Working at 8 / 16 / 32 bits per channel...

Search   Help    Register    Login   Home
Fri, 13 March 2009 08:44 Go to next message
Smurfy (Photoshop expert)
How many of you work at higher bits per channel than 8?

Will gradients be rendered more smoothly in those situations?

What do printers do with those files? Any problems?

When should one work at 16 or 32 bits per channel?

Thanks.
Fri, 13 March 2009 15:56 Go to previous messageGo to next message
nomail (Photoshop expert)
Adam <no@thanks.com> wrote:

> How many of you work at higher bits per channel than 8?

Nobody can tell you how many of us do that.

> Will gradients be rendered more smoothly in those situations?

Perhaps they do, but as most screens do not support more than 8 bits,
you won't see it. Working in 16 bits matters for *editting* those
gradients (and other things).

> What do printers do with those files? Any problems?

Most printers do not use 16 bits, but sending a 16 bits file to a
printer doesn't cause any problems. The driver will change it to 8 bits.


> When should one work at 16 or 32 bits per channel?

32 bits is for HDR (High Dynamic Range) photography only.


--
Johan W. Elzenga johan<<at>>johanfoto.nl
Editor / Photographer http://www.johanfoto.com
Fri, 13 March 2009 18:15 Go to previous messageGo to next message
John J (Photoshop expert)
Johan W. Elzenga wrote:
> 32 bits is for HDR (High Dynamic Range) photography only.

Today I read a claim that editing in 16 or 32-bit reduced rounding
errors, thus increasing quality. Rubbish?

And is this just snake-oil?
http://www.epson.com/cgi-bin/Store/jsp/ProImaging/EpsonInnov ations.do?invMoreInfo=EpsonInv16BitPrinterDrivers
Fri, 13 March 2009 19:17 Go to previous messageGo to next message
gowanoh (Photoshop expert)
This is an ancient, by digital standards, issue.
The answer is the same as the medieval question about how many angels can
sit on the head of a pin.
Oils vs acrylics.
First of all many photographers would be surprised to learn that there is a
good chance they are judging images on a 6 bit LCD panel, particularly if
their computer is named after a fruit.
The widest photo printer gamuts are in the 8 bit realm, offset printing even
less.
Raw images are most commonly captured in 12 bits, some now 14, truncated to
8 bits when opened, or if opened as 16 bit images most processes will be
applied on bits that contain no information.
The idea that processing a 12 or 14 bit raw image in 16 bits is better than
processing it at 8 bits is technically correct with regard to yielding more
smoothly curved plots of the colors at the extremes of the gamut.
Does this make any practical difference when the final image will be viewed
via a medium that is incapable of accurately reproducing that color gamut
anyway?
As I recall Epson used to be very upfront in their literature on the 8 vs 16
bit issue from the printer standpoint but I don't know if they are
officially still as truthy. They showed graphs and actual color mismatches
generated by working in 16 bits.
Does it make sense to send a 16 bit file, most containing empty data, to a
printer driver that will arbitrarily strip it down to 8 bits and think this
will render a better image --if you pay Apple prices for "new" but already
obsolete Intel hardware the answer is obviously yes, but to those who still
have the capacity to reason the answer may not be yes.
If you think working in 8 bits, 16 bits, LAB color or whatever works for you
then that is all that counts.
Fri, 13 March 2009 20:42 Go to previous messageGo to next message
John J (Photoshop expert)
pupick wrote:

> First of all many photographers would be surprised to learn that there
> is a good chance they are judging images on a 6 bit LCD panel,

It would be fair to specify exactly what monitors are/were 6-bit - the
Apple 20" was one, I think. But not all of them are.
Sat, 14 March 2009 04:11 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Fri, 13 Mar 2009 08:44:04 -0400, Adam wrote:

> How many of you work at higher bits per channel than 8?

A significant number of people do

> Will gradients be rendered more smoothly in those situations?

If you mean a gradient from a photograph, no. If you mean a computer
generated gradient, yes, in some situations, particularly if you are
speaking of a mathematical smoothness, and not a visual one.

> What do printers do with those files? Any problems?

Usually no problems at all.

> When should one work at 16 or 32 bits per channel?

8 vs 16 is a personal preference. Some people simply want the extra bits,
periodm and feel that anything less is compromising quality. My main
interest is in producing tools that they can use, rather than convincing
them as to one workflow or another.

16 bits is also required to prevent artificial banding in unusually leggy
color spaces, such as ProPhoto RGB and Wide Gamut RGB. 32 bits is useful
for a number of applications, including, as Johan says, HDR images, and
other high dynamic range images such as combined astronomical images.
--
Mike Russell - http://www.curvemeister.com
Sat, 14 March 2009 10:38 Go to previous messageGo to next message
John J (Photoshop expert)
Mike Russell wrote:
> On Fri, 13 Mar 2009 08:44:04 -0400, Adam wrote:
>
>> How many of you work at higher bits per channel than 8?
>
> A significant number of people do
>
>> Will gradients be rendered more smoothly in those situations?
>
> If you mean a gradient from a photograph, no. If you mean a computer
> generated gradient, yes, in some situations, particularly if you are
> speaking of a mathematical smoothness, and not a visual one.

