________________________________________________
Date: Sat, 13 Nov 2004 16:18:59 -0500
From: todie
Subject: Re: Evolution and color spaces
On Nov 13, 2004, at 11:41 AM, Mike Russell wrote:
Looking at our retinas under a microscope, cones come
in three colors, cyan,
magenta, and yellow. Rods, which are the more
primitive sensor, are
senstive to monochrome only, and used for edge
detection at normal light
levels. So, it may be argued, with anatomical
support, that we are CMYK
creatures, with sharpening in the K channel, not RGB
ones.
This is fun and instructive, but how come we can see
way outside the current CMYK spaces?
(I’m aware that “current” may be
part of the answer)
Laurentiu Todie
___________________________________________________________________________
Date: Sat, 13 Nov 2004 17:08:05 -0600
From: “Mike Davis”
Subject: RE: Color bit depth
Backing away from the trees here a minute and looking
at things as a relative novice, I seem to recall that the ability of the
human eye and brain to recognize (perceive) adjacent similar color
differences is quite limited. If one uses 8 bit colors and generates
sufficient adjacent color differential as the direct result of an edit,
then there may be a case for the use of 16 bit color. However, at the
pixel level, one cannot discern close adjacent color differences without
extreme magnification and an electronic meter. The eye blends colors
quite well and uses a built in AWB to compensate for what we
“think” the right color is or should be. Even at 72 ppi
on a decent monitor, adjacent pixels are blended, thus leaving the 8 vs. 16
argument moot for the most part.
Having said that, I can also imagine an image with an
incredibly narrow density range being artificially expanded with a curve
(Dan’s fans don’t use levels :-)) which would stretch the gaps
between colors and create a wider range between adjacent pixels (black cat
in a coal bin, polar bear on ice), which in turn would show up in favor of
the 16 bit camp after some extreme editing to prove a point. I
suspect that, mathematically, one could crunch some numbers to make a case
for 16 bit under extreme editing conditions. I also suspect that in
the real world, 8 bit is indistinguishable from 16 bit with
“normal” images. I have tried some tests similar to
Dan’s and have yet to “see” anything that would encourage
me to change to 16 bit files. I’m still testing. In the
real world, I don’t stretch black objects from black thru gray to
white.
Mike Davis
mldavis2 AT sbcglobal DOT net
___________________________________________________________________________
Date: Sat, 13 Nov 2004 23:36:37 EST
From: Dan Margulis
Subject: Re: Evolution and color spaces
David Tighe writes,
Awhile back I was reading something about
color-blindness (specifically what
is called red-green color blindness). I recall the
article said it should
more accurately be called “magenta-green”
color blindness. This explanation
made color blindness sound like it was occurring in a
LAB model.
That would be me. I tested a group of color-blind
people and found that their condition could best be approximated as an
extreme inability to perceive variations in the A channel. The group
appeared to be at least as sensitive as normally-sighted people to changes
in the B channel (further testing needed to say this for sure) but even the
ones with the closest to normal vision could not see a difference in the
image when 50% of the contrast in the A channel was removed. The ones who
were most spectacularly colorblind could not see a difference even when the
A channel was inverted, i.e. magenta objects turning green and vice versa.
Dan Margulis
___________________________________________________________________________
Date: Sun, 14 Nov 2004 00:17:56 -0500
From: David Deaubrey Tighe
Subject: Re: Evolution and color spaces
That would be me. I tested a group of color-blind
people and found that their
condition could best be approximated as an extreme
inability to perceive
variations in the A channel.
Dan,
Yes, now I recall you discussing that. Sorry I
couldn’t give you attribution in my original paraphrase.
Was that in your book or from online? I was going
through some old files this week from CompuServe’s Pre Press Forum
and perhaps you mentioned it there?
My graduate advisor, a painter, is color blind.
I’ve also discussed some of these interpretations with him. FYI, I
used to be a commercial pre press guy, now I am an artist/printmaker and
mixing the old world methods with today’s high tech.
David
___________________________________________________________________________
Date: Sun, 14 Nov 2004 09:23:38 EST
From: Dan Margulis
Subject: Re: Evolution and color spaces
David Tighe writes,
Was that in your book or from online? I was going
through some old files
this week from CompuServe’s Pre Press Forum and
perhaps you mentioned it
there?
It was in a column in Electronic Publishing.
Dan Margulis
___________________________________________________________________________
Date: Sun, 14 Nov 2004 10:03:21 EST
From: Dan Margulis
Subject: Re: Color bit depth
Mike Davis writes,
Having said that, I can also imagine an image with an
incredibly narrow
density range being artificially expanded with a curve
(Dan’s fans don’t
use levels :-)) which would stretch the gaps between
colors and create a
wider range between adjacent pixels (black cat in a
coal bin, polar bear
on ice), which in turn would show up in favor of the
16 bit camp after
some extreme editing to prove a point.
I thought that such a thing would happen myself, which
is why I included exactly such images in my tests. To my surprise, there
was no advantage to editing in 16-bit even in such an extreme case. If
anybody else has tried a side-by-side test on such an image and gotten a
different result, I’d be happy to hear about it.
Dan Margulis
___________________________________________________________________________
Date: Sun, 14 Nov 2004 10:00:52 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
Andrew Rodney writes,
This appears to be one of the first or original sets
of discussions (including
the bit about recreational users). You1ll find no
posts by yours truly.
You1ll also find some sensible advise from Bruce
Fraser such as:
The argument concerns Andrew’s revisionist
claims that he has never heard anyone (like, for example, his own
business partners) say that correction in 16-bit creates a night-and-day
difference, or that anyone who doesn’t work that way is a
recreational as opposed to a professional user. I posted two lengthy
quotes showing that he had in fact heard exactly those things from
his friends. Now, we get the above—a reference to an unrelated
group posted a year after the one I cited, where the poster has already
changed his mind considerably.
Because of the possibility that the group may be
misled by this obvious red herring, I reluctantly post the complete,
unedited text of the 2001 posting by the person Andrew now quotes as
taking a soft line on the issue. A reminder that this individual was only
one of several persons posting substantially similar comments at the
time. And, as Andrew’s later post shows, neither this person nor
anybody else in their right mind would say such a thing today. I trust this
brings this episode to a close.
NOTE: With respect to the first paragraph, the poster
was immediately corrected publcly by the person providing me with many of
the raw scans used in my book. The originals *were* untouched, *were* in
16-bit, and had not been toned in any way, whether by setting white and
black point or any other modification. Some of the tested moves were almost
inconceivably extreme, far more than would be encountered in any plausible
real-life scenario. Therefore, they are precisely the kind of files that
the speaker then thought would be so much better in 16-bit that “the
difference really is night and day” and “it’s totally
obvious to anyone who looks.” As those who have read my book know, it
is in fact extremely difficult to tell the two sets of images apart, even
at high magnifications.
My apologies to the group for the waste of bandwidth.
Dan Margulis
*********************************
Bruce Fraser - 02:58pm Aug 12, 2001 Pacific (#46 of
52) I think a lot of people are overlooking the fact that Dan is playing
with a stacked deck here. I very much doubt that his “tests”
start out with a raw scan from an 8-bit capture device. At very least,
black and white points are probably being set in the scanner, probably in
12-bit space. Maybe some rough tonal shaping too.
In which case, the debate simply becomes one of
whether it’s advantageous to do gross corrections in the scanner
software or in Photoshop.
I started using a primarily 16-bit workflow primarily
because most scanner software did such an abysmal job of transferring color
correctly to Photoshop, and because I don’t like making critical
image decisions on a postage-stamp-sized prescan.
From a quality standpoint, it makes little difference
whether you carry out the manipulations on the high-bit data in the scanner
software or in Photoshop. I prefer doing it in Photoshop because it gives
me more control than most scanner software (or maybe, honestly, because I
know Photoshop’s controls better than I know any scanner software),
and because I can see every pixel in Photoshop.
If you really start out with a RAW image in high-bit
form and a raw image downsampled to 8 bits, the difference really is night
and day. But very few people do anything of the kind.
Those of us who have adopted a 16-bit workflow in
Photoshop have done so because we’re less likely to make dumb moves
based on the postage-stamp preview, and because high-bit files in Photoshop
give us much more editing flexibility and headroom than an 8-bit file
produced from manipulating high-bit data on a scanner or digital camera.
But if someone prefers to do the gross manipulations in the scanner or
camera software, I won’t argue with them. It doesn’t really
matter, beyond personal preference and skills, where one manipulates the
high-bit data.
But it’s totally obvious to anyone who looks
that it’s very advantageous to do the big moves on high-bit data.
*********************************
___________________________________________________________________________
Date: Sun, 14 Nov 2004 10:29:59 -0500
From: “Iliah Borg”
Subject: Re: Color bit depth
On Sun, 14 Nov 2004 10:03:21 EST
Dan Margulis wrote:
I thought that such a thing would happen myself, which
is why I included
exactly such images in my tests. To my surprise, there
was no advantage to editing
in 16-bit even in such an extreme case. If anybody
else has tried a
side-by-side test on such an image and gotten a
different result,
I’d be happy to hear about it.
I’m afraid that rather crude math used in some
image editing programs may be one of the reasons.
Best regards,
ib
___________________________________________________________________________
Date: Sun, 14 Nov 2004 10:22:18 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
on 11/14/04 8:00 AM, Dan Margulis wrote:
The argument concerns Andrew’s revisionist
claims that he has never heard
anyone (like, for example, his own business partners)
say that correction in
16-bit creates a night-and-day difference, or that
anyone who doesn’t workthat
way is a recreational as opposed to a professional
user.
No, I said I never said that and challenged you to
quote me on saying that. Big difference.
And, as Andrew’s later post shows, neither
this
person nor anybody else in their right mind would say
such a thing today. I trust
this brings this episode to a close.
That1s an assumption only Dan would make. I suggest
you contact Mr. Schewe and see if indeed he1s of that belief or not. I
posted a reasonable and rational response from Bruce Fraser from that 2001
post. But again, you1ll have to ask Bruce if he still feels that way or
not. Unlike some, I try not
to put words in other people1s mouth or misquote.
This debate which has gone on for years in this and
other forums is simply getting to the point that it1s like a religious
belief of both parties. It1s pointless.
Here are some facts and some color theory:
Fact. Nearly all tone and color corrections (color
space conversions included) cause data loss. It1s simple math. This can1t
be disputed. The question becomes can anyone see this data loss? See it
where? On what output device? At one point in the editing process? The fact
that data loss happens is not something that should keep users awake at
night worrying about but it should be something they recognize.
You1re taking an image and printing out to a halftone
dot once (as far as you1re concerned). Should you be alarmed that editing 8
bit data will cause data loss and be seen? From some or the more open
minded users who have posted on this subject, the answer based on their
experiences is clearly no. They point out that 16 bit files take up double
the file size and slow down processing and workflow and demand more storage
space. Now space isn1t a big issue for some, it is for others. Processing
speeds are getting faster all the time but none the less, that cost money
and takes time. Some have pointed out that the editing options in 16 bit
are less then 8 bit. True indeed. Not theory, fact. Photoshop CS opened up
a lot more options in respect to what users can do in 16 bit. That doesn1t
alter the facts about workflow, speed and storage.
