QTR-Quadtone RIP

When using Linearize-Data figures in the Curve Creator table, the 
curves get "re-shaped" so that the output gets more linear.
Yet at times, the linearity remains debatable.

Clearing the Linearization Table after, and inputting the readings of 
the already-once linearized stepwedge should work as an iterative 
linearization since the curve are built with the first linearization 
and the second linearization should refine the output. Right or wrong ?

I know the Create-Icc should deliver this but I'm trying to really get 
the best possible linearization before the profile building, and this 
can take 2 or 3 iterations in my opinion.

Olivier

Olivier wrote:
> When using Linearize-Data figures in the Curve Creator table, the 
> curves get "re-shaped" so that the output gets more linear.
> Yet at times, the linearity remains debatable.
> 
> Clearing the Linearization Table after, and inputting the readings of 
> the already-once linearized stepwedge should work as an iterative 
> linearization since the curve are built with the first linearization 
> and the second linearization should refine the output. Right or wrong ?
> 
I'm pretty sure that's wrong. The curves won't "remember" the old 
linearisation - they will just be remade with the new one as if that was 
the original.

I've always found the linearisation process to be uncannily accurate. I 
don't know what calculations are being done but it works very well. It 
helps to have reasonably accurate curve set up in the first place. I 
wouldn't do the linearisation until I had got all the stepwedge points 
within about 10% of where they should be.

Howard

That's my understanding too. Linearization is a one shot deal.  If 
you try linearizing again over a previous linearization, your new lin 
data will overwrite the old lin data, and you will not have a good 
linearized profile.  I also agree that QTR does an excellent job of 
linearization and separation of tones.  

Remember, QTR thinks and works in Lab, so  gamma or dot gain based 
data will look different.  If you open a 21 step grayscale in 
Photoshop (for example in gamma 2.2 grayscale mode) and set your eye 
droppers to read the actual pixels on the left and Lab on the right, 
you will see the Lab equivalents of the K values.  Evenly spaced 
grayscale steps don't equate to the same spacing in Lab.  For 
example, 100K is equal to 0L, but 95K is not 5L, it is 1L.  So, in 
Lab terms, there isn't much separation between these two steps.  If 
that is what you are seeing, then it may be that you are working in a 
gamma or dotgain based working space.  Try working in Roy's Lab 
Grayscale space and see if that makes a difference.

Lou


--- In QuadtoneRIP@yahoogroups.com, Howard Shaw <glassman@...> wrote:
>
> Olivier wrote:
> > When using Linearize-Data figures in the Curve Creator table, the 
> > curves get "re-shaped" so that the output gets more linear.
> > Yet at times, the linearity remains debatable.
> > 
> > Clearing the Linearization Table after, and inputting the 
readings of 
> > the already-once linearized stepwedge should work as an iterative 
> > linearization since the curve are built with the first 
linearization 
> > and the second linearization should refine the output. Right or 
wrong ?
> > 
> I'm pretty sure that's wrong. The curves won't "remember" the old 
> linearisation - they will just be remade with the new one as if 
that was 
> the original.
> 
> I've always found the linearisation process to be uncannily 
accurate. I 
> don't know what calculations are being done but it works very well. 
It 
> helps to have reasonably accurate curve set up in the first place. 
I 
> wouldn't do the linearisation until I had got all the stepwedge 
points 

> within about 10% of where they should be.
> 
> Howard
>

Louis Dina wrote:
> That's my understanding too. Linearization is a one shot deal.  If 
> you try linearizing again over a previous linearization, your new lin 
> data will overwrite the old lin data, and you will not have a good 
> linearized profile.  I also agree that QTR does an excellent job of 
> linearization and separation of tones.  

Iterative measurements can be an interesting option if (and 
it is a big 3 fold IF) the measuring instrument, the 
measurements and the printed targets are much more 
consistent than the linearisation fault you want to correct. 
There's more to gain in my opinion with multiple readings of 
more random linearisation targets that are printed at two 
angles to the print direction and building the linearisation 
on the average of the measurements.

For color profile creation the iterative build-up has been 
discussed on the colorsync list maybe two years ago and 
while there's color profiling software that allows iterative 
measurements the parties were divided on the pros of it. 
Considering the more complex shape of a profile and its 
function to represent the printer's color space the 
iterative function in the software could improve the shape. 
The same IFs apply there too. But if that can be achieved 
and when used on good regular calibration routines of 
printer + paper batches to keep the underlying base 
consistent it is in theory not a bad concept. The 
calibration done as suggested with multiple readings etc.