Mike, am I correct that the reason a photograph's graduated tone might
not be as smooth is because it does not follow a regular transition?
Sun, 15 March 2009 18:50 Go to previous messageGo to next message
nomail (Photoshop expert)
John J <nohj@droffats.ten> wrote:

> Johan W. Elzenga wrote:
> > 32 bits is for HDR (High Dynamic Range) photography only.
>
> Today I read a claim that editing in 16 or 32-bit reduced rounding
> errors, thus increasing quality. Rubbish?

No, and I never said it was. However, because there are no input devices
(cameras, scanners) that have *more than* 16 bits/color depth, there is
no reason to work in *more than* 16 bits for single images. Only HDR
gives you the need for 32 bits. That's what I said and I stand by it.

> And is this just snake-oil?
> http://www.epson.com/cgi-bin/Store/jsp/ProImaging/EpsonInnov ations.do?invM
> oreInfo=EpsonInv16BitPrinterDrivers

READ what I said, please. There are indeed *SOME* printers that can
handle 16 bits files. But like I said, *MOST* printers cannot. The
question was: does it create problems? My answer was, no it does not.
I stand by that too.


--
Johan W. Elzenga johan<<at>>johanfoto.nl
Editor / Photographer http://www.johanfoto.com
Sun, 15 March 2009 20:18 Go to previous messageGo to next message
John J (Photoshop expert)
Johan W. Elzenga wrote:
> John J <nohj@droffats.ten> wrote:
>
>> Johan W. Elzenga wrote:
>>> 32 bits is for HDR (High Dynamic Range) photography only.
>> Today I read a claim that editing in 16 or 32-bit reduced rounding
>> errors, thus increasing quality. Rubbish?
>
> No, and I never said it was. However, because there are no input devices
> (cameras, scanners) that have *more than* 16 bits/color depth, there is
> no reason to work in *more than* 16 bits for single images. Only HDR
> gives you the need for 32 bits. That's what I said and I stand by it.

I did not suggest that you mentioned rounding errors, nor did I disagree
with comments about HDR.

I simply asked whether rounding errors are a significant problem. I
think they are not.
Mon, 16 March 2009 01:18 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Sat, 14 Mar 2009 09:38:08 -0500, John J wrote:

> Mike, am I correct that the reason a photograph's graduated tone might
> not be as smooth is because it does not follow a regular transition?

Yes. This is a good description of what causes banding - an irregular
transition in one or more color channels that creates a visible edge or
ripple, aka "band" in an otherwise smooth gradient. When this happens, the
photographer has lost tonal control of the image.
--
Mike Russell - http://www.curvemeister.com
Mon, 16 March 2009 11:04 Go to previous messageGo to next message
nomail (Photoshop expert)
John J <nohj@droffats.ten> wrote:

> Johan W. Elzenga wrote:
> > John J <nohj@droffats.ten> wrote:
> >
> >> Johan W. Elzenga wrote:
> >>> 32 bits is for HDR (High Dynamic Range) photography only.
> >> Today I read a claim that editing in 16 or 32-bit reduced rounding
> >> errors, thus increasing quality. Rubbish?
> >
> > No, and I never said it was. However, because there are no input devices
> > (cameras, scanners) that have *more than* 16 bits/color depth, there is
> > no reason to work in *more than* 16 bits for single images. Only HDR
> > gives you the need for 32 bits. That's what I said and I stand by it.
>
> I did not suggest that you mentioned rounding errors, nor did I disagree
> with comments about HDR.
>
> I simply asked whether rounding errors are a significant problem. I
> think they are not.

Editting in 16 bits does give an advantage over editting in 8 bits, even
though you often do not see it clearly on your monitor. It depends on
the kind of images.


--
Johan W. Elzenga johan<<at>>johanfoto.nl
Editor / Photographer http://www.johanfoto.com
Tue, 07 April 2009 11:43 Go to previous messageGo to next message
Roberto (Photoshop expert)
Actually John rounding errors are a much bigger problem than you think. In
fact it can be used to interesting effect. Back in the days of CS2 Russell
Brown of Adobe created a free script for Photoshop called shake, rattle and
roll. This script made use of the rounding errors in Photoshop to create
some very interesting effects. While the script unfortunately has never been
updated for CS3 or CS4 you can still do it manually.

If you would like to see just how bad and how quickly rounding errors can
affect your images load a low resolution image (high resolution will work
but will take longer) and then rotate the image by 8 degrees using the
transform tool. Now repeat this over and over again. Do it 20 or 30 times
and you will see that the image is slowly destroyed, this is because of
rounding errors in Photoshop. The script I mentioned did other things for
example it would rotate back and forth, spin it around and things like this
to give some very interesting effects. If you would like to get an idea of
the effect keep rotating the image the more you do it the more dramatic the
effect.

I have tried to get Russell to update the script but he says he doesn't have
the time.