Now despite the fact that some only view the world in
CMYK ink on paper, it1s a fact that some users have far greater needs from
their data. There are users who like Mr. Schewe will not only be doing on
going editing to images and printing them out to all nature of devices we
have today, but devices that are now in the lab. In 2001, when the quotes
above were made, there were few devices that would actually accept and use
16 bit data at print time. Today I can drive my 2200 with a RIP
(ImagePrint) that supports, uses and produces better results from 16 bit
files. I didn1t know this would be possible in 2001. I don1t know what will
be possible in 2006. IF my goal is to have a file that retains as much
useful data today and in the future (which is far cry from the service
bureau or print shop), my needs are vastly different.
Let1s not forget what1s coming (very soon) for some
users. HDR! An example can be seen here:
http://www.ict.usc.edu/graphics/HDRShop/
Some users produce imagery that requires by their
choice the least amount of data loss in order to keep the data as pristine
as possible for present and future uses. Others don1t care and have to
handle data quickly and move on. NEITHER is right for the other user. One
does paint the user into a lesser corner then the other. That1s not a color
theory, that1s a color fact. This argument over bit depth is like the
current argument now in vouge (shoot RAW or let the camera process the RAW
data on the fly and toss the RAW away).
The latest take on high bit editing I1ve seen from
Bruce Fraser is in his fine book “Real World Camera RAW2 page 8. It
takes both debates and puts them into a useful perspective:
3Don1t be overly afraid of losing levels-it1s a normal
and necessary part of image editing, and it1s effect can be greatly reduced
by bringing images into Photoshop in 16 bit/channel files rather then 8
bit/channel ones-but simply be aware of the destructive potential Photoshop
edits.2
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Sun, 14 Nov 2004 10:19:08 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Dan Margulis
I thought that such a thing would happen myself, which
is why I included
exactly such images in my tests. To my surprise, there
was no
advantage to editing
in 16-bit even in such an extreme case. If anybody
else has tried a
side-by-side test on such an image and gotten a
different result,
I’d be happy to hear
about it.
16 bits allows you to pile more changes on top of each
other without accruing round-off errors. You can make some global edit
(e.g., Curves, Levels, Hue/Sat, etc.), then decide it’s not quite
right so make another edit, then another, then another, without worrying
about the fact that in 8-bit mode, each tweak can add up to half an lsb
worth of quantization noise in some areas. In most images it won’t
matter, but it frees you from worrying that you’re approaching the
point where you’re visibly degrading the image.
Of course, you can avoid some of this by only doing
changes through adjustment layers, but that’s a major workflow
restriction.
It’s sort of like JPEG compression. I
don’t hesitate to save my finished images as JPEGs, using quality
level 12, and have proved to my satisfaction that you can’t see any
quality difference between it and an uncompressed TIFF, no matter how hard
you look. But if I were repeatedly loading, editing and saving, I
wouldn’t use JPEG for the intermediate files.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Sun, 14 Nov 2004 16:12:02 EST
From: Dan Margulis
Subject: Re: Color bit depth
Paul DeRocco writes:
16 bits allows you to pile more changes on top of each
other without
accruing round-off errors. You can make some global
edit (e.g., Curves,
Levels, Hue/Sat, etc.), then decide it’s not
quite right so make another
edit, then another, then another, without worrying
about the fact that in
8-bit mode, each tweak can add up to half an lsb worth
of quantization noise
in some areas. In most images it won’t matter...
That phrase “most images” seems to suggest
that there must be some images where it *does* matter. Have you ever
encountered any of these images yourself? And if you thought you did, how
did you know that it mattered? Did you perchance run the same move, however
intense they may have been, on a copy of the image that had been converted
to 8-bit, so that you might have known whether it mattered or not? If so,
would you be able to share the image(s) with us? If not, can you point to
someone who has done similar testing?
Of course, you can avoid some of this by only doing
changes through
adjustment layers, but that’s a major workflow
restriction.
Have you tested this for yourself? Or have you seen
anyone else demonstrate it? If you have shown that using adjustment layers
preserves any significant detail that applying the changes without them
doesn’t, would you be willing to share the files? Or could you point
us to anyplace where we could see such files?
It’s sort of like JPEG compression. I
don’t hesitate to save my finished
images as JPEGs, using quality level 12, and have
proved to my satisfaction
that you can’t see any quality difference
between it and an uncompressed
TIFF, no matter how hard you look.
Perhaps no matter how hard *you* look, but most people
wouldn’t have to look all that hard to find the quality loss—a
simple Command-3 with the file open on the screen would demonstrate what I
would characterize as, er, a night-and-day difference.
Now in the case of the very extreme moves I made to
text 16-bit vs. 8-bit, I looked really, really hard—and I
couldn’t find any difference that would affect quality in any
significant way.
But if I were repeatedly loading, editing and saving,
I wouldn’t use JPEG
for the intermediate files.
Neither would I—but I wouldn’t expect
anyone to take my word for it without anything to back it up. Therefore, on
Page 368, Chapter 17 of Professional Photoshop Fourth Edition, I show an
image where use of JPEG compression definitely and incontestably harms a
subsequent correction. That is, whether the original JPEGging could be
detected in print or not, it restricted the things that could be done in
the future. In this particular case, the example was a luminosity blend
into a fleshtone. It showed serious artifacting and, next to one that had
been done without previously JPEGging the file, it would have been a
night-and-day difference.
Have you ever done a similar demonstration with a real
image that shows the same kind of restriction on future action if you
convert to 8-bit too early? If so, would you be willing to share it with
us? If not, would you be able to point us to someone who has?
Dan Margulis
___________________________________________________________________________
Date: Sun, 14 Nov 2004 16:11:11 EST
From: Dan Margulis
Subject: Re: 16 Bit images only into LAB color space
On November 14, 2004, at 12:22 p.m., Andrew Rodney
wrote,
The argument concerns Andrew’s revisionist claims
that he has never heard
anyone (like, for example, his own business partners)
say that correction in
16-bit creates a night-and-day difference, or that
anyone who doesn’t work
that way is a recreational as opposed to a professional
user.
No, I said I never said that and challenged you to
quote me on saying that.
**************************************
But on November 11, 2004, at 9:33 a.m. Andrew Rodney
wrote,
And no I've not seen any text that says
©¯16-bit is absolutely critical,
creates a night
and day difference, that anyone who doesn’t do
it is an amateur, etc., etc©~.
**************************************
Those familiar with Andrew are aware that older
statements tend to disappear from his memory as soon as they are disproven.
However, normally it takes a couple of years. Three days, two hours, 49
minutes must be some kind of new record.
Dan Margulis
___________________________________________________________________________
Date: Sun, 14 Nov 2004 14:59:15 -0700
From: Andrew Rodney
Subject: Re: 16 Bit images only into LAB color space
Once again I have to take up bandwidth to point out
Dan1s misunderstandings. First, I never said:
16-bit creates a night-and-day difference, or that
anyone who doesn’t work
that way is a recreational as opposed to a
professional user.
Second, I’ve never seen anyone suggest this
however I did find this URL after the discussion started which I posted:
http:
//www.photoshoptechniques.com/forum/archive/index.php/t-336.html
No where in that set of posts does anyone say 16-bit
is absolutely critical, or it creates a night and day difference. In that
post, I DID find what I think is the original comment on
3recreational users2 which I also referenced in that
post.
Enough of this liberal editing in an attempt to
suggest I1m a Kerry-like revisionist (which like1s Dan1s claim is bogus).
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Sun, 14 Nov 2004 16:54:10 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Dan Margulis [mailto:Dan Margulis]
That phrase “most images” seems to suggest
that there must be some images
where it *does* matter. Have you ever encountered any
of these images yourself?
And if you thought you did, how did you know that it
mattered? Did you
perchance run the same move, however intense they may
have been, on a copy of the
image that had been converted to 8-bit, so that you
might have known whether it
mattered or not? If so, would you be able to share the
image(s) with us? If not,
can you point to someone who has done similar testing?
When I say “some images” I merely mean
that most of the time I get satisfactory adjustments in one try, and would
undoubtedly get perfectly good results with eight bits. Only occasionally
do I go back and forth, trying this tweak and that tweak, before I’m
satisfied with the results. But the math behind quantization noise is
pretty simple: since it’s basically random on real-world images, you
get typically twice as much added quantization noise if you do four edits
as for one, or four times as much if you do sixteen edits.
Have you tested this for yourself? Or have you seen
anyone else demonstrate
it? If you have shown that using adjustment layers
preserves any significant
detail that applying the changes without them
doesn’t, would you be willing to
share the files? Or could you point us to anyplace
where we could
see such files?
The point of adjustment layers is that you can go back
and tweak the adjustments over and over without adding more quantization
noise, because each time the result is recalculated from the _original_
image. That’s a good thing, but since an adjustment layer is another
step, I generally find it less convenient.
Perhaps no matter how hard *you* look, but most people
wouldn’t have to look
all that hard to find the quality loss—a simple
Command-3 with the file open
on the screen would demonstrate what I would
characterize as, er, a
night-and-day difference.
We must be talking about two different things.
I’m talking about saving an image, after all edits are finished, as
JPEG quality 12. I’m not talking about low-quality JPEG. I can see
compression artifacts at quality 5, maybe quality 8, but never at quality
12. And I’m not talking about saving originals as JPEG, before
editing.
I’ve done A/B comparison on sharp curved edges
from real-world images, such as tree branches, blowing both the before and
after images up to 1000%. After saving and reloading, I can find pixels
that look slightly different, but without labels on the images I could
never tell you which was the before and which was the after.
But it seems odd to me that you would be so certain
that a single JPEG conversion would introduce visible degradation, but not
a long series of eight bit curve manipulations or other simple edits.
Have you ever done a similar demonstration with a real
image that shows the
same kind of restriction on future action if you
convert to 8-bit too early? If
so, would you be willing to share it with us? If not,
would you be able to
point us to someone who has?
Here’s an example. I took a 16-bit ISO 100 image
(from a raw file) that had blue sky in it, cropped a piece of the sky,
copied it as an eight bit image, and then did a whole series of identical
curve, hue/sat, levels and color balance manipulations to the two. The
successive edits were more or less inverses of each other, boosting the
contrast a bit with an S-curve one way, and then cutting it back with an
S-curve the other way, or dropping the saturation a little, and then
boosting it back up, so that I wound up with something that still looked
like blue sky. In the end, I converted the 16-bit image to 8-bit, and saved
both along with the original using quality 12 JPEG. Here they are:
http://www.pbase.com/pderocco/image/36348013 Original
http://www.pbase.com/pderocco/image/36348011 16-bit
edits
http://www.pbase.com/pderocco/image/36348012 8-bit
edits
I’ve also appended the section from the
Photoshop Edit Log, so you can see what the edits were. They’re
fairly strong, but nothing outlandish, compared to what I often do to
images.
