Computer Graphics with Photoshop - Photoshop CS

Author

Photoshop CS - 15-bit 16-bit mode.

JPS@no.komm

2004-05-16, 4:28 am

16-bit mode is actually 15-bit (maximum values are 32,767), according to
the Info tool.

Is this internal, or a bug in the Info tool?
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

Toby Thain

2004-05-16, 9:28 am

JPS@no.komm wrote in message news:<4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>...
> 16-bit mode is actually 15-bit (maximum values are 32,767), according to
> the Info tool.
>
> Is this internal, or a bug in the Info tool?

As Mr Cox will be delighted to inform you, Photoshop's "16"-bit mode
is indeed quasi-15-bit, channels have an internal range of 0-32768
(inclusive). See this recent discussion on Adobe Forums:
http://www.adobeforums.com/cgi-bin/...wcs.8@.3bb3f466

Adobe does not seem to consider it a bug, but those who need to work
with 16 bit data may disagree.

T

jjs

2004-05-18, 7:28 pm

In article <4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>, JPS@no.komm wrote:

> 16-bit mode is actually 15-bit (maximum values are 32,767), according to
> the Info tool.
>
> Is this internal, or a bug in the Info tool?

32767 is 16-bit signed integer max. I wonder what they use the <0 values for.

Chris Cox

2004-05-18, 7:28 pm

Who said it was signed?

The range is 0 to 32768.

Chris

In article <john-1605040852140001@m-0-135.docsis.hbci.com>, jjs
<john@xyzzy.stafford.net> wrote:

> In article <4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>, JPS@no.komm wrote:
>
>
> 32767 is 16-bit signed integer max. I wonder what they use the <0 values for.

jjs

2004-05-18, 7:28 pm

In article <160520041717368533%ccox@mindspring.com>, Chris Cox
<ccox@mindspring.com> wrote:

> Who said it was signed?

Not me. I only set the bait.

> The range is 0 to 32768.

And that's plenty.

Toby Thain

2004-05-18, 7:28 pm

john@xyzzy.stafford.net (jjs) wrote in message news:<john-1605040852140001@m-0-135.docsis.hbci.com>...
> In article <4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>, JPS@no.komm wrote:
>
>
> 32767 is 16-bit signed integer max. I wonder what they use the <0 values for.

Photoshop's quasi-15-bit upper value is +32768, so the 16 bit value is
regarded as unsigned. (See Adobe Forums thread cited above.)

T

Toby Thain

2004-05-18, 7:28 pm

Toby Thain

2004-05-18, 7:28 pm

john@xyzzy.stafford.net (jjs) wrote in message news:<john-1605040852140001@m-0-135.docsis.hbci.com>...
> In article <4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>, JPS@no.komm wrote:
>
>
> 32767 is 16-bit signed integer max. I wonder what they use the <0 values for.

Photoshop's quasi-15-bit upper value is +32768, so the 16 bit value
should be regarded as unsigned. (See Adobe Forums thread cited above.)

T

Stephen H. Westin

2004-05-18, 7:28 pm

Chris Cox <ccox@mindspring.com> writes:

> Who said it was signed?
>
> The range is 0 to 32768.

Which is just beyond a 15-bit unsigned integer. 2^16 was 65,536, last
I checked. That said, I don't know of any digital camera that gives
even 15 useful bits, much less 16.

<snip>

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Mike Russell

2004-05-18, 7:28 pm

Stephen H. Westin wrote:
> Chris Cox <ccox@mindspring.com> writes:
>
>
> Which is just beyond a 15-bit unsigned integer. 2^16 was 65,536, last
> I checked. That said, I don't know of any digital camera that gives
> even 15 useful bits, much less 16.

Yes, but medical data, and CGI data can generate that much data.

There is also the question of dynamic range. Luminance maps (which often
use floating point) and other data generated by combining several
conventional photographs can easlily use up 15-16 bits of dynamic range.

All that said, yes, for photographs you can't see much past 8 bits per
channel gamma encoded, and our output devices are more limited than our
eyes, particularly if you consider that fine art prints are where the money
is, not CRT display, but that does not stop some of us from striving for
more.
--

Mike Russell
www.curvemeister.com
www.geigy.2y.net

Bart van der Wolf

2004-05-18, 7:28 pm

"Mike Russell" <REgeigyMOVE@pacbellTHIS.net> wrote in message
news:rq7qc.49955$sR1.14925@newssvr29.news.prodigy.com...
SNIP
> Yes, but medical data, and CGI data can generate that much data.