The moral of this is don't transform any more than you have to. Each time
you do you damage the image. It may take some doing to notice it, but the
damage is happening from transformation one. This is also just one area of
Photoshop that has rounding errors. Now does this mean that in 16-bit or
32-bit that you have less rounding errors or get less damge I don't know, I
don't care. I never transform an image enough to really worry about it. And,
since just about any output is 8-bit it just isn't worth the trouble in my
opinion. If others thing it is awesome, have fun.

Robert
Tue, 07 April 2009 14:58 Go to previous messageGo to next message
erpy (Photoshop begginer)
John J:
> Johan W. Elzenga wrote:
>> 32 bits is for HDR (High Dynamic Range) photography only.
>
> Today I read a claim that editing in 16 or 32-bit reduced rounding
> errors, thus increasing quality. Rubbish?
>
> And is this just snake-oil?
> http://www.epson.com/cgi-bin/Store/jsp/ProImaging/EpsonInnov ations.do?invMoreInfo=EpsonInv16BitPrinterDrivers
>

That's correct.
16bit and 32bit floating point linear editing yields much more precision
when using many layers with different effects/adjustments.
When using only a few layers it might not be that noticeable...
otherwise it makes much more difference.

As for 32bit editing in Photoshop, it depends whether Adobe is
dedicating part of the precision to HDR data or not.
Anyway, 32bit floating point editing is much more precise than 16bit
(which is integer anyway).

See: http://en.wikipedia.org/wiki/Floating_point
Tue, 07 April 2009 16:21 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Tue, 07 Apr 2009 20:58:57 +0200, erpy wrote:

> As for 32bit editing in Photoshop, it depends whether Adobe is
> dedicating part of the precision to HDR data or not.
> Anyway, 32bit floating point editing is much more precise than 16bit
> (which is integer anyway).

Any examples where any of this matters? Didn't think so.

--
Mike Russell - http://www.curvemeister.com
Tue, 07 April 2009 17:22 Go to previous messageGo to next message
mike[12] (Photoshop begginer)
In article <49db74a3$0$95539$742ec2ed@news.sonic.net>, nospam@nospam.com
says...
> Actually John rounding errors are a much bigger problem than you think. In
> fact it can be used to interesting effect. Back in the days of CS2 Russell
> Brown of Adobe created a free script for Photoshop called shake, rattle and
> roll. This script made use of the rounding errors in Photoshop to create
> some very interesting effects. While the script unfortunately has never been
> updated for CS3 or CS4 you can still do it manually.
>
> If you would like to see just how bad and how quickly rounding errors can
> affect your images load a low resolution image (high resolution will work
> but will take longer) and then rotate the image by 8 degrees using the
> transform tool. Now repeat this over and over again. Do it 20 or 30 times
> and you will see that the image is slowly destroyed, this is because of
> rounding errors in Photoshop. The script I mentioned did other things for
> example it would rotate back and forth, spin it around and things like this
> to give some very interesting effects. If you would like to get an idea of
> the effect keep rotating the image the more you do it the more dramatic the
> effect.
>
I think this has very little to do with the bit-depth of the image and
almost everything to do with the resolution of the image. For example,
create a small 8-bit per channel image (100x100 pixels square is fine)
and draw a vertical black single pixel width line down the centre of the
white background. Now apply an arbitrary rotation like 4.56 degrees
(don't choose an angle that is a factor of 360) and apply iy just once.
Now repeat by rotating in the opposite direction by the same angle.If
you zoom into the image far enough to see the individual pixels as
monotone squares you will note that the line is now a fuzzy shape
approximately 2 pixels wide (with outliers up to 3-4 pixels wide) in
shades of grey. This is not due to any lack of bit depth as the result
is almost identical if you repeat the task with a 16-bit image. Rather
it is due to the fact that, when rotated, the pixels of the line don't
map exactly onto individual new pixels so their colour value is
'smeared' across 4 or more neighbouring pixels. So it is a resolution
issue. If you repeat the experiment with a 10 pixel wide line in a
1000x1000 image, you can see that at the same scale of magnificatiobn
there is less apparent damage to the line.

Mike
Tue, 07 April 2009 21:13 Go to previous messageGo to next message
John J (Photoshop expert)
In article <49db74a3$0$95539$742ec2ed@news.sonic.net>, nospam@nospam.com
> says...
> Actually John rounding errors are a much bigger problem than you think. In
> fact it can be used to interesting effect. Back in the days of CS2 Russell
> Brown of Adobe created a free script for Photoshop called shake, rattle and
> roll. This script made use of the rounding errors in Photoshop to create
> some very interesting effects. While the script unfortunately has never been
> updated for CS3 or CS4 you can still do it manually.
>
> If you would like to see just how bad and how quickly rounding errors can
> affect your images load a low resolution image (high resolution will work
> but will take longer) and then rotate the image by 8 degrees using the
> transform tool. Now repeat this over and over again. Do it 20 or 30 times
> and you will see that the image is slowly destroyed,


I use photoshop for photographs. I don't ever need to manipulate the
image in such bizarre ways.
Wed, 08 April 2009 00:23 Go to previous messageGo to next message
Roberto (Photoshop expert)
Mike we are talking about round errors and if they can have a visible effect
on the image. If you try my suggestion you will see that the round errors
can really have an impact and pretty quickly, surprisingly quickly. I made
and make no statement on wheater 8-bit, 16-bit or 32-bit mode will increase
or reduce rounding errors. I don't know and I really don't care. What I was
trying to show is that round errors are there and they have affect your
images and the rounding errors are quite easy to encounter.

