printing floating point images

radiance-general
1

I wonder if anyone on this list has experience in printing 32 bit floating
point images (radiance pics preferably)? Some years ago I had some vibrant
discussion with Iebele Abel on this matter. I find it a rather frustrating
idea to thrash spending much time and effort getting a physically accurate
rendered image, and thrashing a tremendous amount of image fidelility to get
the image printed, there where it counts most. I've been able to get 16-bit
data images, but what I'm interested in is getting the full scope.
The relevant question indeed is what would be limiting output quality.
Its quite likely paper of photographic means are only able to output so many
colours. I refer to physical limitations, rather than those implied by a
data format.

It seems that film industry has successfully overcome getting lack of colour
fidelity, also thank to the openEXR format (basically a clumsy rip-off of
the radiance image format??) but mostly by actually being able to output the
full scope of the data.

So how would I get this done in print, its quite a concern to me.

Best,

Jelle.

2

Jelle,

This is a very interesting question, and one in which I have an
interest in following. I'll share with you my limited knowledge.

It has been my knowledge from speaking and reading about display
devices (mainly reflective print and transparency film) that film
can record a greater range of brightnesses, allowing it greater
bit depth. Print is hindered by its reflective nature, and the
upper and lower bounds of reflectivity (users of Radiance are
more aware than most that what is normally percieved as white is
probably no more than 70-80% reflective).

I have been assured by professional digital printers that a
good 8-bit-per-channel image gets printed just as well as an
image with grater depth. I may have even been told that the
rasterizer upgrades the bit depth as it upsamples the image.
I can't confirm that, though.

I usually print on a Lightjet 5000. I use ra_tiff to convert my
.pic into a 16-bit-per-channel TIFF, then I use Cinepaint on
Linux to do final adjustments, outputting to an 8bpc TIFF.

Mark

···

On Wed, 1 Dec 2004, Jelle Feringa // EZCT / Paris wrote:

I wonder if anyone on this list has experience in printing 32 bit floating
point images (radiance pics preferably)? Some years ago I had some vibrant
discussion with Iebele Abel on this matter. I find it a rather frustrating
idea to thrash spending much time and effort getting a physically accurate
rendered image, and thrashing a tremendous amount of image fidelity to get
the image printed, there where it counts most. I've been able to get 16-bit
data images, but what I'm interested in is getting the full scope.
The relevant question indeed is what would be limiting output quality.
Its quite likely paper of photographic means are only able to output so many
colours. I refer to physical limitations, rather than those implied by a
data format.

It seems that film industry has successfully overcome getting lack of colour
fidelity, also thank to the openEXR format (basically a clumsy rip-off of
the radiance image format??) but mostly by actually being able to output the
full scope of the data.

So how would I get this done in print, its quite a concern to me.

Best,

Jelle.

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

3

Jelle,

Like Mark I will add in what I know and what we have done in the past to get the best quality prints possible (for a reasonable cost).

In general you need to print to a print device that goes direct rgb. I think that a fair amount of progress has been made for printing digital images (rgb) direct to photographic quality paper (paper makes a huge difference). We have had good success with two devices and one service:

   1. Lambda Printer - this is a big fancy printer that only a large
      service bureau would have (I think that the cost around $500,000
      US). They can print to very large paper sizes (photo paper). I do
      not remember what the resolution is, although we found that you
      could have quite good image quality even at 100dpi. Cost - expensive.
   2. Kodak Pegasus Printer- cannot print as large as the lambda but is
      direct to photo paper. Is more sensitive to image resolution, I
      think they prefer 250dpi. Cost - medium to expensive.
   3. Online Service Bureaus - we have also used Ofoto on occasion.
      Again the print is direct to photo paper. Cost - INEXPENSIVE!

Depending on the device selected, I expect that one could also prepare some kind of calibration to make sure the images are calibrated to the device (which means that you have to make sure your other output devices such as monitors are correctly calibrated as well).

Of course none of this answers your question about bit depth. I guess one question to ask is how much is needed to achieve what you want. Alternatively, another way to look at it is to learn what professional photographers using high end digital cameras do for printing, I can only imagine that this could get quite expensive however.

-Jack

Mark Stock wrote:

···

Jelle,

This is a very interesting question, and one in which I have an
interest in following. I'll share with you my limited knowledge.