Indeed, and changing colorspace, resizing, sharpening, etc. will also use
whatever is available to preseve accuracy, until reduction to 8-b/ch.

Bart

Toby Thain

2004-05-18, 7:28 pm

westin*nospam@graphics.cornell.edu (Stephen H. Westin) wrote in message news:<s0k6zbrm8t.fsf@diesel.graphics.cornell.edu>...
> Chris Cox <ccox@mindspring.com> writes:
>
>
> Which is just beyond a 15-bit unsigned integer. 2^16 was 65,536, last
> I checked. That said, I don't know of any digital camera that gives
> even 15 useful bits, much less 16.

Who said it was a 15-bit unsigned integer? It's a 16-bit unsigned
field... with an internal maximum less than the implied 65535.

And as for the source of 16-bit data, there are many ways to get it...
synthetic renderings, scientific and medical images (as others have
mentioned), etc. Many formats offer well beyond 16 bits per channel.

TIFF and SGI formats, to name two common ones, offer full 16 bits per
channel.

T

>
> <snip>

Stephen H. Westin

2004-05-18, 7:28 pm

toby@telegraphics.com.au (Toby Thain) writes:

> westin*nospam@graphics.cornell.edu (Stephen H. Westin) wrote in message news:<s0k6zbrm8t.fsf@diesel.graphics.cornell.edu>...
>
> Who said it was a 15-bit unsigned integer?

Well, I kinda did. It's 15 bits unsigned, plus one other possible value.
A distinction without a difference, as far as the results are concerned.

> It's a 16-bit unsigned
> field... with an internal maximum less than the implied 65535.
>
> And as for the source of 16-bit data, there are many ways to get it...
> synthetic renderings, scientific and medical images (as others have
> mentioned), etc. Many formats offer well beyond 16 bits per channel.

Definitely true. But I'd bet that well over 99% of the pixels run
through Photoshop come from scanners and digital cameras, and are
displayed on monitors or paper prints with effective dynamic range
within the capability of 8 bits. After all, retouching and alteration
aren't really considered useful things to do to medical imagery...

> TIFF and SGI formats, to name two common ones, offer full 16 bits per
> channel.

Oh, yes. 32-bit float in TIFF, for example.

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Paul J Gans

2004-05-18, 11:28 pm

jjs <john@xyzzy.stafford.net> wrote:
>In article <4p2ea01npnvse7api9gn5lkl947ra9npp6@4ax.com>, JPS@no.komm wrote:

[color=darkred]
>32767 is 16-bit signed integer max. I wonder what they use the <0 values for.

Since most 16-bit material comes from RAW camera images, the question
is: are those true 16-bit images and if so, how does Photoshop
handle them?

----- Paul J. Gans

Paul J Gans

2004-05-18, 11:28 pm

Chris Cox <ccox@mindspring.com> wrote:

>Who said it was signed?

>The range is 0 to 32768.

>Chris

This is wrong no matter how it is interpreted. 16 bits can represent
65536 different values. 15 bits can represent 32768. If Photoshop
16-bit mode can only represent 32768 values then it is *not* a
16-bit mode.

On the other hand, if source images such as those from various
cameras (Canon, Nikon, etc.) claiming to be 16-bit images only
contain 32768 different values, then they are not 16-bit images
either, but 15-bit images.

To make matters worse, even if the range is as Chris states
it above, there are 32769 values in the range from 0 to 32768.

All in all, I suspect one or many of us have slightly mistated
things.

--- Paul J. Gans

>In article <john-1605040852140001@m-0-135.docsis.hbci.com>, jjs
><john@xyzzy.stafford.net> wrote:
[color=darkred]

Bart van der Wolf

2004-05-18, 11:28 pm

"Paul J Gans" <gans@panix.com> wrote in message
news:c8e6qr$3k2$1@reader2.panix.com...
SNIP
> Since most 16-bit material comes from RAW camera images, the question
> is: are those true 16-bit images and if so, how does Photoshop
> handle them?

In most cases they are 12-14 bit/channel images, of which approx. 1-2 bits
noise. But as soon as you sharpen, convert color profile, etc, you can use
all bits the internal engine supports.