Robert
Wed, 08 April 2009 00:28 Go to previous messageGo to next message
Roberto (Photoshop expert)
No per my example no. But, it does show that if you have to do more than one
or two transformations (rotations to straighten a horrizon), resize up or
down, correct for keystoning and the like you are definately damaging your
images as you are then encountering rounding errors. All of the
tranformation tools introduce data damage to your image (except for flip
horzontal and flip verticle. All of the others damage. So if you use any of
the tools (and who knows what other tools) you are damaging your images. As
a photographer myself you can avoid having to use these tools at some point,
it is best to keep their use to a minimum and not to keep applying them over
and over to correct a single issue. Straigtening the horizon with a single
application or rotate is fine, having to apply it several times to do it
isn't. Though honest using the ruler with the rotate command I see no reason
why you would have to apply rotate more than once.

I am sure that many of the filters have rounding errors as well.

My point for my post was that rounding errors are real, they do damage your
images and if you want to see them do it I provided a way to see it. Proof
that they are real in other words. For most they won't be a problem, but one
needs to keep them in mind.

Robert
Wed, 08 April 2009 01:15 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Wed, 8 Apr 2009 09:22:12 +1200, mike wrote:

> I think this has very little to do with the bit-depth of the image and
> almost everything to do with the resolution of the image.

Good point. I did the experiment. Applying repeated rotations to an 8 bit
image, versus a 16 bit image, results in nearly the same result. Doubling
the resolution preserves image detail much better, will both 8 bit and 16
bit producing nearly identical results.

http://www.curvemeister.com/forum/index.php?topic=2569
--
Mike Russell - http://www.curvemeister.com
Wed, 08 April 2009 03:21 Go to previous messageGo to next message
Henk de Jong (Photoshop begginer)
> How many of you work at higher bits per channel than 8?
> Will gradients be rendered more smoothly in those situations?
> What do printers do with those files? Any problems?
> When should one work at 16 or 32 bits per channel?
> Thanks.


This question has been asked many times. Part of it is answered in this
article:
http://www.creativepro.com/article/out-of-gamut-the-high-bit -advantage


With kind regards,
Henk de Jong

http://www.hsdejong.nl
Nepal and Myanmar (Burma) - Photo Galleries
Wed, 08 April 2009 06:00 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Wed, 8 Apr 2009 09:21:42 +0200, Henk de Jong wrote:

> This question has been asked many times. Part of it is answered in this
> article:
> http://www.creativepro.com/article/out-of-gamut-the-high-bit -advantage
>
>
> With kind regards,
> Henk de Jong

Have you tried to duplicate what the article describes? I have.
--
Mike Russell - http://www.curvemeister.com
Wed, 08 April 2009 07:34 Go to previous messageGo to next message
Joel (Photoshop expert)
Mike Russell <groupsRE@MOVEcurvemeister.com> wrote:

> On Wed, 8 Apr 2009 09:22:12 +1200, mike wrote:
>
> > I think this has very little to do with the bit-depth of the image and
> > almost everything to do with the resolution of the image.
>
> Good point. I did the experiment. Applying repeated rotations to an 8 bit
> image, versus a 16 bit image, results in nearly the same result. Doubling
> the resolution preserves image detail much better, will both 8 bit and 16
> bit producing nearly identical results.

In general it pretty much depends on the codition of the photo, or how
much detail and large you want to print. Example

- If you have to work on some detail of a small photo (like 200-400K) then
you may want to switch to 16-bit mode.

- If you want to work on detail of a photo taken by average P&S, then you
may want to give 16-bit a try.

- If you work on a photo taken by high-end DSLR camera with good lens, and
you just want to print 8x10 or little larger then 8-bit should be fine. If
have to work on the EDGE and need to print to something like 20x30 or larger
then you may want to try 16-bit with the combination of Photoshop's Enlarger
feature.

- If you work on a photo taken by high-end camera but with poor lens or some
damaged channel etc. then you may need to give 16-bit mode a try.

There is a HUGE difference between 8-bit and 16-bit, and depend on the
photo or condition you may have to zoom in 200-300% or so to learn the
difference. With the lousy photo you should be able to see without zooming.
Wed, 08 April 2009 08:48 Go to previous messageGo to next message
John J (Photoshop expert)
Robert Barnett wrote:

> My point for my post was that rounding errors are real, they do damage
> your images and if you want to see them do it I provided a way to see
> it.

He who needs to do excessive iterative rotations for a photograph, or
has to make so very many non-adjustment layer manipulations to a
photograph can live with the degradation alleged to occur through
rounding errors.