It has been my knowledge from speaking and reading about display
devices (mainly reflective print and transparency film) that film
can record a greater range of brightnesses, allowing it greater
bit depth. Print is hindered by its reflective nature, and the
upper and lower bounds of reflectivity (users of Radiance are
more aware than most that what is normally percieved as white is
probably no more than 70-80% reflective).

I have been assured by professional digital printers that a
good 8-bit-per-channel image gets printed just as well as an
image with grater depth. I may have even been told that the
rasterizer upgrades the bit depth as it upsamples the image.
I can't confirm that, though.

I usually print on a Lightjet 5000. I use ra_tiff to convert my
.pic into a 16-bit-per-channel TIFF, then I use Cinepaint on
Linux to do final adjustments, outputting to an 8bpc TIFF.

Mark

On Wed, 1 Dec 2004, Jelle Feringa // EZCT / Paris wrote:

I wonder if anyone on this list has experience in printing 32 bit floating
point images (radiance pics preferably)? Some years ago I had some vibrant
discussion with Iebele Abel on this matter. I find it a rather frustrating
idea to thrash spending much time and effort getting a physically accurate
rendered image, and thrashing a tremendous amount of image fidelity to get
the image printed, there where it counts most. I've been able to get 16-bit
data images, but what I'm interested in is getting the full scope.
The relevant question indeed is what would be limiting output quality.
Its quite likely paper of photographic means are only able to output so many
colours. I refer to physical limitations, rather than those implied by a
data format.

It seems that film industry has successfully overcome getting lack of colour
fidelity, also thank to the openEXR format (basically a clumsy rip-off of
the radiance image format??) but mostly by actually being able to output the
full scope of the data.

So how would I get this done in print, its quite a concern to me.

Best,

Jelle.

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
# John E. de Valpine
# president
#
# visarc incorporated
# http://www.visarc.com
#
# channeling technology for superior design and construction

4

Hi!

In general you need to print to a print device that goes direct rgb.

??? You mean a driver that can do the rgb to cmyk translation, without the
need to do that in an application? Just wondering, I never heard about
rgb-printers, do you mean transfering to a film (which is like foto and as
such rgb)?

Regarding printers, it is a big help to use printers that don't offer just
cmyk, but also the light colors (there are 6 and 8 color printers), as this
avoids raster problems in bright regions of your renderings.

Calibration... afaik the radiance output is close to srgb, so if you have an
output profile for your printer (icc), this should be easy. There is also
macbethcal which could be used to make a simple output calibration, however,
I never used it (I tried once).

I can't imagine that the bit per channel question is really relevant (for
inkjet printing), because the limitations of paper and the raster technique
used to mix the colors.

I will have to do some large-format printouts of radiance pictures (which
have to be rendered first :wink: for an exhibition next march, as such I am
interested in the topic.

CU Lars.

5

Hi!

> In general you need to print to a print device that goes direct rgb.

??? You mean a driver that can do the rgb to cmyk translation, without the
need to do that in an application? Just wondering, I never heard about
rgb-printers, do you mean transfering to a film (which is like foto and as
such rgb)?

Regarding printers, it is a big help to use printers that don't offer just
cmyk, but also the light colors (there are 6 and 8 color printers), as this
avoids raster problems in bright regions of your renderings.

This is where the Lightjet family of printers is different. They
upsample the image to ~4000dpi and expose photographic paper with
lasers. I suppose they use red, green, and blue lasers, but I
have no documented proof. I have been very happy with the output.
If you managed to make it to SIGGRAPH in the last 3 years, you
may have seen these samples of Radiance-rendered work, in
large-format, hanging in the Art Gallery.

Calibration... afaik the radiance output is close to srgb, so if you have an
output profile for your printer (icc), this should be easy. There is also
macbethcal which could be used to make a simple output calibration, however,
I never used it (I tried once).

I find that when I call a professional photo/printing business
and start talking technical, I get forwarded to the specialist,
who takes something of an interest once I mention that I have no
ICC profiles, and my work isn't from a Mac or a PC. I've learned
that (if I set my monitor's gamma to ~2.2) what I see on my
monitor looks very much like what gets printed when I dictate
"assign no profile" or "assign sRGB." Maybe the sRGB one looks a
little more "correct," or a little less saturated. I have a hard
time telling the difference.

I can't imagine that the bit per channel question is really relevant (for
inkjet printing), because the limitations of paper and the raster technique
used to mix the colors.