In PBase, you won’t see any difference between
the two edits. But if you open them in PS and blow them up to 1000%, so
that you can see the individual pixels, you can easily see that the 8-bit
image has gotten a little noisier than the 16-bit image. Admittedly, this
is a lot of back-and-forth edits, and the differences aren’t
make-or-break. But the point is only that in 8-bit mode, you have to worry
about eventually degrading the image a little, whereas in 16-bit mode, you
can edit all day without worrying about quantization noise build-up.
—
Ciao,
Paul D. DeRocco
Paul
mailto:pderocco@ix.netcom.com
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 128, 192
point: 255, 255
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 192, 128
point: 255, 255
Hue/Saturation
Hue/Saturation
Without Colorize
Adjustment: hue/saturation
adjustment list
hue/saturation adjustment
Hue: -35
Saturation: 10
Lightness: 0
Hue/Saturation
Hue/Saturation
Without Colorize
Adjustment: hue/saturation
adjustment list
hue/saturation adjustment
Hue: 35
Saturation: -10
Lightness: 0
Levels
Levels
Adjustment: levels adjustment
list
levels adjustment
Channel: composite channel
Input: 40, 220
Levels
Levels
Adjustment: levels adjustment
list
levels adjustment
Channel: composite channel
Output: 40, 220
Color Balance
Color Balance
Shadow Levels: 0, 0, 0
Midtone Levels: 13, 0, -14
Highlight Levels: 0, 0, 0
Without Preserve Luminosity
Color Balance
Color Balance
Shadow Levels: 0, 0, 0
Midtone Levels: -13, 0, 14
Highlight Levels: 0, 0, 0
Without Preserve Luminosity
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 64, 80
point: 192, 176
point: 255, 255
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 80, 64
point: 176, 192
point: 255, 255
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 64, 32
point: 192, 224
point: 255, 255
Curves
Curves
Adjustment: curves adjustment
list
curves adjustment
Channel: composite channel
Curve: point list
point: 0, 0
point: 32, 64
point: 224, 192
point: 255, 255
Hue/Saturation
Hue/Saturation
Without Colorize
Adjustment: hue/saturation
adjustment list
hue/saturation adjustment
Hue: 10
Saturation: -40
Lightness: 0
Hue/Saturation
Hue/Saturation
Without Colorize
Adjustment: hue/saturation
adjustment list
hue/saturation adjustment
Hue: -10
Saturation: 40
Lightness: 0
_________________________________________________________________________
Date: Mon, 15 Nov 2004 13:02:58 -0500
From: Lee Clawson
Subject: Re: Color bit depth
on 11/14/04 4:12 PM, Dan Margulis at Dan Margulis
wrote:
Have you tested this for yourself? Or have you seen
anyone else demonstrate
it? If you have shown that using adjustment layers
preserves any significant
detail that applying the changes without them
doesn’t, would you be willing to
share the files? Or could you point us to anyplace
where we could see such
files?
Dan,
While outside your request the following might be of
interest.
___________________________________
The only source which I know (but I’m unable to
comprehend the science) is work done for motion picture film. At SIGGRAPH
2004 a software engineer from ILM Studio, Florian Kainz presented an update
to Open EXR (http://www.openexr.com/OpenEXRColorManagement.pdf). This is an
open source 16 bit file format developed and used by ILM for output to
motion picture film. They have plug-ins for Photoshop and a public site,
http://www.openexr.com.
Though I wouldn’t want to speak for him, he has
implied that 16 bit files for print or display gain little if anything.
Here’s 2 of the features of OpenEXR (more at the
site):
— High dynamic range:
Pixel data are stored as 16-bit or 32-bit
floating-point numbers. With 16 bits, the representable dynamic range is
significantly higher than the range of most image capture devices: 109 or
30 f-stops without loss of precision, and an additional 10 f-stops at the
low end with some loss of precision. Most 8-bit file formats have around 7
to 10 stops.
— Good color resolution:
with 16-bit floating-point numbers, color resolution
is 1024 steps per f-stop, as opposed to somewhere around 20 to 70 steps per
f-stop for most 8-bit file formats. Even after significant processing
(e.g., extensive color correction) images tend to show no noticeable color
banding.
Lee
___________________________________________________________________________
Date: Mon, 15 Nov 2004 13:41:39 -0500
From: “Iliah Borg”
Subject: Re: Color bit depth
On Mon, 15 Nov 2004 13:02:58 -0500 Lee Clawson
wrote:
Even after significant processing (e.g., extensive
color
correction) images tend to show no noticeable color
banding.
Exactly right. That is why we are using FP for HDR.
Best regards,
ib
___________________________________________________________________________
Date: Mon, 15 Nov 2004 11:14:06 -0700
From: Nathan Wade
Subject: Re: Color bit depth
Dan Margulis writes:
Have you tested this for yourself? Or have you seen
anyone else demonstrate
it? If you have shown that using adjustment layers
preserves any significant
detail that applying the changes without them
doesn’t, would you be willing to
share the files? Or could you point us to anyplace
where we could see such
files?
A script may be coming to a theater near you.
As Dan mentions later in that post, adjustment layers
can help prevent quantization errors that are caused by multiple adjustment
applications to an image.
I’m of the opinion that adjustment layers are
good for workflows where images will be re-purposed. I also think that
adjustment layers, like 16-bit per channel support, are just one of many
features that Adobe has built into Photoshop to compensate for 1.) less
than ideal workflows 2.) user incompetence 3.) the current, dubious nature
of file exchange and editing 4.) the current, and less-than-ideal state of
color management.
I like to think of adjustment layers as an extension
of the ICC profile (embedded in an image). Adjustment layers can be turned
on and off, or changed at will (just as an image can change its assigned
profile, and embed or discard it.) When combined with ICC color
management and perhaps 16-bit per channel data, adjustment layers can help
compensate for minor output profile imperfections with minimal loss or
damage to the original image. It’s all about the original.
Additionally, adjustment layers are scriptable
objects, which has tasty implications for master files, high volume
workflows, and re-purposing.
Nathan Wade
___________________________________________________________________________
Date: Mon, 15 Nov 2004 14:52:15 -0800
From: Rick Gordon
Subject: Adjustment Layers & Quantization (WAS Re:
Color bit depth)
Lee Varis, in this forum in 2002, stated that
adjustment layers operate sequentially on the resulting output of the
layers beneath them — therefore, not preventing quantization errors.
At 8:18 PM -0700 9/6/02, Lee Varis wrote:
It is a common misconception that using
“adjustment layers” is somehow less destructive than simply
doing the same edits one after the other. In order for multiple adjustment
layers to work at all the transforms applied by each successive layer HAVE
to be calculated sequentially - one after the other. There is no difference
once the file is flattened or sent through a printing rip to doing the same
corrections directly on the file one after the other.
The advantage of using adjustment layers is really the
ability to revise a correction later without re-applying (you edit the
layer and its like applying a new correction to the original file instead
of making an additional successive adjustment) - Thats it!
Stacking up adjustment layers is just as destructive
as applying multiple adjustments and should be avoided to the same degree.
If, however, you gain some advantage using other features of layer work
(layer masks, blending options, ect..) than it may be desirable to use
layers rather than successive edits.
Can anyone authoritatively confirm or deny this? Or
has it changed with CS?
Rick Gordon
___________________________________________________________________________
Date: Mon, 15 Nov 2004 16:30:04 -0700
From: Andrew Rodney
Subject: Re: Adjustment Layers & Quantization (WAS
Re: Color bit depth)
on 11/15/04 3:52 PM, Rick Gordon wrote:
Lee Varis, in this forum in 2002, stated that
adjustment layers operate
sequentially on the resulting output of the layers
beneath them — therefore,
not preventing quantization errors.
There1s no damage UNTIL you flatten then the data loss
would be the same as if you applied the corrections once and moved on. At
least that1s my understanding from Chris Cox at Adobe. So Adjustment layers
are good for changing your mind and postponing the eventual data loss until
you1re done editing. Or course you now have 16 bit adjustment layer support
but let1s not open that can of worms again...
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Mon, 15 Nov 2004 17:41:49 -0600
From: Bob Smith
Subject: Re: Adjustment Layers & Quantization (WAS
Re: Color bit depth)
On Nov 15, 2004, at 4:52 PM, Rick Gordon wrote:
Lee Varis, in this forum in 2002, stated that
adjustment layers
operate sequentially on the resulting output of the
layers beneath
them — therefore, not preventing quantization
errors.
I think Lee is probably correct however the savings
that adjustment layers provides is valid for those situations where an
individual adjustment layer is edited numerous times as opposed to making
numerous sequential edits or piling adjustment layers on top of one
another. My ideas about how an image should look may change as I get
deeper into the editing process; or after I’ve seen a proof output at
proper size. I might lighten the image initially and then later
decide that darker and warmer was better. With an adjustment layer
that’s just one correction even though I may have edited (changed my
mind about) the settings in the adjustment layer numerous times before
final output.
Bob Smith
Accurate Image Bob Smith Photographer
Waco Texas USA
http://www.accurateimage.org
___________________________________________________________________________
Date: Mon, 15 Nov 2004 15:53:09 -0800
From: Paul DeRocco
Subject: RE: Adjustment Layers & Quantization (WAS
Re: Color bit depth)
From: Rick Gordon
Lee Varis, in this forum in 2002, stated that
adjustment layers
operate sequentially on the resulting output of the
layers
beneath them — therefore, not preventing
quantization errors.
Well, sure, if you stack up 8-bit adjustment layers,
you get the accumulated quantization error from those layers. The point,
though, is that you can twiddle the adjustments in those layers back and
forth all day long with no increase in quantization error. If you do that
with raw adjustments on the image layer itself, the errors gradually add
up.
—
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 15 Nov 2004 16:20:16 -0800
From: Rick Gordon
Subject: RE: Adjustment Layers & Quantization
Yes, I concur with that. I use them all the time
because that is true. I guess my question was: Does the rounding error from
each layer get passed into the source data for layer, essentially
travelling up the vertical chain as if each operation was done in turn?
Rick Gordon
___________________________________________________
RICK GORDON
EMERALD VALLEY GRAPHICS AND CONSULTING
___________________________________________________
WWW: http://www.shelterpub.com
___________________________________________________________________________
Date: Mon, 15 Nov 2004 18:52:34 -0600
From: “Mike Davis”
Subject: RE: Bit Depth question
I’m not suggesting answers here, just a question
from this novice. Aren’t we comparing apples to oranges when we
talk about noise and image degradation here? File compression
algorithms have nothing to do with bit depth, which is the level of color
differentiation. The number of file pixels is the same in both 8 bit
and 16 bit versions of a given image, we’re just increasing or
decreasing the number of colors (grayscale levels) available for use by
each pixel and hence increasing or decreasing file size. A 16 bit
image has no more resolution detail than an 8-bit image of the same pixel
dimensions, (as long as you don’t JPEG it to death).