Only someone with more monitor time than lifetime can care about such
unrealistic bench-racing metrics that require a damned spreadsheet to
'see' and still have nothing to do with real-world outcomes as it
regards photography.
Wed, 08 April 2009 11:02 Go to previous messageGo to next message
jaSPAMc (Photoshop expert)
John J <nohj@droffats.ten> found these unused words:

>Robert Barnett wrote:
>
>> My point for my post was that rounding errors are real, they do damage
>> your images and if you want to see them do it I provided a way to see
>> it.
>
>He who needs to do excessive iterative rotations for a photograph, or
>has to make so very many non-adjustment layer manipulations to a
>photograph can live with the degradation alleged to occur through
>rounding errors.
>
>Only someone with more monitor time than lifetime can care about such
>unrealistic bench-racing metrics that require a damned spreadsheet to
>'see' and still have nothing to do with real-world outcomes as it
>regards photography.

Another "Close is good enough"?
Wed, 08 April 2009 14:57 Go to previous messageGo to next message
erpy (Photoshop begginer)
Mike Russell:
> On Tue, 07 Apr 2009 20:58:57 +0200, erpy wrote:
>
>
>> As for 32bit editing in Photoshop, it depends whether Adobe is
>> dedicating part of the precision to HDR data or not.
>> Anyway, 32bit floating point editing is much more precise than 16bit
>> (which is integer anyway).
>>
>
> Any examples where any of this matters? Didn't think so.
>
>

Whenever you need a smooth result with many layers using different
blending modes.
You can do this easy test... sometimes you see this banding artifact on
TV adverts too...

- make a new 24bit image, let's say HD 1080 resolution, pick two close
colors, say RGB(0,100,100) and RGB(0,150,150)
- make a radial gradient from the center to one corner of the picture.
- Duplicate twice the layer and set the duplicates to Multiply.
Duplicate twice again and set these other duplicates to Overlay.
- Notice the banding.(100% zoom)
- Convert the image to 16 bit/component...
- The banding is gone.


--
Wed, 08 April 2009 19:09 Go to previous messageGo to next message
mike[12] (Photoshop begginer)
In article <1nb0bd6tmnvj3.dlg@mike.curvemeister.com>,
groupsRE@MOVEcurvemeister.com says...
> On Wed, 8 Apr 2009 09:22:12 +1200, mike wrote:
>
> > I think this has very little to do with the bit-depth of the image and
> > almost everything to do with the resolution of the image.
>
> Good point. I did the experiment. Applying repeated rotations to an 8 bit
> image, versus a 16 bit image, results in nearly the same result. Doubling
> the resolution preserves image detail much better, will both 8 bit and 16
> bit producing nearly identical results.
>
This leads to an interesting possibility that I had not thought of
before.

It is considered good practice by some to 'upscale' an 8 bit image to 16
bits before applying a whole bunch of colour corrections. If the image
is then set back to 8 bits there are less rounding errors in the final
image than if the whole process is carried out at 8 bits (and yes, I
know there is debate as to whether it actually makes much difference for
many images).

But what does appear a valid arguement to me is to up-scale the
resolution of an image as much as possible before applying distortions
(rotations, skew, scale, pinch, spherise, liquidise etc) then down-scale
to the previous resolution. I tried a simple test where I drew a few 1
pixel wide lines and boxes of various colours on a 100x100 image. Then
created a copy at 1000x1000 resolution. Applied identical rotation,
pinch, and skew to both images then resized the larger image back to
100x100. The first image showed significant 'smearing' of line-shape and
colours bled into neighbouring regions - the upscaled and then
downscaled image showed much less of this (in fact most of the 'damage'
to the second image was due to the downscaling process itself).

In a photographic image at 2-3000 pixel resolution this may not make
much difference unless you went wild with the distortion process - but
if you wanted to minimise 'damage' to fine details it might be worth
while in some cases.

Mike
Wed, 08 April 2009 20:52 Go to previous messageGo to next message
Joel (Photoshop expert)
mike <m.fee@irl.cri.replacethiswithnz> wrote:

> In article <1nb0bd6tmnvj3.dlg@mike.curvemeister.com>,
> groupsRE@MOVEcurvemeister.com says...
> > On Wed, 8 Apr 2009 09:22:12 +1200, mike wrote:
> >
> > > I think this has very little to do with the bit-depth of the image and
> > > almost everything to do with the resolution of the image.
> >
> > Good point. I did the experiment. Applying repeated rotations to an 8 bit
> > image, versus a 16 bit image, results in nearly the same result. Doubling
> > the resolution preserves image detail much better, will both 8 bit and 16
> > bit producing nearly identical results.
> >
> This leads to an interesting possibility that I had not thought of
> before.
>
> It is considered good practice by some to 'upscale' an 8 bit image to 16
> bits before applying a whole bunch of colour corrections. If the image
> is then set back to 8 bits there are less rounding errors in the final
> image than if the whole process is carried out at 8 bits (and yes, I
> know there is debate as to whether it actually makes much difference for
> many images).
>
> But what does appear a valid arguement to me is to up-scale the
> resolution of an image as much as possible before applying distortions
> (rotations, skew, scale, pinch, spherise, liquidise etc) then down-scale
> to the previous resolution. I tried a simple test where I drew a few 1
> pixel wide lines and boxes of various colours on a 100x100 image. Then
> created a copy at 1000x1000 resolution. Applied identical rotation,
> pinch, and skew to both images then resized the larger image back to
> 100x100. The first image showed significant 'smearing' of line-shape and
> colours bled into neighbouring regions - the upscaled and then
> downscaled image showed much less of this (in fact most of the 'damage'
> to the second image was due to the downscaling process itself).
>
> In a photographic image at 2-3000 pixel resolution this may not make
> much difference unless you went wild with the distortion process - but
> if you wanted to minimise 'damage' to fine details it might be worth
> while in some cases.
>
> Mike

As I have mentioned in other message that in general it won't make much
difference to a good hi-rez image. BUT some *original* good hi-rez image in
the wrong hand of some RAW experted (the ones with wrong impression of RAW
converter) can damage some color channel, then you may need 16-bit or 32-bit
to get around the damage.