I will have to do some large-format printouts of radiance pictures (which
have to be rendered first :wink: for an exhibition next march, as such I am
interested in the topic.

Another bit of advice: learn about the printer that you're using,
and send it the best possible pixels that you can. Insist that
they print at that resolution. Most printers are used to folks
printing enlargements of bad photos. It helps to talk to the
person who will actually run the machine and explain that there
is minute detail that is important. For the Lightjet 5000, I
always send a 300 dpi file, and I always antialias by pfilt'ing
my final rendering down at least 3:1 (sometimes more).

I hope my experiences have helped. It has taken me quite a few
years to learn how to "talk the talk" and get a good image
printed.

Mark

···

On Wed, 1 Dec 2004, Lars Grobe wrote:

6

Hi Lars,

Yes, as I understand these photographic printers (eg direct to photo paper), such as the Lambda and Pegasus, print RGB not CMYK. The nice thing about this is there is no mucking about with conversion from one colorspace to the other. They use red, green and blue lasers to expose the paper. Do a web search for Durst Lambda to learn more or check out:

http://www.durst-online.com/uk/produkte.asp?pid=2&hid=

I think that the kodak product may be different, perhaps some kind of LED to expose the paper.

-Jack

Lars Grobe wrote:

···

Hi!

In general you need to print to a print device that goes direct rgb.
   
??? You mean a driver that can do the rgb to cmyk translation, without the
need to do that in an application? Just wondering, I never heard about
rgb-printers, do you mean transfering to a film (which is like foto and as
such rgb)?

Regarding printers, it is a big help to use printers that don't offer just
cmyk, but also the light colors (there are 6 and 8 color printers), as this
avoids raster problems in bright regions of your renderings.

Calibration... afaik the radiance output is close to srgb, so if you have an
output profile for your printer (icc), this should be easy. There is also
macbethcal which could be used to make a simple output calibration, however,
I never used it (I tried once).

I can't imagine that the bit per channel question is really relevant (for
inkjet printing), because the limitations of paper and the raster technique
used to mix the colors.

I will have to do some large-format printouts of radiance pictures (which
have to be rendered first :wink: for an exhibition next march, as such I am
interested in the topic.

CU Lars.

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
# John E. de Valpine
# president
#
# visarc incorporated
# http://www.visarc.com
#
# channeling technology for superior design and construction

7

Hi,

cmyk is probably a bit out of date nowadays, it always was a rather poor standard suitable for those big printing machines of former time. As already mentioned, the best way today is to use those new fancy laser printers which use rgb directly without color space conversion.

Apart from color fidelity, another question which might come into play is to apply a tonemapping process on the HDR image before converting it to rgb. The old Radiance had the pquish script, the newer versions have the 'pcomp' program for that purpose. This applys a partially nonlinear compression to preserve detail in the highlight/dark regions which otherwise would be over/underexposed, in other words it allows to compress the image to a contrast range fitting that of the output machine (which definitely is the final limiting factor in the whole chain)

-cb

8

Carsten Bauer wrote:

Hi,

cmyk is probably a bit out of date nowadays, it always was a rather poor standard suitable for those big printing machines of former time. As already mentioned, the best way today is to use those new fancy laser printers which use rgb directly without color space conversion.

not really my field of expertise, but -
probably one should measure the spectral reflectance of each printed primary color to come with a good color calibration ? Or is that purely academic ?
I had found it rather difficult to get a feel what the a laser printer does internally, what happens between the RGB values in the PS or PDF file and the beam intensity. Has the user enough control over that process to justify a calibration ? Anyone got (negative ?) experiences to reproduce colors and shades between self-calibrations of a printer ?

-Peter

···

--
pab-opto, Freiburg, Germany, http://www.pab-opto.de
[see web page to check digital email signature]

9

Carsten Bauer wrote:

Apart from color fidelity, another question which might come into play is to apply a tonemapping process on the HDR image before converting it to rgb. The old Radiance had the pquish script, the newer versions have the 'pcomp' program for that purpose.

I think you mean 'pcond'. And yes, the tonemapping that pcond offers is very useful for generating "more printable" images. I use it all the time. Stuff like pcond -h is what separates Radiance from the rest!

···

-------------------
    Rob Guglielmetti
www.rumblestrip.org

10

Peter Apian-Bennewitz wrote:

Carsten Bauer wrote:

cmyk is probably a bit out of date nowadays, it always was a rather
poor standard suitable for those big printing machines of former time.