So we’re talking about potentially
“damaging” or losing intermediate grayscale levels when we
fragment the histogram by editing? Actually, we’re creating
gaps by stretching the tonal values that were absent from the captured
image. We’re not losing color information, just shifting values
to correct for off-color or add contrast. The only difference I can
see is that the 16 bit histogram is “denser” and hence shows
less alteration. But what are we losing? Both 8 bit and 16 bit
images will be printed on a printer with only 4 to 7 colors available to
each pixel, dithered to re-create each pixel color value. Even with
7-color printers, 7 colors cubed is less than 8 bits cubed.
Wouldn’t it take a 16-color printer to do justice to a 16-bit
image? And if such were available, could we really see the difference
with the limited human capacity to differentiate between adjacent colors at
that level, even at 1000% magnification? Surely not at 100% where
individual pixels are not visible to the naked eye on a 300 ppi printout or
onscreen at 72 ppi.
Mike Davis
mldavis2 AT sbcglobal DOT net
___________________________________________________________________________
Date: Mon, 15 Nov 2004 19:14:40 -0800
From: Paul DeRocco
Subject: RE: Adjustment Layers & Quantization
From: Rick Gordon
Yes, I concur with that. I use them all the time
because that is
true. I guess my question was: Does the rounding error
from each
layer get passed into the source data for layer,
essentially
travelling up the vertical chain as if each operation
was done in turn?
I don’t know. I expect that in PS7 it worked
that way, but I suppose it’s possible that in PS CS, all the
intermediate calculations are done in 16 bits, only truncating back to 8
bits at the end. One could probably construct a test to find out, but that
I leave, as they say, as an exercise for the reader.
—
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 15 Nov 2004 17:38:15 -0800
From: Rick Gordon
Subject: Re: Color bit depth
[Not trying to beat a dead horse...]
My question regarding the possibility of a 16-bit
advantage in Lab is primarily concerning the granularity of near-neutral
values. When you increment the center point of a Lab curve from 0 to 1 or
-1, etc., there is a distinct, and easily perceptible shift in the source
neutral tones — to the point where I sometimes would want to split
the level of change. Could THIS be a target zone for quantifying a 16-bit
vs. 8-bit difference with a Lab move?
Rick Gordon
___________________________________________________
RICK GORDON
EMERALD VALLEY GRAPHICS AND CONSULTING
___________________________________________________
WWW: http://www.shelterpub.com
___________________________________________________________________________
Date: Mon, 15 Nov 2004 21:13:07 -0700
From: Ron Kelly
Subject: Re: RE: Bit Depth question
On 15-Nov-04, at 5:52 PM, Mike Davis wrote:
So we’re talking about
potentially “damaging” or losing
intermediate
grayscale levels when we fragment the histogram by
editing?
Mike:
This sounds like an apt description of this question.
Good on ya.
Ron Kelly
___________________________________________________________________________
Date: Tue, 16 Nov 2004 11:43:52 -0000
From: Stephen Marsh
Subject: Re: Color bit depth
Rick Gordon writes:
My question regarding the possibility of a 16-bit
advantage in Lab
is primarily concerning the granularity of
near-neutral values. When
you increment the center point of a Lab curve from 0
to 1 or -1,
etc., there is a distinct, and easily perceptible
shift in the source
neutral tones — to the point where I sometimes
would want to split
the level of change.
As things stand, you would need to separate this move
from the other edits. If using a direct adjustment, then fade xx%. If using
an adjustment layer, then set the layer to xx% opacity. Crude, but the best
we have.
I have mentioned this before and been confused by the
answers. I appreciate Dan’s POV that working in fine LAB increments
is hard for image editing as in it is not like dialing in colours in a
colour picker and that LAB is best for the big moves, often not the small
ones (where CMYK offers more control than RGB, just as modest gamut RGB
offers more control than LAB). I thought that there were two terms...ICC
LAB where the AB values were expressed as integers, while CIE LAB was
expressed as a A1.3 etc.
It is not the naming convention as such, just that
some software has the ability to dial in colours in finer icrements than a
value of 1, which is where you are coming from.
Perhaps you should look into RGB or CMYK edits rather
than LAB for such moves, perhaps using color blend mode?
Could THIS be a target zone for quantifying a 16-
bit vs. 8-bit difference with a Lab move?
As previously discussed, the maths indicate the
converting an 8bpc image to LAB will produce a unique loss of levels and
quantization.
The same is true for high bit data, but as there are
more bits the the high bit advocates have less issues with the move - but
most of them still wish that LAB would go away and that users would only
deal with RGB and perhaps CMYK.
Performing the move with high bit data, even if the
original is not high bit has minor advantage - but since the high bit data
is not true one still has loss of unique levels, just less quantization
(not that one can see this anyway, pre or post extreme edits).
The math/theory is sound. The problem starts in the
real world when one applies the theory to production workflows, as it can
be harder to justify “intangibles”.
Stephen Marsh.
___________________________________________________________________________
Date: Tue, 16 Nov 2004 18:42:52 +0000
From: Martin Orpen
Subject: Re: Color bit depth
on 15 Nov 2004 13:02:58, Lee Clawson wrote:
Pixel data are stored as 16-bit or 32-bit
floating-point numbers. With 16
bits, the representable dynamic range is significantly
higher than the range
of most image capture devices: 109 or 30 f-stops
without loss of precision,
and an additional 10 f-stops at the low end with some
loss of precision.
Most 8-bit file formats have around 7 to 10 stops.
If you take a closer look at the information on
openexr you’ll note that the data is described as
“linear”.
Machines use linear data, humans prefer non-linear -
and if you’d like to know why, take a look at the two images in this
story on our site:
<http:
//prometheus.idea-digital.com/phpbb2/viewtopic.php?t=299>
Big transformations like the move from linear to
non-linear benefit from using numbers with lots of decimal places.
You can choose to do it in Photoshop in 16-bit mode
yourself if you’re bored - but it’s much easier to have
dedicated software/hardware do the job for you and then work on the results
in Photoshop in plain old 8 bit.
As I mention in the story, I’ve wasted more time
than most people in trying to develop *killer* 16-bit workflows to scan
negatives and transparencies. My conclusion is that it’s a waste of
time.
Applying the initial edits during the scanning, and
working on the 8-bit (non-linear) results produces a better image, faster.
A good retoucher can produce better looking images
than any machine - but I really don’t see the point of extending that
argument back to the interpretation of raw data values.
I’m fully prepared to admit that a computer can
work out the answer to:
R’ = 1.099 R^0.45 - 0.099
A lot faster and a lot more times without complaining
than I’d ever care to
:-)
Why spend time on additional editing that can only be
*perceived* by a histogram?
—
Martin Orpen
Idea Digital Imaging Ltd — The Image Specialists
http://www.idea-digital.com
___________________________________________________________________________
Date: Tue, 16 Nov 2004 13:26:45 -0500
From: Dan Margulis
Subject: RE: Adjustment Layers & Quantization
Rick Gordon writes,
Yes, I concur with that. I use them all the time
because that is true. I guess my question was: Does the rounding error from
each layer get passed into the source data for layer, essentially
travelling up the vertical chain as if each operation was done in turn?
If you’re asking, if you have three adjustment
layers and then flatten the image, does it give you exactly and precisely
the same file as if you had applied the three adjustments consecutively
without using layers, yes, it does, contrary to at least one article.
Dan Margulis
___________________________________________________________________________
Date: Tue, 16 Nov 2004 20:31:00 -0500
From: Dan Margulis
Subject: Re: Color bit depth
Rick Gordon writes,
My question regarding the possibility of a 16-bit
advantage in Lab is primarily concerning the granularity of near-neutral
values. When you increment the center point of a Lab curve from 0 to 1 or
-1, etc., there is a distinct, and easily perceptible shift in the source
neutral tones — to the point where I sometimes would want to split
the level of change. Could THIS be a target zone for quantifying a 16-bit
vs. 8-bit difference with a Lab move.
As a practical matter, no. First of all, hitting any
specific A or B value with a curve isn’t easy in LAB no matter how
many bits you have. The curves are too powerful—they aren’t
meant for finetuning things. Unless you’re in a horrendous rush, you
should use LAB to come close to where you want to be and assume that minor
touchup is going to be necessary in RGB or CMYK thereafter.
I think what you’re asking for is the ability to
access a value of .5A. It’s theoretically possible in 16-bit but not
in 8-bit. However, there’s no practical way of knowing whether
you’ve hit it because Photoshop only reads out in 1-point increments.
So, you’d be stuck with putting a layer of 1A over a layer of 0A and
changing the opacity. If you had such a layered file, you could convert it
directly into RGB or CMYK without flattening, and, provided you used
Convert to Profile, which uses 20 bits in its computations last I knew,
you’d get the same results whether the layered file was in 8-bit or
16-bit.
The short answer is that it’s a whole lot easier
to do these very fine moves ABL—meaning, anywhere but LAB.
Dan Margulis
___________________________________________________________________________
Date: Tue, 16 Nov 2004 20:16:15 -0500
From: Dan Margulis
Subject: RE: Color bit depth
Paul DeRocco writes,
But it seems odd to me that you would be so certain
that a single JPEG conversion would introduce visible degradation, but not
a long series of eight bit curve manipulations or other simple edits.
As explained in my last post, nobody contends that
anybody but an expert can determine whether a printed piece has been
JPEGged first, unless it’s printed at a much larger size than its
resolution justifies. However, JPEGging puts in certain artifacts that may
become harmful if the file is used for other purposes. I am certain that
this effect exists because I can see it, I can perform tests that make it
evident, and because I can construct a demonstration file that shows clear
damage to a real image that is used in a plausible real-world scenario.
A long series of edits may or may not cause such an
effect. If you are saying, a long series of edits done in 8-bit can cause
the effect where doing them in 16-bit wouldn’t, I am uncertain of
this because I can’t see it, none of my tests indicate that
it’s there, and I can’t prepare any demonstration that it
exists using any real-world color photograph and any plausible real-world
scenario. Furthermore, none of the proponents of the workflow can either,
and to judge by the lame series of files you offered links to, neither can
you. Yet I am given to understand that there is a "night and
day"", "totally obvious” gain in quality, and that
anyone who doesn’t do things that way is a "recreational"
user of Photoshop as opposed to a professional.
I do not see what is so hard to understand about
"real image" and "plausible real-world scenario"?. What
you have posted is neither. It demonstrates conclusively that if you
correct in 16-bit you will get a result that isn’t pixel-for-pixel
identical with an 8-bit correction, but that’s about it.
You posted a flat color containing no detail
whatsoever, except for artifacting that you described as noise. Images of
flat colors containing zero valid detail are not typical of real-world
work. Ordinarily the idea is to enhance the detail in an image, not
suppress it. Ordinarily the idea is to make the image look better than the
original, not worse. Your example stands all three of these concepts on
their head. Correcting in 8-bit brought out the detail in the image
slightly better than correcting in 16-bit did, as often is the case.
Therefore, the version done in 8-bit is a technically superior correction
to the one done in 16-bit, which you would see if there were anything else
in the image except for this flat area. It is only your insistence that,
unlike anything found in the real world, all the detail in the image is bad
that causes you to think that there’s a problem.