Years ago, I used to hand around Dpewview.com retouch forum to help other
members, and I have seen many good images being destroyed by owners.
Wed, 08 April 2009 21:18 Go to previous messageGo to next message
John J (Photoshop expert)
Sir F. A. Rien wrote:
> John J <nohj@droffats.ten> found these unused words:
>
>> Robert Barnett wrote:
>>
>>> My point for my post was that rounding errors are real, they do damage
>>> your images and if you want to see them do it I provided a way to see
>>> it.
>> He who needs to do excessive iterative rotations for a photograph, or
>> has to make so very many non-adjustment layer manipulations to a
>> photograph can live with the degradation alleged to occur through
>> rounding errors.
>>
>> Only someone with more monitor time than lifetime can care about such
>> unrealistic bench-racing metrics that require a damned spreadsheet to
>> 'see' and still have nothing to do with real-world outcomes as it
>> regards photography.
>
> Another "Close is good enough"?

If one cannot find a difference in these strange abstracted deviations
from 'the the thng itself' (a photographic image) then the result is
FAPP, nill, just usenet typing and meaningless. Let the bit-twi8ddler
have their way; it has nothing to do with compelling imagery.
Wed, 08 April 2009 23:08 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Wed, 08 Apr 2009 06:34:11 -0500, Joel wrote:

> There is a HUGE difference between 8-bit and 16-bit, and depend on the
> photo or condition you may have to zoom in 200-300% or so to learn the
> difference.

I'd like to see an image that demonstrates this.
--
Mike Russell - http://www.curvemeister.com
Wed, 08 April 2009 23:10 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Wed, 08 Apr 2009 20:57:07 +0200, erpy wrote:

> Mike Russell:
>> On Tue, 07 Apr 2009 20:58:57 +0200, erpy wrote:
>>
>>
>>> As for 32bit editing in Photoshop, it depends whether Adobe is
>>> dedicating part of the precision to HDR data or not.
>>> Anyway, 32bit floating point editing is much more precise than 16bit
>>> (which is integer anyway).
>>>
>>
>> Any examples where any of this matters? Didn't think so.
>>
>>
>
> Whenever you need a smooth result with many layers using different
> blending modes.
> You can do this easy test... sometimes you see this banding artifact on
> TV adverts too...
>
> - make a new 24bit image, let's say HD 1080 resolution, pick two close
> colors, say RGB(0,100,100) and RGB(0,150,150)
> - make a radial gradient from the center to one corner of the picture.
> - Duplicate twice the layer and set the duplicates to Multiply.
> Duplicate twice again and set these other duplicates to Overlay.
> - Notice the banding.(100% zoom)
> - Convert the image to 16 bit/component...
> - The banding is gone.

Gradient doesn't count - gotta be a photograph for me to count it as an
example.
--
Mike Russell - http://www.curvemeister.com
Thu, 09 April 2009 09:35 Go to previous messageGo to next message
John Passaneau (Advanced Photoshop user)
>
> Another "Close is good enough"?
>

It is if your producing photographs. Maybe not if your producing pixels.

John Passaneau
Thu, 09 April 2009 12:26 Go to previous messageGo to next message
jaSPAMc (Photoshop expert)
John J <nohj@droffats.ten> found these unused words:

>Sir F. A. Rien wrote:
>> John J <nohj@droffats.ten> found these unused words:
>>
>>> Robert Barnett wrote:
>>>
>>>> My point for my post was that rounding errors are real, they do damage
>>>> your images and if you want to see them do it I provided a way to see
>>>> it.
>>> He who needs to do excessive iterative rotations for a photograph, or
>>> has to make so very many non-adjustment layer manipulations to a
>>> photograph can live with the degradation alleged to occur through
>>> rounding errors.
>>>
>>> Only someone with more monitor time than lifetime can care about such
>>> unrealistic bench-racing metrics that require a damned spreadsheet to
>>> 'see' and still have nothing to do with real-world outcomes as it
>>> regards photography.
>>
>> Another "Close is good enough"?
>
>If one cannot find a difference in these strange abstracted deviations
>from 'the the thng itself' (a photographic image) then the result is
>FAPP, nill, just usenet typing and meaningless. Let the bit-twi8ddler
>have their way; it has nothing to do with compelling imagery.