The pigments on a print will almost always be cymk (sometimes
augmented with additional colors). This has nothing to do with
"big machines of former time", but simply with the laws of
physics.

As already mentioned, the best way today is to use those new fancy
laser printers which use rgb directly without color space conversion.

The only difference is that *you* don't need the printer's color
profile for doing the conversion yourself. The conversion module
of the printer has that built in. To think that there will be no
conversion at all (not sure if you really do think so) would be
naive. This is even the case for machines that directly expose
photographic paper with rgb lasers (which of course will activate
cym[k] pigments in the paper, introducing yet another "conversion").

not really my field of expertise, but -
probably one should measure the spectral reflectance of each printed
primary color to come with a good color calibration ? Or is that purely
academic ?

If you're really picky, then that would be the most accurate
method to generate the color profile for a printer. In practise,
visual comparisons against a reference print may be good enough
in most cases.

I had found it rather difficult to get a feel what the a laser printer
does internally, what happens between the RGB values in the PS or PDF
file and the beam intensity.

That's what color profiles are there for. In the times of color
profiles, the mechanics of how that reflectance gets on paper
should be irrelevant to the user.

All of this of course assumes that you know the color space of
your own data. It seems that the Radiance color space is quite
close to SRGB, which is highly convenient. Once you know that,
you either use the color profile of your printer to do a
(SRGB->"printer rgb" or SRGB->"Printer cymk") conversion
yourself, or you let the printer know that it will get SRGB data
and have it do the same conversion internally. If you get bad
results that way, then either the color profile is wrong or the
printer is broken.

At least that's how I understand the theory... :wink:

-schorsch

···

--
Georg Mischler -- simulations developer -- schorsch at schorsch com
+schorsch.com+ -- lighting design tools -- http://www.schorsch.com/

11

Wow, thanks for all your feedback!

Seems like the lightjet is an interesting machine.
Generally I'm on par with Jack, I share his view on quality / cost effective
printing.
I had to think about the idea of colout fidelity seeing the art work of Yves
Klein.
Its interesting to see art which main focus is pure colour.
The interesting thing here is that with Radiance, it really isn't difficult
at all to produce this pure kind of colour, your just not being able to get
it out of your machine. I've been printing with Durst Lambda on light boxes,
which is quite an excellent technique, but in the light of pure pigment
colours, everything just looks faint. Hence the quest for 32 bit floating
point output. Looking at the VIPS software
(Supports a large range of numeric types Images can contain 8-, 16- and
32-bit integers, signed or unsigned, 32- and 64-bit floats, and 64- and
128-bit complex numbers, http://www.vips.ecs.soton.ac.uk/vips.php) it seems
that this issue addressed here is relevant.
What good is all that data if your not able to get its full output. Actually
I tried to be in touch with the developers at Durst, to give them the
stunning idea of making a floating point Lamba driver for free, alas, no
response. Looks like this is a dead end so far. Thanks for all your
responses!

Jelle.

12

Greg,

Thanks for the valuable feedback!

Concerning the limited dynamic range of paper-like based prints, we've moved
to using transparencies and lightboxes for our output. We've been producing
this output with Durst Lambda printers, which use the same kind of colour
interpolation as the Lightjet printer.
Deep black is difficult to produce in this manner, since the material is
indeed quite reflective, and the backlit nature of lightboxes inherently
doenst do justice to pure black. The most beautifully printed black I've
seen so far was done with Gyclee printers, using up to 11 'ultra wide
gamuth' inks. These prints were done on matte paper, what's beautiful about
it is that the black is almost like powder, the detail and depth in the
black is nothing but astonishing. (see http://www.colorspan.com)

The idea of working with a Hollywood based company is an option in the sense
that its sure they have the technology to do it, but the size of ouput is
for the film industry is rather different than what I'm looking for (35mm
for ordinary film, 70mm for IMAX). Still I'll see if this could be an option
we could pursue.

"Ultimately, I think HDR displays are the best way to appreciate
lighting simulations. Until then, we're kind of stuck with low
dynamic-range, tone-mapped prints and displays."

I couldn't agree more. Somehow I have the feeling hacking a 32 bit colour
Durst lambda driver could be an option. Since I'm not a programmer/physicist
I'm not sure if this suggestion is valid, but what if one could produce
multiple exposed lambda prints? Could building up the exposure increase the
depth of the printed output.

