QTR-Quadtone RIP

Is it necessary to have steps of 5% or 2% in the targets to let them 
function in the QTR linearisation and profiling tools?  In other words; 
would a 33 or 34 step target with 3% increase and one 4% or 1% step 
compromise the algorithms of the tools?

-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

On 06/04/2012 03:45 PM, Ernst Dinkla wrote:
> Is it necessary to have steps of 5% or 2% in the targets to let them
> function in the QTR linearisation and profiling tools? In other words;
> would a 33 or 34 step target with 3% increase and one 4% or 1% step
> compromise the algorithms of the tools?

No reply yet. Could be a stupid question and do not hesitate to rub that 
in then, I have written stupid questions before.

Let me explain why I wrote that question. I am looking for a compromise 
between the 21 and 51 step greyscale targets. The last can create issues 
when either the basic tone range of the printed target is too irregular 
and/or the density measurements are not increasing properly. The 21 step 
does it well on those aspects but if I create RGB curves with Photoshop 
etc the curve control points are close to the 21 number and I think that 
the linearity etc could be improved by a curve based on a higher, odd 
target number to take out interference between the two data sets.
To create an accurate, regular 8-bit target there is no other sensible 
choice than identical steps of 20+1 or 50+1 so 5%K or 2%K steps. I can 
make a 16-bit (15-bit+1 actually) target with 35+1 steps and am already 
halfway that work but if there is a bottleneck on 16 bit in either Color 
Port, Measure Tool, QTR-Create-ICC-RGB tool's linearisation, the 
application color management in printing, then that work is idle. 
Whether the driver in the end delivers 8-bit data to the printer is of 
less concern to me but up to that point I would like to have 16-bit 
processing. 16-bit driver would be nice too but is not available in all 
cases, some rounding off at that stage does not worry me.


-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

I don't know the answer for certain, but my gut feeling is that your idea will work.

I've played around a bunch with the QTR tools, and written some code that does similar functions. Fundamentally, the tools are generating curves from pairs of data: an input (an expected value) and an output (a measured value). You can look at this as a series of X,Y points. QTR probably uses some form of spline algorithm (see <http://en.wikipedia.org/wiki/Spline_(mathematics)>) to interpolate the full curves using those points. Different algorithms have different requirements: some need a certain number of points, some need to have the points spaced at equidistant intervals. I don't know which algorithm QTR uses.

Given that the QTR tools don't seem to need any configuration to indicate the particular patches (they're just read from the CGATS file, as you know), I'd reckon that there's no assumption being made as to the specific intervals between patches. Hence, having arbitrary patches as you're describing will probably work.

--John

On 7 Jun 2012, at 3:36 AM, Ernst Dinkla wrote:

> On 06/04/2012 03:45 PM, Ernst Dinkla wrote:
>> Is it necessary to have steps of 5% or 2% in the targets to let them
>> function in the QTR linearisation and profiling tools? In other words;
>> would a 33 or 34 step target with 3% increase and one 4% or 1% step
>> compromise the algorithms of the tools?
> 
> No reply yet. Could be a stupid question and do not hesitate to rub that 
> in then, I have written stupid questions before.
> 
> Let me explain why I wrote that question. I am looking for a compromise 
> between the 21 and 51 step greyscale targets. The last can create issues 
> when either the basic tone range of the printed target is too irregular 
> and/or the density measurements are not increasing properly. The 21 step 
> does it well on those aspects but if I create RGB curves with Photoshop 
> etc the curve control points are close to the 21 number and I think that 
> the linearity etc could be improved by a curve based on a higher, odd 
> target number to take out interference between the two data sets.
> To create an accurate, regular 8-bit target there is no other sensible 
> choice than identical steps of 20+1 or 50+1 so 5%K or 2%K steps. I can 
> make a 16-bit (15-bit+1 actually) target with 35+1 steps and am already 
> halfway that work but if there is a bottleneck on 16 bit in either Color 
> Port, Measure Tool, QTR-Create-ICC-RGB tool's linearisation, the 
> application color management in printing, then that work is idle. 
> Whether the driver in the end delivers 8-bit data to the printer is of 
> less concern to me but up to that point I would like to have 16-bit 
> processing. 16-bit driver would be nice too but is not available in all 
> cases, some rounding off at that stage does not worry me.
> 
> 
> -- 
> Met vriendelijke groet, Ernst
> 
> Dinkla Grafische Techniek
> Quad, piëzografie, giclée
> www.pigment-print.com
> 
> 
> 
> ------------------------------------
> 
> Yahoo! Groups Links
> 
> 
>

On 06/07/2012 07:47 PM, John Labovitz wrote:

> Given that the QTR tools don't seem to need any configuration to
> indicate the particular patches (they're just read from the CGATS file,
> as you know), I'd reckon that there's no assumption being made as to the
> specific intervals between patches. Hence, having arbitrary patches as
> you're describing will probably work.
>
> --John

John,

Thank you, gives me confidence enough to go that much easier route.