Bart

Paul J Gans

2004-05-18, 11:28 pm

Stephen H. Westin <westin*nospam@graphics.cornell.edu> wrote:
>Chris Cox <ccox@mindspring.com> writes:

[color=darkred]
>Which is just beyond a 15-bit unsigned integer. 2^16 was 65,536, last
>I checked. That said, I don't know of any digital camera that gives
>even 15 useful bits, much less 16.

><snip>

RAW mode in Canon and other cameras and NEF mode in Nikon
cameras supposedly produces an image with 16 bits per color
channel per pixel.

Camera makers supply software to converst such images to TIFFs
(or JPEGS). The user can adjust the ranges converted thus going
for shadow detail or highlight detail as they see fit.

I've never had reason to doubt that those images did have 16 bits
per color channel. Perhaps I should have.

In any event, the practical question is whether to load the
RAW or NEF mode image into Photoshop (it will, with Adobe
supplied plug-ins, accept either) and do the conversion there
or to do the conversion in the manufacturers' less-capable
software.

However if Photoshop is indeed throwing away half the color
range....

---- Paul J. Gans

Bill Hilton

2004-05-18, 11:28 pm

>From: Paul J Gans gans@panix.com
>
>RAW mode in Canon and other cameras and NEF mode in Nikon
>cameras supposedly produces an image with 16 bits per color
>channel per pixel.
>
>I've never had reason to doubt that those images did have 16 bits
>per color channel. Perhaps I should have.

Actually most of the digital cameras output 12 bits/channel or 8 bits/channel,
not 16. I have the Canon 10D and 1Ds and both are 12 bits max, even though the
files open in 16 bit mode in Photoshop.

Also, many scanners are 12 bit or 14 bit/channel, with some of the newer ones
true 16 bits. I think my first film scanner was 12 bits (Nikon LS-2000) and
the one I'm using now is 14 bits/channel (LS-8000).

All 10 bit, 12 bit and 14 bit files get saved as if they were 16 bits because
it's easier to pack the data into 2 eight bit bytes, even if they have to fill
the upper bits with zeros. These files all open in 16 bit mode in Photoshop
but actual data is limited to what the capture device provided. Maybe that's
where you got the idea that all these images actually have 16 bits/channel, but
most of them don't.

I think 12 bits for photographic images is very useful but much beyond that and
you're reaching the point of diminishing returns, based on editing thousands of
8, 12 and 14 bit/channel images.

>In any event, the practical question is whether to load the
>RAW or NEF mode image into Photoshop (it will, with Adobe
>supplied plug-ins, accept either) and do the conversion there
>or to do the conversion in the manufacturers' less-capable
>software.

They are still 12 bits (at least the Canon RAW files).

>However if Photoshop is indeed throwing away half the color
>range....

LOL ... don't think so :)

Bill

Bill Hilton

2004-05-18, 11:28 pm

>From: Paul J Gans gans@panix.com

>On the other hand, if source images such as those from various
>cameras (Canon, Nikon, etc.) claiming to be 16-bit images only
>contain 32768 different values, then they are not 16-bit images
>either, but 15-bit images.

12 bits/channel ... see my other post.

Paul J Gans

2004-05-18, 11:28 pm

Bart van der Wolf <bvdwolf@no.spam> wrote:

>"Paul J Gans" <gans@panix.com> wrote in message
>news:c8e6qr$3k2$1@reader2.panix.com...
>SNIP
[color=darkred]
>In most cases they are 12-14 bit/channel images, of which approx. 1-2 bits
>noise. But as soon as you sharpen, convert color profile, etc, you can use
>all bits the internal engine supports.

>Bart

Ok. Thanks.

----- Paul J. Gans

Paul J Gans

2004-05-18, 11:28 pm

Bill Hilton <bhilton665@aol.comedy> wrote:
[color=darkred]
>Actually most of the digital cameras output 12 bits/channel or 8 bits/channel,
>not 16. I have the Canon 10D and 1Ds and both are 12 bits max, even though the
>files open in 16 bit mode in Photoshop.

>Also, many scanners are 12 bit or 14 bit/channel, with some of the newer ones
>true 16 bits. I think my first film scanner was 12 bits (Nikon LS-2000) and
>the one I'm using now is 14 bits/channel (LS-8000).