--- I realize discussing the particulars of a machine for output is somewhat
off-topic for this list... ---

Sorry for my uninformed openEXR bashing... consider it an over enthousiastic
expression for my affection of Radiance :wink:
Thanks for the link/paper considering this topic, it's a wealth of
information to get to the bottom of this! Looking forward for this book!

Cheers,

Jelle.

13

Georg Mischler wrote:

Peter Apian-Bennewitz wrote:

Carsten Bauer wrote:

cmyk is probably a bit out of date nowadays, it always was a rather
poor standard suitable for those big printing machines of former time.

The pigments on a print will almost always be cymk (sometimes
augmented with additional colors). This has nothing to do with
"big machines of former time", but simply with the laws of
physics.

hmm, if it were for the physics alone, we wouldn't need the k, so we'll have to take chemistry and engineering (and of course physiology) into account, also. In most high-quality printing jobs (e.g. art catalogues) additional pigments are used to overcome the cmyk (saturation)-limitations.

But instead of quarreling about this I simply wanted to say that the new RGB laser printers usually do a very good job with their automatic operation, at least for images without strong color bias.
I admit that I haven't bothered so far about what's going on inside the machine, they're very expensive, so I believe you that some complicated stuff is working inside.)

I also think it's out of question that the contrast range and color saturation of machine made prints cannot compare to those of transparencies or oil paintings. But the color resolution of those laser prints is fairly high so you can get a lot of information from the image onto the print also. I noticed this once in a negative sense, when I hoped that some faint splotches which were hardly visible on the screen would finally disappear on the print due to limited color space resolution - they didn't, and remained nicely visible.. :frowning:

ahmm, and yes, Rob, I meant of course pcond..

-cb

14

One follow-up on this. Among the reasons that we were so happy with the Lambdas and Pegasus (and I think also Ofoto, but I have much fewer data points) is that we can process the images as we would for viewing normally on our displays and then send them to print as is (with tif conversion) without further mucking about.

-Jack

Carsten Bauer wrote:

···

Georg Mischler wrote:

Peter Apian-Bennewitz wrote:

Carsten Bauer wrote:

cmyk is probably a bit out of date nowadays, it always was a rather
poor standard suitable for those big printing machines of former time.

The pigments on a print will almost always be cymk (sometimes
augmented with additional colors). This has nothing to do with
"big machines of former time", but simply with the laws of
physics.

hmm, if it were for the physics alone, we wouldn't need the k, so we'll have to take chemistry and engineering (and of course physiology) into account, also. In most high-quality printing jobs (e.g. art catalogues) additional pigments are used to overcome the cmyk (saturation)-limitations.

But instead of quarreling about this I simply wanted to say that the new RGB laser printers usually do a very good job with their automatic operation, at least for images without strong color bias.
I admit that I haven't bothered so far about what's going on inside the machine, they're very expensive, so I believe you that some complicated stuff is working inside.)

I also think it's out of question that the contrast range and color saturation of machine made prints cannot compare to those of transparencies or oil paintings. But the color resolution of those laser prints is fairly high so you can get a lot of information from the image onto the print also. I noticed this once in a negative sense, when I hoped that some faint splotches which were hardly visible on the screen would finally disappear on the print due to limited color space resolution - they didn't, and remained nicely visible.. :frowning:

ahmm, and yes, Rob, I meant of course pcond..

-cb

15

Jelle Feringa // EZCT / Paris wrote:

....

The idea of working with a Hollywood based company is an option in the sense
that its sure they have the technology to do it, but the size of ouput is
for the film industry is rather different than what I'm looking for (35mm
for ordinary film, 70mm for IMAX). Still I'll see if this could be an option
we could pursue.

you might want to check
    http://www.arri.de/prod/digital/arrilaser/index.php
which apparently is used in Hollywood too.

···

--
pab-opto, Freiburg, Germany, http://www.pab-opto.de
[see web page to check digital email signature]

16

> not really my field of expertise, but -
> probably one should measure the spectral reflectance of each printed
> primary color to come with a good color calibration ? Or is that purely
> academic ?

If you're really picky, then that would be the most accurate
method to generate the color profile for a printer. In practise,
visual comparisons against a reference print may be good enough
in most cases.

By the way, if you have the hardware, generating a profile (icc) is not too
difficult, and you can re-use it as often as wanted...

Lars.