QTR-Quadtone RIP

I know the question has been asked in the past regarding QTR dithering 
method but the reply was basically "try and see by yourself".

Has anyone some feed back on the various methods or could I be re-
directed to a (past) discussion,url,book...

Thanks.

Olivier

John, thanks for the below. In fact I'm looking for an explanation of 
how the different dithering methods calculate the print and the 
difference between each.

Olivier

--- In QuadtoneRIP@yahoogroups.com, "John Moody" <moodymz3@y...> wrote:
> The reply has not changed, but I have been using eventone for matte 
paper at

> 1440 super, and ordered for luster at 2880 dpi with the R200 printer.
> 
> Best regards,
> John Moody
>

Based on this 3 links :

http://gimp-print.sourceforge.net/developer-html/gimpprint_26.html
http://man.digipedia.pl/man/man7/gimpprint-dithers.7.html
http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id=00BR5g

it's very hard to find out the effects of each dithering methods 
(except "fast" and "very fast" are not adapted to fine BW prints).

The thread from photonet states that only ordered,fast and Adaptive 
Hybrid are implemented in QTR ?

The one from sourceforge seems to recommend Adaptive Hybrid since 
ordered is only preferable for 4colors none variable dot sized 
printers, A/H would use error diffusion in dark & halftones and ordered 
in light tones.


None of these papers is too explicite. Any additonal comments from 
better experimented people than me...

Olivier


--- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> wrote:
> John, thanks for the below. In fact I'm looking for an explanation of 
> how the different dithering methods calculate the print and the 
> difference between each.
> 
> Olivier
> 
> --- In QuadtoneRIP@yahoogroups.com, "John Moody" <moodymz3@y...> 
wrote:
> > The reply has not changed, but I have been using eventone for matte 
> paper at
> > 1440 super, and ordered for luster at 2880 dpi with the R200 
printer.

> > 
> > Best regards,
> > John Moody
> >

There's really only two dithering methods that I think are worth looking at.
They are Ordered and Adaptive Hybrid.  From looking at the code both Floyd
and Eventone just do Adaptive Hydrid.  Fast and Very Fast aren't that good as
far as I can tell.  

The results from Ordered and Adaptive are pretty close -- they are different but
overall give similar results.  On any given image there may be visible differences
but I had a hard time conclusively saying one was clearly and always better than the other.
Early on I remember seeing an occasional yet rare artifact in the Adaptive dither
on specific test prints - not real photos.  That combined with the noticeably faster
algorithm for Ordered, I always use Ordered.

Ordered dither is the only one that I can say I mostly understand.  There seem to
be some docs that imply that it doesn't do well with variable dot sizes.  I don't in
general find that to be true but I have been experimenting with the handling of
variable dot sizes.

Roy


--- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> wrote:

> Based on this 3 links :
> 
> http://gimp-print.sourceforge.net/developer-html/gimpprint_26.html
> http://man.digipedia.pl/man/man7/gimpprint-dithers.7.html
> http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id=00BR5g
> 
> it's very hard to find out the effects of each dithering methods 
> (except "fast" and "very fast" are not adapted to fine BW prints).
> 
> The thread from photonet states that only ordered,fast and Adaptive 
> Hybrid are implemented in QTR ?
> 
> The one from sourceforge seems to recommend Adaptive Hybrid since 
> ordered is only preferable for 4colors none variable dot sized 
> printers, A/H would use error diffusion in dark & halftones and ordered 
> in light tones.
> 
> 
> None of these papers is too explicite. Any additonal comments from 
> better experimented people than me...
> 
> Olivier
> 
> 
> --- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> wrote:
> > John, thanks for the below. In fact I'm looking for an explanation of 
> > how the different dithering methods calculate the print and the 
> > difference between each.
> > 
> > Olivier
> > 
> > --- In QuadtoneRIP@yahoogroups.com, "John Moody" <moodymz3@y...> 
> wrote:
> > > The reply has not changed, but I have been using eventone for matte 
> > paper at
> > > 1440 super, and ordered for luster at 2880 dpi with the R200 
> printer.
> > > 
> > > Best regards,
> > > John Moody
> > >

Thanks Roy,

Since it's a Gimp based processing I'll try to get more info on the 
Gimps forum.

Could you possibly let me know the Gimp version used in QTR : there's 
been some questionning about it inb the past and I saw the latest QTR 
beta uses files dates July 5th 05 : have you updated the print engine 
or anything recently ?