>All 10 bit, 12 bit and 14 bit files get saved as if they were 16 bits because
>it's easier to pack the data into 2 eight bit bytes, even if they have to fill
>the upper bits with zeros. These files all open in 16 bit mode in Photoshop
>but actual data is limited to what the capture device provided. Maybe that's
>where you got the idea that all these images actually have 16 bits/channel, but
>most of them don't.

>I think 12 bits for photographic images is very useful but much beyond that and
>you're reaching the point of diminishing returns, based on editing thousands of
>8, 12 and 14 bit/channel images.

[color=darkred]
>They are still 12 bits (at least the Canon RAW files).

[color=darkred]
>LOL ... don't think so :)

>Bill

Thanks. I certainly hoped Photoshop wasn't throwing
any information away.

By the way (I've not tried this yet) I assume that one
can save the image using the upper 8 bits out of 12 and
save it again using the lower 8. These two can then
be combinded to give an image having mor detail in
the highlights and the shadows than an ordinary 8 bitter
would have. Or am I off-base?

---- Paul J. Gans

Toby Thain

2004-05-19, 4:28 am

westin*nospam@graphics.cornell.edu (Stephen H. Westin) wrote in message news:<s03c5xsrrt.fsf@diesel.graphics.cornell.edu>...
> [...] But I'd bet that well over 99% of the pixels run
> through Photoshop come from scanners and digital cameras, and are
> displayed on monitors or paper prints with effective dynamic range
> within the capability of 8 bits. After all, retouching and alteration
> aren't really considered useful things to do to medical imagery...

Fair enough, but Photoshop still has plenty of conventional "image
processing" features - histogramming, filtering, convolutions, format
conversions, channel arithmetic - aside from its main use (attaching
celebrity heads to anonymous bodies).

Returning to the topic, the fact that it "munges" 16 bit data into a
smaller dynamic range is perhaps one of the strongest indications that
the program is not really aimed at serious image processing
applications.

>
>
> Oh, yes. 32-bit float in TIFF, for example.

As well as true 16-bit integer channels. And any HDR image.
T

Stephen H. Westin

2004-05-20, 9:28 am

bhilton665@aol.comedy (Bill Hilton) writes:

>
>
> 12 bits/channel ... see my other post.

from a non-cooled digital camera. Chilled cameras routinely achieve 14
or 15 bits, and I'm sure the proper application of liquid nitrogen can
improve on that. But since almost all digital cameras used for general
photography are unchilled, 12 bits is the best possible at the moment.

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Stephen H. Westin

2004-05-20, 9:28 am

toby@telegraphics.com.au (Toby Thain) writes:

> westin*nospam@graphics.cornell.edu (Stephen H. Westin) wrote in message news:<s03c5xsrrt.fsf@diesel.graphics.cornell.edu>...

<snip>

>
> As well as true 16-bit integer channels. And any HDR image.

Well, the Ward HDR format actually has 8 bits per channel mantissa,
with an 8-bit exponent shared among the three channels. So it's not
true 16-bit precision. That's one of the reasons for OpenEXR, logLUV
TIFF, and the like.

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Bart van der Wolf

2004-05-20, 9:28 am

"Stephen H. Westin" <westin*nospam@graphics.cornell.edu> wrote in message
news:s0lljor09u.fsf@diesel.graphics.cornell.edu...
> bhilton665@aol.comedy (Bill Hilton) writes:
SNIP
>
> from a non-cooled digital camera. Chilled cameras routinely achieve 14
> or 15 bits, and I'm sure the proper application of liquid nitrogen can
> improve on that.

At the risk of wandering too much away from Photoshop ;-):
If the camera's ADC is only 12-bit, even liquid Nitrogen can't change
that...

Bart

Stephen H. Westin

2004-05-20, 9:28 am

"Bart van der Wolf" <bvdwolf@no.spam> writes:

> "Stephen H. Westin" <westin*nospam@graphics.cornell.edu> wrote in message
> news:s0lljor09u.fsf@diesel.graphics.cornell.edu...
> SNIP
>
> At the risk of wandering too much away from Photoshop ;-):
> If the camera's ADC is only 12-bit, even liquid Nitrogen can't change
> that...