-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

I'm surprised Roy hasn't responded to this.

To my understanding, the linearization in QTR assumes equally-spaced points. i.e. if you entered 5 points, it would assume 0,25,50,75,100. 11 points would be 0, 10, 20...100, etc. You could do 4% intervals, wouldn't that be a 26-point linearization?

33 or 34 linearization points would very slightly perturb the linearization curve, you could run some evaluations to determine where that perturbation would have the least effect.


--Greg

--- In QuadtoneRIP@yahoogroups.com, Ernst Dinkla <e.dinkla@...> wrote:

>
> On 06/07/2012 07:47 PM, John Labovitz wrote:
> 
> > Given that the QTR tools don't seem to need any configuration to
> > indicate the particular patches (they're just read from the CGATS file,
> > as you know), I'd reckon that there's no assumption being made as to the
> > specific intervals between patches. Hence, having arbitrary patches as
> > you're describing will probably work.
> >
> > --John
> 
> John,
> 
> Thank you, gives me confidence enough to go that much easier route.
> 
> 
> -- 
> Met vriendelijke groet, Ernst
> 
> Dinkla Grafische Techniek
> Quad, piëzografie, giclée
> www.pigment-print.com
>

Hi Ernst,

I've been out of town and touch for a while so I'm just getting to things.

Like Greg said, the linearization numbers are always assumed evenly spaced
for however number there are.  The tricky part comes when you make the
targets.
With 8 bits (256 values) files and K in percents (100 values) roundoff
errors and
uneven steps are likely.  16 bit files are the best solution but even
that is tricky
because you can't enter 16 values.  A posterized 16 bit gradient will
give correct
values, or starting with 33 even 8 bits values (0-160) and then
stretching it out
in 16 bit mode.  Once you have a good evenly spaced 16 bit file its OK
to convert
to 8 bit - the spacing will still be good.

Finally there's the reading of the target.  I think you get a lot of
benefit from averaging
multiple patches (more important than more steps).  MeasureTool with a reference
file works great for this because the ref file tells all the gray
values.  (Note: use
decimal values for gray if needed -- e.g. 0 3.33 6.67 10 13.33 ...).
Without a ref file
QTR makes a best guess based on how many patches total there are -- works
well for most cases but you may be able to stump it.

Roy

On Thu, Jun 7, 2012 at 11:21 AM, Ernst Dinkla
<e.dinkla@...> wrote:
> On 06/07/2012 07:47 PM, John Labovitz wrote:
>
>> Given that the QTR tools don't seem to need any configuration to
>> indicate the particular patches (they're just read from the CGATS file,
>> as you know), I'd reckon that there's no assumption being made as to the
>> specific intervals between patches. Hence, having arbitrary patches as
>> you're describing will probably work.
>>
>> --John
>
> John,
>
> Thank you, gives me confidence enough to go that much easier route.
>
>
> --
> Met vriendelijke groet, Ernst
>
> Dinkla Grafische Techniek
> Quad, piëzografie, giclée
> www.pigment-print.com
>
>
>
> ------------------------------------
>
> Yahoo! Groups Links
>
>
>



-- 
Roy Harrington
roy@...
www.harrington.com

On 12-06-07 10:47 PM, "Roy Harrington" <roy@...> wrote:

> Finally there's the reading of the target.  I think you get a lot of
> benefit from averaging
> multiple patches (more important than more steps).

Roy, are you talking about taking readings from multiple prints of the
target, or about reading a single target print multiple times (or both)?

Terry.

On 06/12/2012 04:49 AM, Terry Ritz wrote:
> On 12-06-07 10:47 PM, "Roy Harrington" <roy@...
> <mailto:roy%40harrington.com>> wrote:
>
>  > Finally there's the reading of the target. I think you get a lot of
>  > benefit from averaging
>  > multiple patches (more important than more steps).
>
> Roy, are you talking about taking readings from multiple prints of the
> target, or about reading a single target print multiple times (or both)?
>
> Terry.

There could be two reasons for multiple measurements of the same 
density. The printer can be less consistent so you print more of the 
same patches to measure. The measuring can be inconsistent and measuring 
the same densities several times + averaging reduces the faults. The 
21x4 target covers both. If both printer and measuring are consistent 
(for example measuring per patch) then there will be little gain by 
printing more patches and measuring them more times. With my old Epson 
9000 custom quad printer + Spectrocam I used to measure strips of 
targets that were placed in different directions on the sheet.