I use two terms: calibration and linearisation.
The first doesn't have to be a linearisation but sets the 
printer to fixed output numbers based on target readings, 
including the Dmax. So as I understand it, if that printer 
delivers more Dmax than the factory prescribes the inklimit 
at 100% will also be set lower for that channel. The 
uncalibrated channel could give a higher inkload (which 
isn't always a higher Dmax) than the calibrated channel.
Linearisation is done between paper white and the Dmax 
measured and as the first and last could vary the steps in 
between could vary. A profile made earlier on the first 
linearisation may not represent the printer after a new 
linearisation. The linearised and unlinearised channel 
should have the same Dmax.
What manufacturers did implant in the functions and how they 
describe them is not always clear, hybrids of the two are 
possible too and do not have to be bad either.

For sophisticated color profile creation with an iterative 
build up you better start with a calibration to squeeze out 
the last inconsistency in the base before the sculpturing 
and polishing happens on top. I doubt that this all is 
needed for B&W printing with QTR.

Again, there's more to gain in my opinion with multiple 
readings of more random linearisation targets that are 
printed at two angles to the print direction and building 
the linearisation on the average of the measurements. Same 
for the QTR profile.


-- 
Met vriendelijke groeten, Ernst


|  Dinkla Grafische Techniek  |
|     www.pigment-print.com    |
|             ( unvollendet )            |

Ernst Dinkla wrote:

> Again, there's more to gain in my opinion with multiple 
> readings of more random linearisation targets that are 
> printed at two angles to the print direction and building 
> the linearisation on the average of the measurements. Same 
> for the QTR profile.
> 

Ernst

Could you summarise what the advantages of random targets are over 
ordered ones? They are rather inconvenient for those of us with more 
primitive devices.

Also could you please explain what you mean by printing the target at 
two angles. Do you mean in portrait & in landscape modes? If so why 
would that be necessary?

thanks & regards
Howard

--- In QuadtoneRIP@yahoogroups.com, Ernst Dinkla <E.Dinkla@...> wrote:
...
> Again, there's more to gain in my opinion with multiple 
> readings of more random linearisation targets that are 
> printed at two angles to the print direction and building 
> the linearisation on the average of the measurements. Same 
> for the QTR profile.

Ernst, I have no experience on this matter specifically, but thought
you might find this interesting. When some of us met Bill Atkinson, he
had done exhaustive tests with regard to measurement. His findings
were that series of readings with the least possible difference from
one patch to the next gave consistently more accurate results than
random patch strips. So he ordered his charts accordingly.
Of course this was for color profiling, and large charts, and I
suspect the differences were small, but there you are for what it's
worth...
Tyler

Howard Shaw wrote:
> Ernst Dinkla wrote:
> 
>> Again, there's more to gain in my opinion with multiple 
>> readings of more random linearisation targets that are 
>> printed at two angles to the print direction and building 
>> the linearisation on the average of the measurements. Same 
>> for the QTR profile.
>>
> 
> Ernst
> 
> Could you summarise what the advantages of random targets are over 
> ordered ones? They are rather inconvenient for those of us with more 
> primitive devices.
> 
> Also could you please explain what you mean by printing the target at 
> two angles. Do you mean in portrait & in landscape modes? If so why 
> would that be necessary?
> 
> thanks & regards
> Howard

In row reading spectrometers and densitometers can be 
influenced by a light patch before reading a dark patch, the 
more contrast the more likely to happen, sensor memory. So 
reading greyscale step wedges going upwards in density or 
downwards should give different readings as well. On the 
Epson 9000 quad I also observed that I got a more equal 
output if the head had been busy before printing the target 
than if it just had to start with the target itself. Next to 
that comes banding (can be almost invisible) in the inkjet 
printed target and a dominant grain direction in the paper. 
The print will lay down its ink different and/or the 
spectrometer gets another reflection from the surface. For 
example (textured) art papers made on the foudrinier, or 
worse, canvas. There is also the inconsistency of the manual 
measurements in row reading. So printing multiple random 
targets at different angles to the paper and print direction 
isn't a bad idea. One could arrange them all at 45 degrees 
but on the 9000 I had two in one direction and one in the 
other covering the width of a roll together. Faster in 
printing, the roll could stay on the printer, less paper 
waste and more related to the printing done in practice, 
drying between passes etc. I have an old helmet type hair 
dryer on a box to speed up drying of the targets.

Whether it is all necessary depends on how good the 
measuring method is, the quality of equipment and how anal 
one is :-)

The HP Z3100 that I have now prints the calibration target 
(16 patches per channel, 11-12 channels) in the paper 
transport direction. Steps are not random but go up and down 
in that direction. Quite big patches. After drying time it 
will measure per patch in the head direction, measurements 
on the dark patches are done longer than the light ones. 
With profiles one can either let the target dry on the 
printer or take it out and do the measurements later on. The 
Z3100 ink settles faster so it works well on the printer.