That's absolutely correct. But I wasn't thinking in that direction; I
was talking about physical limitations as they stand today. Without
cooling, bits beyond 12 (for a DSLR-type sensor) or even 10 (for a
consumer-type sensor) will be pretty much pure noise.

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Bill Hilton

2004-05-20, 9:28 am

>> 12 bits/channel (for Canon and Nikon dSLR's)... see my other post.

>From: westin*nospam@graphics.cornell.edu (Stephen H. Westin)
>
>from a non-cooled digital camera. Chilled cameras routinely achieve 14
>or 15 bits, and I'm sure the proper application of liquid nitrogen can
>improve on that.

LOL ... I think the LCD would freeze with liquid nitrogen :) The Canon dSLR's
are rated to function down to 0 C, below that you're on your own.

Bill

Stephen H. Westin

2004-05-20, 9:28 am

bhilton665@aol.comedy (Bill Hilton) writes:

>
>
> LOL ... I think the LCD would freeze with liquid nitrogen :) The Canon dSLR's
> are rated to function down to 0 C, below that you're on your own.

I did say the *proper* application. Apparently astronomers do this all
the time, cooling just the sensor chip and of course building the
"camera" from scratch. Our cooled camera runs the chip at -25C, and
has the chip in a vacuum chamber to prevent the formation of
condensation and frost on the sensor. Noise is pretty low...

--
-Stephen H. Westin
Any information or opinions in this message are mine: they do not
represent the position of Cornell University or any of its sponsors.

Paul J Gans

2004-05-20, 9:28 am

Bill Hilton <bhilton665@aol.comedy> wrote:
[color=darkred]
[color=darkred]
>LOL ... I think the LCD would freeze with liquid nitrogen :) The Canon dSLR's
>are rated to function down to 0 C, below that you're on your own.

That's a bummer. What does one do in the snow? Both my
film slrs function quite well below zero C. So does my Nikon
950.

---- Paul J. Gans

Chris Cox

2004-05-23, 12:08 pm

In article <c8e74j$3k2$2@reader2.panix.com>, Paul J Gans
<gans@panix.com> wrote:

> Chris Cox <ccox@mindspring.com> wrote:
>
>
>
>
> This is wrong no matter how it is interpreted.

No, it is correct.

> 16 bits can represent
> 65536 different values.

Yes, it can.

> 15 bits can represent 32768. If Photoshop
> 16-bit mode can only represent 32768 values then it is *not* a
> 16-bit mode.

No, it represents 32769 values. I said the range was 0 to 32768.

> On the other hand, if source images such as those from various
> cameras (Canon, Nikon, etc.) claiming to be 16-bit images only
> contain 32768 different values, then they are not 16-bit images
> either, but 15-bit images.

So far none of them produce more than 14 bits/channel (until you get to
some really expensive scientific cameras).

Chris

Paul J Gans

2004-05-23, 7:14 pm

Chris Cox <ccox@mindspring.com> wrote:
>In article <c8e74j$3k2$2@reader2.panix.com>, Paul J Gans
><gans@panix.com> wrote:

[color=darkred]
>No, it is correct.

[color=darkred]
>Yes, it can.

[color=darkred]
>No, it represents 32769 values. I said the range was 0 to 32768.

Semantics. Sixteen bits can represent 65536 values and they
are there even if Photoshop only uses the range you give. It
is a strange range, but that's esthetics.

[color=darkred]
>So far none of them produce more than 14 bits/channel (until you get to
>some really expensive scientific cameras).

I agree that fewer bits are significant, but the camera outputs
16 bits. We humans know that several of the least significant
bits are useless and the software can implement that. But the
hardware format contains many more values.

---- Paul J. Gans

Toby Thain

2004-05-24, 12:14 pm

Paul J Gans <gans@panix.com> wrote in message news:<c8qhs8$3ot$1@reader2.panix.com>...
> Chris Cox <ccox@mindspring.com> wrote:
> ...
^^^^^^[color=darkred]
>
> I agree that fewer bits are significant, but the camera outputs
> 16 bits. We humans know that several of the least significant
> bits are useless and the software can implement that. But the
> hardware format contains many more values.

"So far" being the key words. Sooner or later they'll have to fix it.
It will probably mean a leapfrog to 32 or 64 bit channels (&/or
float/double/HDR).

Toby

>
> ---- Paul J. Gans