Similarly, if you are saying that exaggerating the
detail is bad, then both the 16-bit and the 8-bit “corrections”
are worse than the original, again, contrary to any real-world practice.
Also, gross lightening followed by gross darkening,
then gross desaturation followed by gross saturation, followed by
conflicting color balance commands is not a real-world production scenario.
Anybody following such a workflow has more than enough problems to keep
themselves busy without worrying about bit depth.
So again, I ask. Do you have any REAL COLOR PHOTOGRAPH
on which you have applied REAL CORRECTIONS that appears to work better when
the corrections are applied in 16-bit rather than 8-bit mode? If so, would
you mind sharing it with us? If not, and it’s just your hunch that
somehow, someday, somewhere it might possibly produce a better result, fine
if you have the disk space to support the habit. However, I question that
it justifies suggesting to other people that they double the size of their
workfiles.
Dan Margulis
___________________________________________________________________________
Date: Tue, 16 Nov 2004 21:51:41 -0500
From: “jc castronovo”
Subject: Re: Color bit depth
For once I have to agree with you Dan. This thread has
prompted me to do a series of tests, and however bad I can make the
histograms look after attempting to do serious damage in 8 bits vs. doing
the same corrections in 16 bits, the results look the same. They’re
the same on the monitor, they’re the same on a proof and
they’re the same even on Ektachrome film output.
Now I know that there are cases where it matters
because I’ve seen it before and I fixed the problem by using 16 bit
file corrections, but now that I want to make it happen, I can’t. My
conclusion is that for most work and almost all users, 8 bit workflow is
all that we need unless we run into a problem with that unique image.
john castronovo
___________________________________________________________________________
Date: Tue, 16 Nov 2004 19:30:39 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Dan Margulis
As explained in my last post, nobody contends that
anybody but an
expert can determine whether a printed piece has been
JPEGged
first, unless it’s printed at a much larger size
than its
resolution justifies. However, JPEGging puts in
certain artifacts
that may become harmful if the file is used for other
purposes. I
am certain that this effect exists because I can see
it, I can
perform tests that make it evident, and because I can
construct a
demonstration file that shows clear damage to a real
image that
is used in a plausible real-world scenario.
I just created a couple of crops out of some images,
and did a simple test. I saved the image as 8-bit TIFF, saved it again as
JPEG quality 12, reloaded the JPEG, and saved it also as 8-bit TIFF. When I
load the two TIFFs into PS or ThumbsPlus and blow it up to, say, 800%, so
that I can see the individual pixels, and toggle back and forth between
them, I can’t see any difference. If I apply a really strong contrast
boost, strong enough to ruin the picture, then I can see that the pixels
aren’t quite identical. But since I only do this on _finished_
images, not ones that are going to undergo major edits, cranking the
contrast in order to see a tiny difference isn’t a fair test.
The two pairs of files are here:
http://pderocco.dynip.com/junk/JPEG/
Can you tell me which one of each pair looks better?
A long series of edits may or may not cause such an
effect. If
you are saying, a long series of edits done in 8-bit
can cause
the effect where doing them in 16-bit wouldn’t,
I am uncertain of
this because I can’t see it, none of my tests
indicate that it’s
there, and I can’t prepare any demonstration
that it exists using
any real-world color photograph and any plausible
real-world
scenario. Furthermore, none of the proponents of the
workflow can
either, and to judge by the lame series of files you
offered
links to, neither can you. Yet I am given to
understand that
there is a “night and day”?,
“totally obvious”? gain in quality,
and that anyone who doesn’t do things that way
is a
“recreational”? user of Photoshop as
opposed to a professional.
I never said that. I quite explicitly said that it was
a small difference, and then only after long sequences of edits. My point
was only that in 16-bit mode, you don’t have to worry about
encroaching on that point, if you’re doing extensive experimenting on
an image.
I do not see what is so hard to understand about
“real image”?
and “plausible real-world scenario”?. What
you have posted is
neither. It demonstrates conclusively that if you
correct in
16-bit you will get a result that isn’t
pixel-for-pixel identical
with an 8-bit correction, but that’s about it.
You posted a flat color containing no detail
whatsoever, except
for artifacting that you described as noise. Images of
flat
colors containing zero valid detail are not typical of
real-world
work. Ordinarily the idea is to enhance the detail in
an image,
not suppress it. Ordinarily the idea is to make the
image look
better than the original, not worse. Your example
stands all
three of these concepts on their head. Correcting in
8-bit
brought out the detail in the image slightly better
than
correcting in 16-bit did, as often is the case.
Therefore, the
version done in 8-bit is a technically superior
correction to the
one done in 16-bit, which you would see if there were
anything
else in the image except for this flat area. It is
only your
insistence that, unlike anything found in the real
world, all the
detail in the image is bad that causes you to think
that there’s
a problem.
But it _is_ a real-world image. It’s a piece of
sky. And that wasn’t “detail” that was being emphasized,
it was quantization error that was being introduced. Quantization error
looks like noise, and the only place you notice noise in real-world images
is in a noiseless color, the most common example of which is blue sky.
You’ll never notice noise, whether from 8-bit roundoff errors, or for
that matter from using ISO 200 instead of ISO 100, in a picture of grass,
or trees, or ocean waves, or someone’s hair, because the noise is
drowned out by the image detail. But I gave up on my DiMage 7 and bought a
10D precisely because large prints of blue sky looked like blurry blue
confetti.
Also, gross lightening followed by gross darkening,
then gross
desaturation followed by gross saturation, followed by
conflicting color balance commands is not a real-world
production
scenario. Anybody following such a workflow has more
than enough
problems to keep themselves busy without worrying
about bit depth.
So again, I ask. Do you have any REAL IMAGE on which
you have
applied REAL CORRECTIONS that appears to work better
when the
corrections are applied in 16-bit rather than 8-bit
mode? If so,
would you mind sharing it with us? If not, and
it’s just your
hunch that somehow, someday, somewhere it might
possibly produce
a better result, fine if you have the disk space to
support the
habit. However, I question that it justifies
suggesting to other
people that they double the size of their workfiles.
You’re asking the impossible, because when I do
what you call “REAL CORRECTIONS”, I’m not doing tests. In
order to do a test, I have to do _precisely_ the same edits on two files.
In order to satisfy your request, I would have to do this laborious process
on every image I edit, until I just happen to encounter one that for
whatever reason prompts me to do an unusually large number of edits.
Remember, I said quite clearly that for most images, I quickly home in on
what I want, without a lot of trial and error, and that in that vast
majority of cases, eight bit editing would be fine.
As to disk size, that’s an irrelevancy. One may
occasionally save an image part way through an edit, and come back to it
later, but one doesn’t normally save a file, close it, then reopen
and reload it, between each of a large number of edits.
—
___________________________________________________________________________
Date: Wed, 17 Nov 2004 15:05:33 -0500
From: Dan Margulis
Subject: Re: Color bit depth
John Castronovo writes,
For once I have to agree with you Dan. This thread has
prompted me to do a
series of tests, and however bad I can make the
histograms look after
attempting to do serious damage in 8 bits vs. doing
the same corrections in
16 bits, the results look the same. They’re the
same on the monitor, they’re
the same on a proof and they’re the same even on
Ektachrome film output.
Well, if you’ve tested it and that’s what
you’ve found, then you have to agree, just as I would have to agree
if somebody actually came up with a file that showed one of those
night-and-day, you’re-an-amateur-if-you-don’t-do-this kind of
differences.
It continues to amaze that all the people who actually
test the idea report that there isn’t any there there, and yet the
people who haven’t actually tested it (or perhaps have, and
don’t care for what the results showed) continue to hype it.
Dan Margulis
___________________________________________________________________________
Date: Wed, 17 Nov 2004 16:44:39 -0500
From: Dan Margulis
Subject: RE: Color bit depth
Paul DeRocco writes,
But it _is_ a real-world image. It’s a piece of
sky.
When you can *sell* a piece of sky that has no cloud,
no detail, nothing but a flat tint, then you will have a real-world image.
And that wasn’t “detail” that was
being emphasized, it was quantization error that was being introduced.
Unless you’ve taken a couple of statistical
analysis courses, you should refrain from using terms that you don’t
understand. There is no error where there isn’t sufficient precision
in the reference sample to demonstrate one. Given the magnitude of the
changes in your file, the 8-bit sample is probably mathematically more
correct than the 16-bit one, which looks like it’s assigning a
mathematical significance to the 11th and 12th bits, which are basically
random numbers.
In all image-processing operations, as in all
averaging operations, the more data, the smoother the result—but not
necessarily the better the result. Working with 16-bit creates a smoother
result than working with 8-bit, just as working with higher resolution as
opposed to lower resolution will. In the case of resolution—as
I’ve shown, with real-world images, in my books—the effect is
large enough for a casual observer to prefer the printed version of an
image with the correct resolution versus one with resolution that’s
too high or too low. In the case of 8-bit vs. 16-bit, the difference is so
trivially small that only an expert can even see it except at high
magnification.
Quantization error looks like noise,
There is *no* noise in your 8-bit sample that
isn’t in your 16-bit sample. The two samples treat transitions
differently, but the difference is so slight, in spite of the enormously
non-real-world moves you made, that you have to go to high magnification to
see it. I reiterate that had you included the rest of the image (IOW,
anything with real detail) you would probably have preferred the 8-bit to
the 16-bit file.
You’re asking the impossible, because when I do
what you call “REAL CORRECTIONS”, I’m not doing tests. In
order to do a test, I have to do _precisely_ the same edits on two files.
In order to satisfy your request, I would have to do this laborious process
on every image I edit, until I just happen to encounter one that for
whatever reason prompts me to do an unusually large number of edits.
It only seems impossible to those people who advocate
the 16-bit workflow. At least half a dozen people on this list have tried
side-by-side tests, using the sorts of images that the advocates claim work
better in 16-bit, and the sorts of corrections that they say will create a
problem if done in 8-bit. Me, whenever I find any method that seems to work
better than the currently popular one, I make sure I’ve got enough
backup to make the case if I have to, and I know other authors do the same.
But the bottom line is, since you haven’t run
such tests, you have never personally seen even a single image that you
know for a fact works better in 16-bit than 8-bit. Fair enough. If you ever
do find one, please let us know.
Dan Margulis
___________________________________________________________________________
Date: Wed, 17 Nov 2004 18:16:14 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
Look, I’ve had enough of this. First you made
the radical claim that 16 bits _never_ makes any difference in the quality
of an image. I responded by making the rather innocuous claim that it is
occasionally possible to tell the difference, but only in extreme cases.
You seemed to find this outrageous, and challenged me to prove it. So I
constructed what I regard as a real-world test, cropping a piece of blue
sky out of an image, doing some strong but not outlandish manipulations to
it in both 8-bit and 16-bit mode. This you dismissed as not a real-world
image because the entire image was sky. Well, duh, the sky was cropped
_from_ a real image, which I did because the full images, saved at JPEG
quality 12, would have been six megabytes, and I clearly said that the
effect would only be visible in things like sky.