-- 
Met vriendelijke groeten, Ernst


|  Dinkla Grafische Techniek  |
|     www.pigment-print.com    |
|             ( unvollendet )            |

Tyler Boley wrote:
> --- In QuadtoneRIP@yahoogroups.com, Ernst Dinkla <E.Dinkla@...> wrote:
> ...
>> Again, there's more to gain in my opinion with multiple 
>> readings of more random linearisation targets that are 
>> printed at two angles to the print direction and building 
>> the linearisation on the average of the measurements. Same 
>> for the QTR profile.
> 
> Ernst, I have no experience on this matter specifically, but thought
> you might find this interesting. When some of us met Bill Atkinson, he
> had done exhaustive tests with regard to measurement. His findings
> were that series of readings with the least possible difference from
> one patch to the next gave consistently more accurate results than
> random patch strips. So he ordered his charts accordingly.
> Of course this was for color profiling, and large charts, and I
> suspect the differences were small, but there you are for what it's
> worth...
> Tyler

Tyler,

I'm sure that David has nice automatic equipment that may 
read the targets patch wise and he started making his 
excellent profiles for Epson 9600's with some support from 
Epson (read: well calibrated printers). Not linearising with 
a  manual Spectrocam the customised Epson 9000 made for dye 
inks loaded with third party quad inks :-)

-- 
Met vriendelijke groeten, Ernst


|  Dinkla Grafische Techniek  |
|     www.pigment-print.com    |
|             ( unvollendet )            |

Ernst Dinkla wrote:

> I'm sure that David has nice automatic equipment that may 
> read the targets patch wise and he started making his 
> excellent profiles for Epson 9600's with some support from 
> Epson (read: well calibrated printers). Not linearising with 
> a  manual Spectrocam the customised Epson 9000 made for dye 
> inks loaded with third party quad inks :-)

Busy day yesterday. Bill its should be.



-- 
Met vriendelijke groeten,Ernst


|  Dinkla Grafische Techniek  |
|     www.pigment-print.com    |
|             ( unvollendet )            |

> > I'm sure that David has nice automatic equipment that may 
> > read the targets patch wise and he started making his 
> > excellent profiles for Epson 9600's with some support from 
> > Epson (read: well calibrated printers). Not linearising with 
> > a  manual Spectrocam the customised Epson 9000 made for dye 
> > inks loaded with third party quad inks :-)
> 
> Busy day yesterday. Bill its should be.

Ernst, the idea of multiple linearizations is different from tracking 
printer instabilities and/or reading errors. Your points are all 
valid but imho do not tackled the issue I'm referring to.
Linearization is a simple function where you simply swap input-output 
values and get the corrections needed. With a single chanel output, 
probability is high that this simple function will achieve the aim. 
Now with ink split curves and overlaps, I doubt the function remains 
that simple because it's my understanding there's no simple model of 
reversing multiple density outputs due to dot gain, light scaterring, 
overlapping, etc... So sampling iterative adjustments is the best 
option to optimize the output untill you reach the point of tracking 
noise. This is the way linearization is proceeded in commercial 
packages such as Studioprint and this is also what I have experienced 
with QTR : once the output is linearized even on a 51stepwedge I 
don't get a clean linear scale with error above dE 2, read with a 
DTP41 uv-cut in Colorport. Alternative readings with a DTP20-
ColorShop (aka Pulse) deliver the same results and multiple readings 
remain in a dE 0.5 acceptable range after proper drying time.

Anyway, I agree with you that a proper calibration and in the quad 
case, a proper ink split curve shaping is critical. At the time being 
this is my depressing concern, far ahead linearization. I know 
there's a computation to do it but I don't know which. 