Aw gee, you meant the 'artistic critics' judging my photos' 'technical
merit' have been wrong? It's really the image that counts?
Thu, 09 April 2009 12:41 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Thu, 09 Apr 2009 09:26:36 -0700, Sir F. A. Rien wrote:

> John J <nohj@droffats.ten> found these unused words:
>
>>Sir F. A. Rien wrote:
>>> John J <nohj@droffats.ten> found these unused words:
>>>
>>>> Robert Barnett wrote:
>>>>
>>>>> My point for my post was that rounding errors are real, they do damage
>>>>> your images and if you want to see them do it I provided a way to see
>>>>> it.
>>>> He who needs to do excessive iterative rotations for a photograph, or
>>>> has to make so very many non-adjustment layer manipulations to a
>>>> photograph can live with the degradation alleged to occur through
>>>> rounding errors.
>>>>
>>>> Only someone with more monitor time than lifetime can care about such
>>>> unrealistic bench-racing metrics that require a damned spreadsheet to
>>>> 'see' and still have nothing to do with real-world outcomes as it
>>>> regards photography.
>>>
>>> Another "Close is good enough"?
>>
>>If one cannot find a difference in these strange abstracted deviations
>>from 'the the thng itself' (a photographic image) then the result is
>>FAPP, nill, just usenet typing and meaningless. Let the bit-twi8ddler
>>have their way; it has nothing to do with compelling imagery.
>
> Aw gee, you meant the 'artistic critics' judging my photos' 'technical
> merit' have been wrong? It's really the image that counts?

They may have been wrong, or they may have been right. The only way to
know would be to see some of your pix.

I am biased in favor of people who bring real names, as well as real
photographs, to the table. The others, I agree with Robert, are just so
much blather.
--
Mike Russell - http://www.curvemeister.com
Thu, 09 April 2009 18:45 Go to previous messageGo to next message
erpy (Photoshop begginer)
Mike Russell ha scritto:
> On Wed, 08 Apr 2009 20:57:07 +0200, erpy wrote:
>
>
>> Mike Russell:
>>
>>> On Tue, 07 Apr 2009 20:58:57 +0200, erpy wrote:
>>>
>>>
>>>
>>>> As for 32bit editing in Photoshop, it depends whether Adobe is
>>>> dedicating part of the precision to HDR data or not.
>>>> Anyway, 32bit floating point editing is much more precise than 16bit
>>>> (which is integer anyway).
>>>>
>>>>
>>> Any examples where any of this matters? Didn't think so.
>>>
>>>
>>>
>> Whenever you need a smooth result with many layers using different
>> blending modes.
>> You can do this easy test... sometimes you see this banding artifact on
>> TV adverts too...
>>
>> - make a new 24bit image, let's say HD 1080 resolution, pick two close
>> colors, say RGB(0,100,100) and RGB(0,150,150)
>> - make a radial gradient from the center to one corner of the picture.
>> - Duplicate twice the layer and set the duplicates to Multiply.
>> Duplicate twice again and set these other duplicates to Overlay.
>> - Notice the banding.(100% zoom)
>> - Convert the image to 16 bit/component...
>> - The banding is gone.
>>
>
> Gradient doesn't count - gotta be a photograph for me to count it as an
> example.
>


Oh, well... someone please ban gradients from Photoshop because "they
don't count"... :))) Very scientific, or even logical, argument.... and
a good laugh, thank you! :))

I thought the argument was about the "16 bits difference" per-se... not
"for photographs". And I was actually right.
If you read the OP, he asks in fact:

"Will gradients be rendered more smoothly in those situations?"

the answer is: "Yes"

"When should one work at 16 or 32 bits per channel?"

the answer is: "Whenever you use many blending layers" ... and even more
so in conjuction with high resolutions.








--
Thu, 09 April 2009 18:57 Go to previous messageGo to next message
erpy (Photoshop begginer)
Mike Russell ha scritto:
> On Wed, 08 Apr 2009 20:57:07 +0200, erpy wrote:
>
>
>> Mike Russell:
>>
>>> On Tue, 07 Apr 2009 20:58:57 +0200, erpy wrote:
>>>
>>>
>>>
>>>> As for 32bit editing in Photoshop, it depends whether Adobe is
>>>> dedicating part of the precision to HDR data or not.
>>>> Anyway, 32bit floating point editing is much more precise than 16bit
>>>> (which is integer anyway).
>>>>
>>>>
>>> Any examples where any of this matters? Didn't think so.
>>>
>>>
>>>
>> Whenever you need a smooth result with many layers using different
>> blending modes.
>> You can do this easy test... sometimes you see this banding artifact on
>> TV adverts too...
>>
>> - make a new 24bit image, let's say HD 1080 resolution, pick two close
>> colors, say RGB(0,100,100) and RGB(0,150,150)
>> - make a radial gradient from the center to one corner of the picture.
>> - Duplicate twice the layer and set the duplicates to Multiply.
>> Duplicate twice again and set these other duplicates to Overlay.
>> - Notice the banding.(100% zoom)
>> - Convert the image to 16 bit/component...
>> - The banding is gone.
>>
>
> Gradient doesn't count - gotta be a photograph for me to count it as an
> example.
>