Today with the Z3200 or the customised K5400 + an Eye 1 Pro, I use one 
21 step target and reading per patch. The Z3200 calibration target has 
16+1 patches for each channel and is measured automatically patch for 
patch by the Z3200 integrated spectrometer, the black patches etc are 
measured longer to give a better measurement. The QTR linearisation I 
put on top is based on the 21 step target but I think that the 34 step 
linearisation may improve on that. The same for the K5400 quad or 
tritone (gloss) that has ACV curves for partitioning.

Finished the 16-bit 34 step target some days ago, will put it into 
practice at the end of the week I hope. I will then add it to the 
targets for download on my site. Thank you Roy for the instructions, I 
used another 16-bit route first that was accurate too but this one is 
faster.


-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

On 06/08/2012 06:47 AM, Roy Harrington wrote:

> Like Greg said, the linearization numbers are always assumed evenly spaced
> for however number there are. The tricky part comes when you make the
> targets.

Thank you for the reply, I have made a 34 step target.

Something related: I use ACV curves to make grey ink partitions \ufffd la 
Paul Roark for a HP K5400 office inkjet loaded with grey inks. In 
Photoshop of course and print the trials directly from Photoshop through 
the HP driver. When finished I embed the curves in an RGB profile with 
the QTR tool for that. That works when there is also a linearisation 
curve added at the same time. But I rather like to add the linearisation 
curve later on as I normally print from Qimage and think I should create 
the linearisation curve based on printing from Qimage with the 
unlinearised RGB profile and then linearise that profile with a target 
output from Qimage. So I think I should embed an ineffective 
"linearisation" curve instead first and wonder how simple the data file 
for that could be. ACV curves do not work in Qimage.


-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

On 12-06-12 2:43 AM, "Ernst Dinkla" <e.dinkla@...> wrote:

> There could be two reasons for multiple measurements of the same
> density. The printer can be less consistent so you print more of the
> same patches to measure. The measuring can be inconsistent and measuring
> the same densities several times + averaging reduces the faults. The
> 21x4 target covers both. If both printer and measuring are consistent
> (for example measuring per patch) then there will be little gain by
> printing more patches and measuring them more times. With my old Epson
> 9000 custom quad printer + Spectrocam I used to measure strips of
> targets that were placed in different directions on the sheet.
> 
> Today with the Z3200 or the customised K5400 + an Eye 1 Pro, I use one
> 21 step target and reading per patch.

What I'm wondering is whether multiple prints, or multiple readings, would
net the largest benefit. The answer of course is, "it depends"!  ;-)

I obviously need to do some testing to determine whether my printer and/or
spectro are weaker links in the curve building chain.

> The QTR linearisation I
> put on top is based on the 21 step target but I think that the 34 step
> linearisation may improve on that. The same for the K5400 quad or
> tritone (gloss) that has ACV curves for partitioning.
> 
> Finished the 16-bit 34 step target some days ago, will put it into
> practice at the end of the week I hope. I will then add it to the
> targets for download on my site.

I am very curious to hear about how your testing with the 34 step target
goes. I am hopeful that it will help to minimize the "bumps" that can occur
during linearization.

Thanks Ernst.

Terry.

> Something related: I use ACV curves to make grey ink partitions \ufffd la
> Paul Roark for a HP K5400 office inkjet loaded with grey inks. In
> Photoshop of course and print the trials directly from Photoshop through
> the HP driver. When finished I embed the curves in an RGB profile with
> the QTR tool for that. That works when there is also a linearisation
> curve added at the same time. But I rather like to add the linearisation
> curve later on as I normally print from Qimage and think I should create
> the linearisation curve based on printing from Qimage with the
> unlinearised RGB profile and then linearise that profile with a target
> output from Qimage. So I think I should embed an ineffective
> "linearisation" curve instead first and wonder how simple the data file
> for that could be. ACV curves do not work in Qimage.

In addition: are the readable target measurement numbers at the top in 
the created ICC file actually used when the profile is active in the 
printing workflow or are they just there for documentation? In other 
words can I change them there for the reason described above or will 
there not be a difference in printing after the change and the profile 
has unreadable data that actually does the job?


-- 
Met vriendelijke groet, Ernst

Dinkla Grafische Techniek
Quad, pi\ufffdzografie, gicl\ufffde
www.pigment-print.com

Hi Ernst-

Have you finished this target? I looked for it on your site, but
couldn't find it.

-- 
Regards,

Bill Morse
Wm. Morse Editions

http://www.MorseEditions.com/

On 12-06-12 2:43 AM, "Ernst Dinkla" <e.dinkla@...> wrote:

> The QTR linearisation I
> put on top is based on the 21 step target but I think that the 34 step
> linearisation may improve on that. The same for the K5400 quad or
> tritone (gloss) that has ACV curves for partitioning.
>
> Finished the 16-bit 34 step target some days ago, will put it into
> practice at the end of the week I hope. I will then add it to the
> targets for download on my site.