Olivier

Olivier wrote:
>  
>>> I'm sure that David has nice automatic equipment that may 
>>> read the targets patch wise and he started making his 
>>> excellent profiles for Epson 9600's with some support from 
>>> Epson (read: well calibrated printers). Not linearising with 
>>> a  manual Spectrocam the customised Epson 9000 made for dye 
>>> inks loaded with third party quad inks :-)
>> Busy day yesterday. Bill its should be.
> 
> Ernst, the idea of multiple linearizations is different from tracking 
> printer instabilities and/or reading errors. Your points are all 
> valid but imho do not tackled the issue I'm referring to.
> Linearization is a simple function where you simply swap input-output 
> values and get the corrections needed. With a single chanel output, 
> probability is high that this simple function will achieve the aim. 
> Now with ink split curves and overlaps, I doubt the function remains 
> that simple because it's my understanding there's no simple model of 
> reversing multiple density outputs due to dot gain, light scaterring, 
> overlapping, etc... So sampling iterative adjustments is the best 
> option to optimize the output untill you reach the point of tracking 
> noise. This is the way linearization is proceeded in commercial 
> packages such as Studioprint and this is also what I have experienced 
> with QTR : once the output is linearized even on a 51stepwedge I 
> don't get a clean linear scale with error above dE 2, read with a 
> DTP41 uv-cut in Colorport. Alternative readings with a DTP20-
> ColorShop (aka Pulse) deliver the same results and multiple readings 
> remain in a dE 0.5 acceptable range after proper drying time.
> 
> Anyway, I agree with you that a proper calibration and in the quad 
> case, a proper ink split curve shaping is critical. At the time being 
> this is my depressing concern, far ahead linearization. I know 
> there's a computation to do it but I don't know which. 
> 
> Olivier

I know it is a different approach. Maybe because I started 
with more mediocre equipment that I'm more skeptic about 
consistency in the total linearisation method and used 
multiple targets + measurement averaging as the better 
option for my system. On the other hand I did the 
partitioning good at that time and got nice linear output.

You could have a point that correction on the overlaps may 
not work out as needed but I think the QTR linearisation on 
the partitioned channels total as theoretically better than 
calibrating individual channels and the partitioning done 
after that like happens on my Z3100. It could be compensated 
with an active partitioning in the Z3100 though. And I have 
no reason to complain about the Z3100 results. How 
Studioprint does that I do not know. If QTR's linearisation 
corrects the output in fine enough steps and the printer is 
consistent in its output it should cope with overlaps, 
dotgain etc. If that isn't the case iterative measurements 
will not help either. Of course when iterative measuring is 
implanted in QTR's linearisation this should mean finer 
adjustment on the output if the adjustment was not fine 
enough already. Roy is the one qualified to give an answer 
on that but he was already qualified when he wrote that 
linearisation tool so I guess it must be working correctly.


-- 
Met vriendelijke groeten, Ernst


|  Dinkla Grafische Techniek  |
|     www.pigment-print.com    |
|             ( unvollendet )            |

> You could have a point that correction on the overlaps may 
> not work out as needed but I think the QTR linearisation on 
> the partitioned channels total as theoretically better than 
> calibrating individual channels and the partitioning done 
> after that like happens on my Z3100. 

To clear misunderstanding, I don't mean QTR linearization routine is 
any better or worse than others. It's simply that there's to my 
knowledge no modelling of overlapping densities. So one can not treat 
a multiple density output as a single density output.

>Of course when iterative measuring is 
> implanted in QTR's linearisation this should mean finer 
> adjustment on the output if the adjustment was not fine 
> enough already. Roy is the one qualified to give an answer 
> on that but he was already qualified when he wrote that 
> linearisation tool so I guess it must be working correctly.

I tried a double linearizeation in QTR without output, just 
evaluating the curve graph : it looks like it works.

Olivier

Hi everyone
I read The measurement target with ColorPort With my X-rite DTP20 UV. To be honest i don't 
remember The default Observer Angle in ColorPort. Does This setting (observer angle) will 
effect the linearized process a lot? Currently im set observer angle to 10.

Thanks,

The QTR linearization is not per channel so overlaps of different inks
don'tchange.
 So there's no iterative issue and it isn't possible.
I think the biggest variables are the printer and the reader.   The 21x4
wedge is
the most effective because it'll average 4 patches and readings for each
step.
It works very well as long as there aren't a lot of wiggles in the basic
output -- you
can't fix that very well with a linearization correction.

Roy


On Jan 17, 2008 6:55 AM, Olivier <odesmais@...> wrote:

> > You could have a point that correction on the overlaps may
> > not work out as needed but I think the QTR linearisation on
> > the partitioned channels total as theoretically better than
> > calibrating individual channels and the partitioning done
> > after that like happens on my Z3100.
>
> To clear misunderstanding, I don't mean QTR linearization routine is
> any better or worse than others. It's simply that there's to my
> knowledge no modelling of overlapping densities. So one can not treat
> a multiple density output as a single density output.
>
> >Of course when iterative measuring is
> > implanted in QTR's linearisation this should mean finer
> > adjustment on the output if the adjustment was not fine
> > enough already. Roy is the one qualified to give an answer
> > on that but he was already qualified when he wrote that
> > linearisation tool so I guess it must be working correctly.
>
> I tried a double linearizeation in QTR without output, just
> evaluating the curve graph : it looks like it works.
>
> Olivier
>
>
>
>
> Yahoo! Groups Links
>
>
>
>


[Non-text portions of this message have been removed]