To reassure you that I’m not one of those
mindless techno-geeks who blindly believes that bit-for-bit accuracy
matters, I mentioned that once I’ve produced a finished image, I
don’t mind saving it as quality 12 JPEG, because my extensive,
careful tests have been unable to detect a difference between it and an
uncompressed TIFF. In response, you suddenly became the bit-for-bit
techno-geek, making the radical claim that you could _always_ tell the
difference when an image had been saved as JPEG. So again, I constructed a
demonstration, choosing a couple of small crops from real-world images,
containing sharp contrasty edges, since that’s where you most easily
see JPEG artifacts, and put two TIFF versions of them on my server (since
PBase won’t accept TIFFs), one of each pair having been saved as JPEG
and reloaded. You haven’t even acknowledged their existence, let
alone told me which was which.
Now, you accuse me of being ignorant of statistical
analysis, and of using terms I don’t understand. It happens that
I’m not a professional photographer, I’m a professional digital
signal processing engineer specializing in communication systems. Questions
of noise, statistics, and so on, are what I deal in every day, and I doubt
there are more than a couple people on this list who know more about such
mathematics, more likely none.
I began with a very moderate claim. I was careful not
to overstate my case: I did not say that 8-bit sucks, or that 8-bit is
unprofessional, or that 8-bit is likely to cause problems for people, and
they should always use 16-bit. Yet judging by the testy tone of your
responses, this obviously really upset you. When you challenged me, I did
what you asked and produced some samples that showed what I meant, which
you’ve tried to dismiss as not “real-world” enough for
you. Fine—we live in different worlds. But on your side, you claim to
have made tests and have found the opposite, and mention other people who
have made similar tests, yet I don’t actually see the tests. And
frankly I don’t care, because nothing you could show me would alter
the fact that I can see the difference in my own test.
I then made another moderate claim about light JPEG
compression, you again bristled that anyone would dare make a claim you
disagree with, so again I produced some example images which you’ve
ignored. And again, you claim to have counterevidence, but all I see is
your claim, not your counterevidence.
Your tone throughout this debate has been mildly rude.
I consider the debate over, and I don’t really care who
“won”. Feel free to edit everything in 8-bit mode and save all
your images as uncompressed TIFF. I’ve spent way too much time on
this argument already.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Thu, 18 Nov 2004 07:08:33 -0700
From: Andrew Rodney
Subject: Re: Color bit depth
on 11/17/04 7:16 PM, Paul D. DeRocco wrote:
When you challenged me, I did what you asked and
produced
some samples that showed what I meant, which
you’ve tried to dismiss as not
“real-world” enough for you.
Welcome to the club Paul. You basically got the same
response as Bruce Lindbloom and others.
Personally I see nothing wrong with your cropped
example of a sky. Last time I looked, a photo of the sky was 3real world2
...
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Thu, 18 Nov 2004 11:16:53 -0500
From: “Hirsch, Steve”
Subject: RE: Color bit depth
Hi,
I’ve found this whole discussion very
interesting actually; sorry to see that top-notch, intelligent
professionals such as yourselves can’t engage in a heated discussion
without it getting personal. I’m going to throw my two cents in here
just for fun, hoping I can contribute slightly to the dialog and confirm
Paul’s findings.
We publish more than 20 major magazines including
Elle, Woman’s Day, Car and Driver, Popular Photography and many
others. We did an extensive test on JPEG compression levels and bit depth
modes with actual image content from our magazines, printing Kodak
Approvals of a sample suite of images saved separately in Photoshop JPEG
compression levels from 2 through 12. We showed these samples to a room
full of industry professionals and color experts and not one of these
experienced pair of eyes could tell the difference between a compression
level of 6 and a level of 10 or 12. Only images saved in compression level
4 or lower showed artifacts on the proof. We settled on a standard file
format of 8-bit JPEG level 8 for all of our image workflows since no one
could tell the difference in either proof or print between the uncompressed
TIFFS and the JPEGs.
The bit-for-bit techno-geeks can’t carry the day
here; I used to be one myself and have learned that prepress, printing and
color management is still an art, not purely a science, and always has
been.
Steve
Steven Hirsch, Systems Manager
Hachette Filipacchi Media, U. S.
212-767-6536
___________________________________________________________________________
Date: Thu, 18 Nov 2004 11:26:08 -0500
From: Lee Clawson
Subject: Re: Re: Color bit depth
on 11/16/04 1:42 PM, Martin Orpen wrote:
If you take a closer look at the information on
openexr you’ll note that the
data is described as “linear”.
Machines use linear data, humans prefer non-linear -
and if you’d like to
know why, take a look at the two images in this story
on our site:
<http:
//prometheus.idea-digital.com/phpbb2/viewtopic.php?t=299
Martin,
Re: OpenEXR. The “linear” data, the
specific shape of data, never bothered me.
Other fields (film) also have this
conversation/debate. Like Dan, yourself and others I don’t find any
noticeable differences in my work. With regard to Open EXR, I do like the
concept of developing extended dynamic range. Today, I agree there’s
nothing there, especially for print, but I do keep looking at it.
Lee
___________________________________________________________________________
Date: Thu, 18 Nov 2004 09:58:05 -0700
From: Andrew Rodney
Subject: Re: Color bit depth
on 11/18/04 9:16 AM, Hirsch, Steve wrote:
We showed these samples to a room full of industry
professionals and color
experts and not one of these experienced pair of eyes
could tell the
difference between a compression level of 6 and a
level of 10 or 12. Only
images saved in compression level 4 or lower showed
artifacts on the proof.
For halftone output to that linescreen, fine. But
there are plenty of OTHER output devices where this panel *may* have seen
otherwise.
If you’re concerned with one output type and
quality, that’s one thing. If you don’t know or you’re
printing to a more demanding output device (a true contone device), a lot
more testing is in order.
Andrew Rodney
http://digitaldog.net
___________________________________________________________________________
Date: Thu, 18 Nov 2004 14:13:10 -0500
From: Jim Rich
Subject: Re: Color bit depth
On 11/18/04 11:58 AM, “Andrew Rodney”
wrote:
If you’re concerned with one output type and
quality, that’s one thing. If
you don’t know or you’re printing to a
more demanding output device (a true
contone device), a lot more testing is in order.
Andrew,
Testing is definitely in order, but...
I it is my understanding that ALL modern day output
devices use a core technology of 8 bits for printing on media.
Also, when we were verifying the 8 v16 bit argument
a few years ago, it was found than when each type of image was
output to inkjet, Lamda, and a film recorder the visual results were
the same. That is there was not degradation from multiple edits on either 8
or 16 bit images.
While there might be one oddball device somewhere that
can somehow take more than 8 bits and will provide better results than 8
bits is a hard argument to buy into considering there are so many 8
bit output devices.
Jim Rich
___________________________________________________________________________
Date: Thu, 18 Nov 2004 13:50:04 -0700
From: Andrew Rodney
Subject: Re: Color bit depth
on 11/18/04 12:13 PM, Jim Rich wrote:
I it is my understanding that ALL modern day output
devices use a core
technology of 8 bits for printing on media.
Nope, my Epson 2200 driver with the ImagePrint RIP
will use the 16 bit data. Talk to John P at ColorByte about this. I think
either the Lightjet or Lambda (can1t recall which) will also access the
additional data and use it.
Also, when we were verifying the 8 v16 bit argument
a few years ago, it was
found than when each type of image was output to
inkjet, Lamda, and a film
recorder the visual results were the same. That
is there was not
degradation from multiple edits on either 8 or 16 bit
images.
Obviously the edits made would play a role. We don1t
know what edits a file will undergo nor do we know what output devices will
be able to use data wise in the coming years.
While there might be one oddball device somewhere that
can somehow take
more than 8 bits and will provide better results than
8 bits is a hard
argument to buy into considering there are so many 8
bit output devices.
Unless you1re using one of those odd ball devices or
more come onto the scene. If your goal is to have the data in such cases,
the idea of keeping the data (or a copy of the data) in high bit makes
sense.
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Thu, 18 Nov 2004 17:07:28 -0500
From: Jim Rich
Subject: Re: Color bit depth
On 11/18/04 3:50 PM, “Andrew Rodney” wrote:
Unless you1re using one of those odd ball devices or
more come onto the
scene. If your goal is to have the data in such cases,
the idea of keeping
the data (or a copy of the data) in high bit makes
sense.
Like I said (and have researched) there are only 8 bit
print engines in most All of the printers in todays market.
How if you have a rip that massages 16 bit data
differently than 8 bit data. Cool. But the printer is still getting 8 bits
of data as does all Epsons, light jets and lamdas last time I looked at
their specs. And while your example has some validity is it more of an
exception than the rule. As for different print engines coming into
the printing scene? Is there a large market demand for high bit printing?
Probably not but...
Positioning for the future is one thing and it is not
so bad. I agree that it does not hurt if you can afford it to save
raw high bit images and work with them in 8 or 16 bit. But for
printing most all images today, 8 bit images work just as good as 16 bit
images. The evidence is too overwhelming. Like it or not it is a fact!
Jim Rich
___________________________________________________________________________
Date: Thu, 18 Nov 2004 14:49:11 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
Actually, the Epson gets a stream of bits that turn
the nozzles on and off. It is the Epson driver that is limited to 8-bit RGB
input. A RIP is free to take advantage of more bits, in its dither
algorithm, although as long as its input color space has a reasonable
gamma, eight bits is plenty. I’ve never seen any banding in test
gradients on my Epsons, using the stock driver.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Thu, 18 Nov 2004 19:31:18 -0500
From: Jim Rich
Subject: Re: Color bit depth
On 11/18/04 5:49 PM, Paul DeRocco wrote:
Actually, the Epson gets a stream of bits that turn
the nozzles on and off.
It is the Epson driver that is limited to 8-bit RGB
input. A RIP is free to
take advantage of more bits, in its dither algorithm,
although as long as
its input color space has a reasonable gamma, eight
bits is plenty. I’ve
never seen any banding in test gradients on my Epsons,
using the stock
driver.
Really?
Yes the Epson driver is limited but it is my
understanding is that the Epson printer can only take up to 8 bits
for printing no matter how many bits you throw at the rip.
However, I am willing to consider another point of
view. Could you comment more on this in relation to why
other rip vendors don1t take advantage of that? Is it demand or some
other factors?
Thanks
Jim Rich
___________________________________________________________________________
Date: Thu, 18 Nov 2004 19:17:11 -0500
From: Dan Margulis
Subject: RE: Color bit depth
Steve Hirsch writes,
I’m going to throw my two cents in here just for
fun, hoping I can contribute slightly to the dialog and confirm
Paul’s findings.
What Paul was attempting to argue with me about
doesn’t apply to what follows.
We did an extensive test on JPEG compression levels
and bit depth modes with actual image content from our magazines, printing
Kodak Approvals of a sample suite of images saved separately in Photoshop
JPEG compression levels from 2 through 12. We showed these samples to a
room full of industry professionals and color experts and not one of these
experienced pair of eyes could tell the difference between a compression
level of 6 and a level of 10 or 12.