Oh, and BTW, the digital photography argument would be too painful for
users like you to take on.
That's because it would show you both why you don't notice much
difference between an 8bit and a 16bit photo and at the same time how
you're being ripped off for good even with the most expensive camera on
the whole market. (well, *maybe* excluding top-of-the-line Hasselblads)







--
Thu, 09 April 2009 19:36 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Fri, 10 Apr 2009 00:57:50 +0200, erpy wrote:

> Oh, and BTW, the digital photography argument would be too painful for
> users like you to take on.
> That's because it would show you both why you don't notice much
> difference between an 8bit and a 16bit photo and at the same time how
> you're being ripped off for good even with the most expensive camera on
> the whole market. (well, *maybe* excluding top-of-the-line Hasselblads)

I think you, and to be fair many others, are making the simplest of
tautologies: "more bits is more better". Your discussion of an 8 bit
gradient versus a 16 bit gradient can be equally well applied to a 16 bit
gradient versus a 32 bit gradient, 32 bit and 64 bit, and so on. As Buzz
Lightyear says, "to infinity, and beyond!".

At some point, I'm sure you will agree, the number of bits becomes "lost in
the noise".For a photograph, the bedrock of whether this reasoning is valid
or not should be in the form of an image that shows the effect. For a
computer generated gradient, there is no noise, and therefore the tautology
holds in Buzz Lightyear.

My challenge, which still stands, is to present a photographic image, in a
normal color space such as sRGB or Adobe RGB, that edits better in 8 bit
than 16 bit. Do you have one, or not? That's all, very simple.

Here are some examples of quality variation that we can agree is real:

1) variation of image quality with jpeg quality - easily demonstrated, and
not a source of discussion.
2) variation of image quality in the face of repeated transforms as a
function of bit depth, and of image resolution. I presented an example
illustrating this earlier in this thread.
3) variation of image quality for indexed versus 24 bit RGB images. Again,
an example of this is easily provided.
4) variation of image quality with ISO setting.
5) variation of image quality with aperture setting.
6) variation of image quality with image stabilization.
7) the difference in image quality between a high quality jpeg and an
uncompressed TIFF or raw image.

Each of the above are easily demonstrated using *photographic images*.
But ah, according to you (and to be fair, others), variation of image
quality versus bit depth is blazingly obvious. But, unlike the above seven
examples, it is not so obvious that you can point to an example photograph,
just a computer generated gradient which d nothing to do with nothing.

--
Mike Russell - http://www.curvemeister.com
Thu, 09 April 2009 20:29 Go to previous messageGo to next message
John J (Photoshop expert)
Mike Russell wrote:
> [...]
> They may have been wrong, or they may have been right. The only way to
> know would be to see some of your pix.
>
> I am biased in favor of people who bring real names, as well as real
> photographs, to the table. The others, I agree with Robert, are just so
> much blather.

Mike, what would you like? URLs to digital representations of pictures?
Thu, 09 April 2009 20:37 Go to previous messageGo to next message
Mike Russell (Photoshop expert)
On Thu, 09 Apr 2009 19:29:37 -0500, John J wrote:

> Mike, what would you like? URLs to digital representations of pictures?

Sounds fine.

BTW, just to fill in a little bit, Rien and I have had this discussion
before. His name is a play on the words "ca ne fait rien" meaning, "that
doesn't matter at all". He is tickled, a little, when people recognize the
pun.
--
Mike Russell - http://www.curvemeister.com
Thu, 09 April 2009 21:07 Go to previous messageGo to next message
Joel (Photoshop expert)
Mike Russell <groupsRE@MOVEcurvemeister.com> wrote:

<snip>
> At some point, I'm sure you will agree, the number of bits becomes "lost in
> the noise".For a photograph, the bedrock of whether this reasoning is valid
> or not should be in the form of an image that shows the effect. For a
> computer generated gradient, there is no noise, and therefore the tautology
> holds in Buzz Lightyear.

Talking about rockbed then why not using ROCK as an example. Now

- Throwing (100) 100x100" rocks on the ground

- Then also (10,000) 1x1" rocks on the ground

And which one looks smoother (flatter, more evenly)?

That's pretty much same with 8-bit and 16-bit, or you will get more pixel
from 16-bit then 8-bits, and if you spread them around then the more pixel
the smoother.
Thu, 09 April 2009 21:51 Go to previous messageGo to previous message
Mike Russell (Photoshop expert)
On Thu, 09 Apr 2009 20:07:19 -0500, Joel wrote:

> That's pretty much same with 8-bit and 16-bit, or you will get more pixel
> from 16-bit then 8-bits, and if you spread them around then the more pixel
> the smoother.

You've got it backwards. If you keep the file size the same, an 8 bit
image will store exactly twice the number of pixels as a 16 bit image.

IOW, if you are going to double the size of the file, it's better to have
twice as many 8 bit pixels, as it is to have the same number of 16 bit per
channel pixels.
--
Mike Russell - http://www.curvemeister.com
Create a new topic Submit Reply
Previous Topic: Drobo or Raid?
Goto Forum:

Adobe Photoshop

Adobe Photoshop techniques & help
Photoshop for Windows
Photoshop for Mac
Technical questions

Photoshop Elements

Adobe Photoshop Elements