Similar tests have been run elsewhere with similar
results. In one particularly notable test, GATF (I am pretty sure it was
GATF) also included uncompressed images. Everything was output to
Matchprints and a large number of experts misidentified the uncompressed
images as JPEGs. A protest was later lodged that some of the Matchprints
were misregistered and that this accounted for the result. Nevertheless, it
proved the point.
I tested the idea myself around 10 years ago. On a
Matchprint, I could tell the difference between a JPEG of around Quality 9
and an uncompressed image, but could only be sure with the aid of a loupe.
I certainly would not expect to be able to tell the difference between
levels of JPEG compression from any kind of color proof.
What I was discussing with Paul was that images that
have been JPEGged may restrict certain kinds of later correction.
Dan Margulis
___________________________________________________________________________
Date: Thu, 18 Nov 2004 17:38:52 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Jim Rich
Yes the Epson driver is limited but it is my
understanding is that the
Epson printer can only take up to 8 bits for
printing no matter how many
bits you throw at the rip.
Nope. The ESC/P protocol consists of a stream of bits
that determine which nozzles are on or off for each dot position. For
printers that have variable drop size, they send two bits per dot.
That’s what goes across the piece of wire, not RGB values.
However, I am willing to consider another point of
view. Could you comment
more on this in relation to why
other rip vendors don1t take
advantage of that? Is it demand or some other factors?
Internally, I’m sure that 16 bits are used,
since on a linear scale eight bits aren’t nearly enough to avoid
banding. However, with a simple input transfer function, the printer can be
made to have a more reasonable gamma, like 2.2, at which point
there’s really no need for any more than eight bits, at least not on
an inkjet whether the dither noise will tend to mask faint banding.
You can find out more about real-world dither
algorithms if you do a web search on gimp-print. I found a very informative
PDF file that explained the algorithms it uses.
—
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Thu, 18 Nov 2004 19:08:17 -0800
From: Peter Figen
Subject: Jpeg artifacts
While I agree that for the most part jpeg compression
is for all practical purposes invisible in any sort of output, I have
recently come across one scenario where even a #12 jpeg was unacceptable.
In an image where I was hired to add a gradated blue angelic halo around a
model’s head, which in itself was hard enough to do with no banding,
the final tiff file had a very smooth gradation all the way out to paper
white at the edges. After emailing a #12 jpeg to the client, they
complained of banding in the halo. The tiff showed no evidence and I
didn’t even think to look at the jpeg because I had never ever seen
anything, especially in a #12 compression that would ever show. I resaved a
jpeg from the tiff, and there was a very harsh banding in the cyan channel
that was clearly caused by jpeg. Now, I don’t know if this type of
example or even more specifically, the cyanish blue color made this
gradation more prone, but I now just don’t even bother with jpegs in
situation that have subtle, colored gradations to white.
Peter Figen
___________________________________________________________________________
Date: Fri, 19 Nov 2004 15:23:23 -0800
From: Paul DeRocco
Subject: RE: Jpeg artifacts
From: Peter Figen
While I agree that for the most part jpeg compression
is for all
practical purposes invisible in any sort of output, I
have
recently come across one scenario where even a #12
jpeg was
unacceptable.
I’d like to see that. Could you post a small
crop from that image somewhere, in uncompressed TIFF form?
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 22 Nov 2004 12:31:22 -0500
From: Jim Rich
Subject: Re: Color bit depth
On 11/18/04 8:38 PM, Paul DeRocco wrote:
Internally, I’m sure that 16 bits are used,
since on a linear scale eight
bits aren’t nearly enough to avoid banding.
However, with a simple input
transfer function, the printer can be made to have a
more reasonable gamma,
like 2.2, at which point there’s really no need
for any more than eight
bits, at least not on an inkjet whether the dither
noise will tend to mask
faint banding.
Paul,
I have done a little more research on this and have
found that, as you mentioned, the Epson uses a stream of bits that
determine which nozzles are on or off for each dot position. And from that
data, that there are four levels for creating dots on the print that
can be addressed. 0=off. 1= a small dot. 2 = a medium dot. 3= a large
dot.
As for using more than 8 bits internally in a rip to
avoiding banding on images such as blends and gradients, I
agree. And I have found that it really depends on the rip
vendor. However, it might be in the order of 10 or 12 bits and not
16. Though I understand that it is possible to use 16 bits. But...It
depends on other factors of the rip.
The example that I understood goes like this: If
the rip used 12 bits internally, it would have 4096 possibilities for
processing the shades of tones of an image. And that using 16
bits internally would offers over 65,000 shades of tones.
In this example, the rip vendor tests
each of these bit depth options and finds that there are no
perceptible differences in the prints from images or blends once they
got above 12 bits. But the speed hit was tremendous when using
the 16 bit implementation because the rip had to process over 60,000
more tones per image.
Now what does this have to do with image quality of
continuous tone images?
Theorectically, I think would be a quality difference
in images using more than 8 bits in a rip such as using 10 or 12 bits
or even 16 bits. However, practically, it depends on the rip vendors
implementation.
For example most third party rips use JAWs to emulate
PostScript. And today JAWs only works with 8 bits. So a rip
vendor has to figure out how to integrate that into their product without
taking quality hit. And from what I understand, that after an image is
processed via JAWs the image information is sometimes scaled up to a higher
bit depth and then processed more in the rip.
In that rip scenario any high bit image is normalized
to 8 bits before being scaled up to say 12 bits. In that situation it
then takes the high-bit argument back up stream where this list has beaten
it black and blue.
So it would seem to me in the situation I have
outlined, that sending a rip a high bit image would be a waste of disk
space and processing time.
And of course, each rip vendor develops its products
differently so they MIGHT have a different way to process high bit images,
but then again they MIGHT NOT.
What do you think?
Jim Rich
___________________________________________________________________________
Date: Mon, 22 Nov 2004 10:49:58 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Jim Rich
As for using more than 8 bits internally in a rip to
avoiding banding on
images such as blends and gradients, I agree.
And I have found
that it really depends on the rip vendor.
However, it might be in the order of 10
or 12 bits and not 16. Though I understand that it is
possible to use 16
bits. But...It depends on other factors of
the rip.
But computers don’t do 10-bit arithmetic, they
do 8-bit or 16-bit or 32-bit.
The point I was trying to make is that the algorithm
that dithers the image, converting it from however many bits down to 1 or
2, by its very nature operates in the linear (gamma = 1) domain. In this
domain, eight bits isn’t enough, so I expect the software is written
to use 16 bits, since that’s the next size up. Also, even though
there’s no such thing as a negative amount of light, the dither
algorithm can produce negative numbers (and numbers larger than 255 on an
8-bit scale), which need to be properly clipped. So the software is almost
certainly written to use short ints in C.
However, eight bits is only insufficient in the linear
domain, because you’d get nasty banding in the shadows. If the scale
is curved by applying a gamma transfer function, then eight bits is
generally good enough. So the driver accepts eight bit input, applies a
simple gamma curve, translating it to 16-bit linear, and then applies the
dither algorithm to that to reduce it to 1 or 2 bits per dot.
The end result is that the number of gray levels you
can get is limited by the 8-bit input, even though the dither algorithm
itself isn’t limited to 256 levels. However, the input 256 levels
aren’t spaced out evenly on a linear scale, but are squeezed together
at the low end by the gamma curve, to avoid banding.
Theorectically, I think would be a quality difference
in images using more
than 8 bits in a rip such as using 10 or 12 bits
or even 16
bits. However, practically, it depends on the
rip vendors implementation.
Eight bits can be marginal in B&W images. In color
gradients, the mere fact that the transitions between values don’t
line up between the three color channels masks the gradients. In B&W,
it is possible in some cases to see very faint banding, or so I’ve
heard. I’ve not seen it on my inkjets, though.
When it comes to PostScript engines, however, I have
no clue how they deal with images. I only send photographs to my printers,
not full laid-out pages.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 22 Nov 2004 14:28:36 -0500
From: Jim Rich
Subject: Re: Color bit depth
On 11/22/04 1:49 PM, Paul DeRocco wrote:
Hmm. I am not a big fan of doing math ( much less
teaching it) , but I always thought that the binary counting system let you
or the computer count, at least, in whole numbers, so using 10, 12,
14, or 16 bits was possible.
When it comes to PostScript engines, however, I have
no clue how they deal
with images. I only send photographs to my printers,
not full laid-out
pages.
Some raster rips use portions of JAWs for
things like scaling that is why I mentioned it.
It sounds like for your implementation
that you might need high bit images. And I might not understand that.
However, for most of the main stream workflows that use tools like
the Epson print driver and third party rips that interface to an ink jet
printer it appears that 8 bit images will print just fine.
Thanks for your comments
Jim Rich
___________________________________________________________________________
Date: Mon, 22 Nov 2004 12:17:12 -0700
From: Andrew Rodney
Subject: Re: Color bit depth
on 11/22/04 11:49 AM, Paul D. DeRocco wrote:
But computers don’t do 10-bit arithmetic, they
do 8-bit or 16-bit or 32-bit.
Here1s how Photoshop works (this from one of the main
engineer’s):
Photoshop deals with images either in 1 bit, 8 bit or
16 bit space. That’s it. An image may only use values
that can be represented by some other bit quantity, but once brought into
Photoshop, it gets rounded up to the smallest value that can contain it and
all values in that space are fair game from then on. So bringing in a
12 bit file will place it in 16 bit space and any image manipulation may
well add values that require more than 12 bits to save, so the file has
effectively become 16 bit. The high-bit representation in Photoshop has
always been “15 1” bits (32767 (which is the total
number of values that can be represented by 15 bits of precision) 1).
This requires 16 bits of data to represent is called “16
bit”. It is not an arbitrary decision on how to display this
data, it is displaying an exact representation of the exact data Photoshop
is using, just as 0-255 is displayed for 8 bit files.
Andrew Rodney
http://digitaldog.net/
___________________________________________________________________________
Date: Mon, 22 Nov 2004 12:06:25 -0800
From: mac townsend
Subject: Re: Color bit depth
A Harlequin rip (outputting to an imagesetter or plate
setter) is happy to provide thousands of grey levels if it is fed 16-bit
data.
Mac Townsend
Adcom Graphics Digital Imaging
Fairfield, California
___________________________________________________________________________
Date: Mon, 22 Nov 2004 18:21:18 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Jim Rich
Hmm. I am not a big fan of doing math ( much less
teaching it) , but I
always thought that the binary counting system let you
or the computer
count, at least, in whole numbers, so using 10,
12, 14, or 16 bits was
possible.
Well, sure, the programmer can add extra code to make
sure the range of results is limited to any arbitrary range, but the basic
instruction set of the CPUs used in PCs does integer arithmetic only on
8-bit bytes, 16-bit half-words, 32-bit words, and in some cases 64-bit
double-words.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 22 Nov 2004 18:23:04 -0800
From: Paul DeRocco
Subject: RE: Color bit depth
From: Andrew Rodney
This requires 16 bits of data to represent is called
“16 bit”. It is
not an arbitrary decision on how to display this data,
it is displaying an
exact representation of the exact data Photoshop is
using, just
as 0-255 is displayed for 8 bit files.
That makes sense. I suspect things are limited to 15
bits so that overflow is more easily handled.
Ciao,
Paul D. DeRocco
___________________________________________________________________________
Date: Mon, 22 Nov 2004 20:58:46 -0800
From: Mike Russell
Subject: Re: Color bit depth
Paul D. DeRocco wrote:
That makes sense. I suspect things are limited to 15
bits so that
overflow is more easily handled.
Good call. FYI, a parallel discussion in Usenet on the
unique 15 bit internal representation used by Photoshop brought forth these
items from Chris Cox just a few hours ago:
1) Because a
shift by 15 (divide by 32768) is much faster than a divide by 65535.
One of the most common operations is
(value1*value2 + (maxValue/2)) /maxValue
With 0..255 we can pull some tricks to
make the divide reasonably fast.
For 0..65535 the tricks take quite a bit
more time (and serialize theoperation), or we have to use a multiply by
reciprocal
For 0..32768, we can just use a shift.
2) A lot fewer overflows of 32 bit
accumulators
This is still a problem.
When 64 bit processors become the norm
(and the @#!^&$ OS allows a
fully 64 bit application), then that
becomes less of a problem.
3) The 2^N maximum value also has some
benefits when dealing with
subsampled lookup tables that require
interpolation.
4) the 2^N maximum value also has
benefits to blending operations that
need a middle value (for 0..255 it was
pretty random whether 127 or 128
was used for the middle).
Mike Russell
www.curvemeister.com
www.geigy.2y.net
___________________________________________________________________________
Date: Tue, 23 Nov 2004 16:49:20 EST
From: Dan Margulis
Subject: Re: Color bit depth
Paul DeRocco writes,
Look, I’ve had enough of this. First you made
the radical claim that 16
bits_never_ makes any difference in the quality of an
image.
Neither I nor anybody else I’ve heard of has
ever made such a claim. Nobody can possibly test every kind of image in
every possible production scenario.
Ordinarily I would not respond to such a post, but now
that a few days have passed, there are so many incorrect statements of fact
that I need to, in case the thread ever gets archived.
What I and other people have said is that we have
tried to produce images demonstrating that applying adjustments to a
real-world color photograph in 16-bit could produce a better result under
any conceivable set of real-world circumstances, however far-fetched, than
doing the same thing in 8-bit. So far, however, we haven’t been able
to do it in even a single image, despite having used exactly the sorts of
images and the sorts of maneuvering that we have been assured produces a
night-and-day, totally obvious,
you-are-a-recreational-user-if-you-don’t-do-this difference.
Because I have given up on ever finding an image
myself, I am reduced to hoping that someone like yourself can find that
elusive single image which will demonstrate that 16-bit is better under at
least some real-world circumstances.
I responded by making the rather innocuous claim that
it is occasionally
possible to tell the difference, but only in extreme
cases. You seemed to find
this outrageous, and challenged me to prove it.
It’s possible to tell the difference if you look
hard enough, so I believe you were suggesting not just a difference, but a
superiority. If that is correct, I don’t find the claim outrageous,
and I didn’t ask you to prove it. I politely inquired whether you had
ever actually tested one of these extreme cases in any real image with real
corrections and had found evidence that correcting it in 16-bit worked any
better than doing exactly the same thing in 8-bit. Your reply indicated
that you were not personally aware of any such image, which answered my
question and should have terminated the thread.
So I constructed what I regard as a real-world test,
cropping a piece of blue sky
out of an image, doing some strong but not
outlandish manipulations to
it in both 8-bit and 16-bit mode. This you dismissed
as not a real-world image
because the entire image was sky. Well, duh, the
sky was cropped _from_ a
real image.
A real-world image is one that could conceivably be
used standalone in any professional context, however bizarre. You say that
what you posted is a sky, but it isn’t identifiable as such. It
isn’t identifiable as anything other than a flat tint of blue, not
usable by itself in any conceivable professional context.
A real-world correction is one that a user, however
incompetent, however ill-advised, might possibly, by any stretch of the
imagination, use in any scenario, however improbable, to improve the image;
or any series of such moves. With your flat tint, I would accept six
different attempts to blur it as a real-world correction, albeit an
incompetent one. Attempts to exaggerate noise in detail-less areas, with no
possible countervailing gain, are not. Repeated attempts to correct and
then cancel are not.
Your “not outlandish” method consisted,
according to the log, of fourteen separate drastic changes, in seven pairs,
each half of which was designed to reverse the other. That is, you started
with a drastic set of curves followed by a set of curves to get back to
where you started; then went to drastic changes in both Hue and Saturation
followed by drastic changes back; then a drastic applications of Levels
followed by a drastic return to the original; then a drastic application of
Color Balance followed by a drastic Color Balance move in the opposite
direction; then a second set of drastic curves followed by a drastic
countermove; then a third set of drastic curves followed by a drastic
countermove; and finally a second drastic Hue/Saturation move followed by a
drastic return to the original; and after all this you got something that
you state you need to go to 1200% on the monitor to see a difference, and
which I stated is technically better in the 8-bit version.
To reassure you that I’m not one of those
mindless techno-geeks who blindly
believes that bit-for-bit accuracy matters, I
mentioned that once I’ve
produced a finished image, I don’t mind saving
it as quality 12 JPEG, because my
extensive, careful tests have been unable to detect a
difference between it and
an uncompressed TIFF. In response, you suddenly became
the bit-for-bit
techno-geek.
Ridiculous. What I did was AGREE with you that
workflows that involve repeated JPEGgings of the same image are probably
not advisable. I don’t regard JPEGging as causing more than trivial
problems, and I certainly don’t recommend that people stop JPEGging
their images. The point is in any event moot because, unless JPEGs start
supporting layers and alpha channels and LAB mode, few people are going to
make multiple saves.
However, in re-reading the post, I see that I
overlooked a key phrase when responding to one point. I said to detect a
JPEG, a user could simply look at the third channel on screen, where there
would be a “night and day” difference. I overlooked that you
had specified a *Quality 12* JPEG. With that quality, there certainly
isn’t a night and day difference; in fact most non-experts probably
wouldn’t see one at all. I apologize for being sloppy.
Now, you accuse me of being ignorant of statistical
analysis, and of using
terms I don’t understand.
I made no such accusation. I said “if you have
never taken statistical courses” The word “if” is not
generally taken as an accusation. If you understand the field, that’s
great. If you are of a scientific bent, then I would suggest that you do
what all scientists do when confronted with real-world results that
don´t accord with what their theories predict.
I began with a very moderate claim. I was careful not
to overstate my case:
I did not say that 8-bit sucks, or that 8-bit is
unprofessional, or that
8-bit is likely to cause problems for people, and they
should always use 16-bit.
Yet judging by the testy tone of your responses, this
obviously really upset
you.
It’s correct that you never said any of the
above things. Inasmuch as you didn’t say them (which might indeed
have upset me and/or made me testy), I was neither upset nor, as I recall,
testy. The only thing I wanted to know was whether you were personally
aware of even a single image on which 16-bit corrections worked better than
on 8-bit. You’ve answered.
When you challenged me, I did what you asked and
produced some samples that
showed what I meant, which you’ve tried to
dismiss as not “real-world” enough
for you. Fine—we live in different worlds. But
on your side, you claim to
have made tests and have found the opposite, and
mention other people who have
made similar tests, yet I don’t actually see the
tests.
Both Jim Rich and I have done extensive testing that
we have made public, using real-world images and extremely heavy, yet
conceivable, correction methods, included repeated heavy adjustments to
images that were originally so flat as to be unidentifiable. I have tested
at least half a dozen sky images. I have indicated to you where both my
printed results and the source files can be found. Jim Rich circulated his
own printed results widely. As to Martin Orpen, John Castronovo and the
others who have claimed to have run side-by-side tests, AFAIK nobody has
ever asked them to make them public.
I then made another moderate claim about light JPEG
compression, you again bristled that anyone would dare make a claim you
disagree with, so again I
produced some example images which you’ve
ignored.
You need to decide whether I ignored it or bristled at
it. The former interpretation is correct. Nobody is criticizing your JPEG
workflow. It simply has no relevance to a thread on bit depth.
And again, you claim to have counterevidence, but all
I see is your claim,
not your counterevidence.
I stated the exact location of the real-world,
real-correction, night-and-day difference, image that demonstrates beyond
doubt that at least in certain cases, certain plausible moves with certain
JPEGged files result in visible quality loss that would not have occurred
without the JPEGging. The purpose in discussing the example is not to
disparage the use of JPEG, or to say that there is no workaround for the
problem with the particular image, but simply to suggest the kind of
example that people can use to illustrate whether one method works better
than another. It is the type of demonstration that the advocates of a
16-bit workflow have been conspicuously unable to provide.
Your tone throughout this debate has been mildly rude.
I consider the
debate over, and I don’t really care who
“won”.
List members frown on rudeness and I do not
deliberately indulge in it. I apologize if you have taken anything as being
rude.
As to the “debate”, there never was one.
It was a strictly factual question, namely, are you personally aware
of any real-world color photograph that you know for a fact actually
benefited from correction in 16-bit as opposed to having done exactly the
same things in 8-bit? Since you aren’t, there’s nothing to
debate, despite your efforts to drag the JPEG irrelevancy into it.
However, if you (or any other list member) ever does
happen to run across such an image, you are most cordially invited to
notify me offline so that we can make arrangements to verify it and put it
out there for people to see. This would be a significant service to
everyone. If there is in fact a point where the use of 16-bit might become
desirable, I’d certainly like to tell my classes about it.
As always, my only requests would be that it be a
color photograph, however bad or unusual, that might conceivably be used by
itself in professional context; that the steps or series of steps taken to
“correct” it, however incompetent, are at least conceivable;
and that the provider be willing to release it for publication to
demonstrate the limited point of whether 16-bit correction can ever serve
any useful purpose at all.
Dan Margulis
___________________________________________________________________________
Date: Fri, 3 Dec 2004 20:15:08 -0000
From: "Bob Frost"
Subject: Re: Re: Color bit depth
Dan,
Having read this interesting series of posts, it seems
to me that you have given this group the wrong name; it clearly should be
RealWorldColor, not ColorTheory!
Bob Frost.
___________________________________________________________________________
Date: Fri, 03 Dec 2004 21:42:55 -0000
From:Stephen Marsh
Subject: Re: Color bit depth
Bob, I think that is where the APPLIED comes in -
Applied Colour Theory.
The high bit theory is not in debate (maths), just the
application of the theory (results).
Stephen Marsh.
___________________________________________________________________________
Date: Fri, 3 Dec 2004 15:10:42 -0700
From: Ron Kelly
Subject: Re: Re: Color bit depth
Bob:
That was tried, but voted out as too controversial.
The consensus was that Real Color could only be 11.5 bits/channel, which is
only possible to use in the hypothetical color space, and therefore *not*
real.
RK
Adobe Photoshop training classes are taught in the US by Sterling Ledet & Associates, Inc.