Thanks again 

Olivier

--- In QuadtoneRIP@yahoogroups.com, "Roy Harrington" <roy@h...> wrote:
> 
> There's really only two dithering methods that I think are worth 
looking at.
> They are Ordered and Adaptive Hybrid.  From looking at the code 
both Floyd
> and Eventone just do Adaptive Hydrid.  Fast and Very Fast aren't 
that good as
> far as I can tell.  
> 
> The results from Ordered and Adaptive are pretty close -- they are 
different but
> overall give similar results.  On any given image there may be 
visible differences
> but I had a hard time conclusively saying one was clearly and 
always better than the other.
> Early on I remember seeing an occasional yet rare artifact in the 
Adaptive dither
> on specific test prints - not real photos.  That combined with the 
noticeably faster
> algorithm for Ordered, I always use Ordered.
> 
> Ordered dither is the only one that I can say I mostly understand.  
There seem to
> be some docs that imply that it doesn't do well with variable dot 
sizes.  I don't in
> general find that to be true but I have been experimenting with the 
handling of
> variable dot sizes.
> 
> Roy
> 
> 
> --- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> wrote:
> > Based on this 3 links :
> > 
> > http://gimp-print.sourceforge.net/developer-html/gimpprint_26.html
> > http://man.digipedia.pl/man/man7/gimpprint-dithers.7.html
> > http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id=00BR5g
> > 
> > it's very hard to find out the effects of each dithering methods 
> > (except "fast" and "very fast" are not adapted to fine BW prints).
> > 
> > The thread from photonet states that only ordered,fast and 
Adaptive 
> > Hybrid are implemented in QTR ?
> > 
> > The one from sourceforge seems to recommend Adaptive Hybrid since 
> > ordered is only preferable for 4colors none variable dot sized 
> > printers, A/H would use error diffusion in dark & halftones and 
ordered 
> > in light tones.
> > 
> > 
> > None of these papers is too explicite. Any additonal comments 
from 
> > better experimented people than me...
> > 
> > Olivier
> > 
> > 
> > --- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> 
wrote:
> > > John, thanks for the below. In fact I'm looking for an 
explanation of 
> > > how the different dithering methods calculate the print and the 
> > > difference between each.
> > > 
> > > Olivier
> > > 
> > > --- In QuadtoneRIP@yahoogroups.com, "John Moody" 
<moodymz3@y...> 
> > wrote:
> > > > The reply has not changed, but I have been using eventone for 
matte 

> > > paper at
> > > > 1440 super, and ordered for luster at 2880 dpi with the R200 
> > printer.
> > > > 
> > > > Best regards,
> > > > John Moody
> > > >

Hi Oliver,

The gimp-print version I started with is way back -- 4.3.5.   I haven't
changed much of the dither code but have tweaked many of the tables
to get much more accurate parameters such as ink drop sizes.

The whole dithering issue is interesting and there's a lot of 
non-intuitive
aspects of the problems.  The eye catches amazingly subtle patterns.
Speed of algorithms is very important. The shear amount of data that has
to get processed is quite astounding:  consider an R800 printing an 8x10
at 2880x1440 dpi.    That's 8x10 x 2880x1440 locations on the page.
That's 330 million locations, x 8 inks, 16bit = 2byte internals.
Total of 5 gigabytes processed for an 8x10!
What about the guy doing a banner on his 9600 -- 40 in x 25 feet?

On the other hand the idea of having multiple lighter grays is to make 
the
dither less and less visible.  As long as there isn't an obvious pattern
that sticks out like a sore thumb its usually OK.

One difficulty with some of the fancier algorithms is that they are ink 
dependent.
With different inks they may show no benefit or may actually be worse.

Roy

On Tuesday, September 27, 2005, at 07:55  PM, Olivier wrote:

> Thanks Roy,
>
> Since it's a Gimp based processing I'll try to get more info on the
> Gimps forum.
>
> Could you possibly let me know the Gimp version used in QTR : there's
> been some questionning about it inb the past and I saw the latest QTR
> beta uses files dates July 5th 05 : have you updated the print engine
> or anything recently ?
>
> Thanks again
>
> Olivier
>
> --- In QuadtoneRIP@yahoogroups.com, "Roy Harrington" <roy@h...> wrote:
>>
>> There's really only two dithering methods that I think are worth
> looking at.
>> They are Ordered and Adaptive Hybrid.  From looking at the code
> both Floyd
>> and Eventone just do Adaptive Hydrid.  Fast and Very Fast aren't
> that good as
>> far as I can tell.
>>
>> The results from Ordered and Adaptive are pretty close -- they are
> different but
>> overall give similar results.  On any given image there may be
> visible differences
>> but I had a hard time conclusively saying one was clearly and
> always better than the other.
>> Early on I remember seeing an occasional yet rare artifact in the
> Adaptive dither
>> on specific test prints - not real photos.  That combined with the
> noticeably faster
>> algorithm for Ordered, I always use Ordered.
>>
>> Ordered dither is the only one that I can say I mostly understand.
> There seem to
>> be some docs that imply that it doesn't do well with variable dot
> sizes.  I don't in
>> general find that to be true but I have been experimenting with the
> handling of
>> variable dot sizes.
>>
>> Roy
>>
>>
>> --- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...> wrote:
>>> Based on this 3 links :
>>>
>>> http://gimp-print.sourceforge.net/developer-html/gimpprint_26.html
>>> http://man.digipedia.pl/man/man7/gimpprint-dithers.7.html
>>> http://www.photo.net/bboard/q-and-a-fetch-msg?msg_id=00BR5g
>>>
>>> it's very hard to find out the effects of each dithering methods
>>> (except "fast" and "very fast" are not adapted to fine BW prints).
>>>
>>> The thread from photonet states that only ordered,fast and
> Adaptive
>>> Hybrid are implemented in QTR ?
>>>
>>> The one from sourceforge seems to recommend Adaptive Hybrid since
>>> ordered is only preferable for 4colors none variable dot sized
>>> printers, A/H would use error diffusion in dark & halftones and
> ordered
>>> in light tones.
>>>
>>>
>>> None of these papers is too explicite. Any additonal comments
> from
>>> better experimented people than me...
>>>
>>> Olivier
>>>
>>>
>>> --- In QuadtoneRIP@yahoogroups.com, "Olivier" <odesmais@y...>
> wrote:
>>>> John, thanks for the below. In fact I'm looking for an
> explanation of
>>>> how the different dithering methods calculate the print and the
>>>> difference between each.
>>>>
>>>> Olivier
>>>>
>>>> --- In QuadtoneRIP@yahoogroups.com, "John Moody"
> <moodymz3@y...>
>>> wrote:
>>>>> The reply has not changed, but I have been using eventone for
> matte
>>>> paper at
>>>>> 1440 super, and ordered for luster at 2880 dpi with the R200
>>> printer.
>>>>>
>>>>> Best regards,
>>>>> John Moody
>>>>>
>
>
>
>
>
>
> Yahoo! Groups Links
>
>
>
>
>
>
-
Roy Harrington
roy@...
Black & White Photo Gallery
http://www.harrington.com

Roy Harrington wrote:

> Speed of algorithms is very important. The shear amount of data that has
> to get processed is quite astounding:  consider an R800 printing an 8x10
> at 2880x1440 dpi.    That's 8x10 x 2880x1440 locations on the page.
> That's 330 million locations, x 8 inks, 16bit = 2byte internals.
> Total of 5 gigabytes processed for an 8x10!
> What about the guy doing a banner on his 9600 -- 40 in x 25 feet?

According to Robert Krawitz the next version of Gimp-print aka 
Gutenberg can handle 32 channels with 16 bit input. So it 
isn't getting less complex. Will there be anyting gained for 
B&W printing if you ever base QTR on the newer versions of 
Gimp-print ?

On EvenTone and Ordered dithering there was something written 
that's interesting, EvenTone usually unloads the droplets 
often in one pass even when more passes are used:

Quote:

  Robert L Krawitz wrote:

    >Graeme Gill wrote:
    >   Good point. So to use this type of screening in the 
real world, it
    >   would be desirable to have the different color planes 
screens
    >   de-correlated once the dot density grow high enough, 
thereby
    >   getting the benefits in the highlights where it 
counts, without
    >   triggering banding and pattern issues.
    >
    >Why would this *trigger* banding issues?  At worst it 
wouldn't help
    >solve them, but I don't see how it would actually trigger any
    >problems.

    Because if you have two screens that are correlated, they 
need to
    be printed in exact alignment (to sub pixel precision) to 
avoid
    moiri type effects between them (assuming the pixel 
density isn't
    sparse). Banding effectively displaces pixels (generally 
sub pixel
    sized displacements) in a pseudo random "banding" fashion, 
so moiri
    effects amplify the banding displacements. If the screens are
    completely uncorrellated, then banding displacements will 
have much
    less visible effect.

Good point.

One other thing that I've had good experiences with (mostly for a
different reason) is to use a very good screen (Raph Levien's 
EvenTone
Screening) with some perturbation.  In this case, the purpose 
of the
perturbation is to break up screening artifacts that create
patterning, but this slight decorrelation may also favorably 
affect
banding.  The EvenTone Screen algorithm suffers from banding 
more than
the Ordered (farthest neighbor matrix) dither algorithm that's our
mainstay, and watching the printer actually print suggests 
that this
would be the case.  It's very common for something printed with
EvenTone to print the vast majority of its drops in one pass, 
even if
4 or 8 passes are being used.  Ordered dithering doesn't 
suffer from
this, and is faster.  The problem with ordered dithering, of 
course,
is the noise that's visible in the midtones.

Using the dither matrix to perturb EvenTone screening 
partially breaks
this up and reduces the banding a bit.

-- Robert Krawitz

End of Quote

                    --
           Ernst Dinkla


www.pigment-print.com
(         unvollendet         )

--- In QuadtoneRIP@yahoogroups.com, Ernst Dinkla <E.Dinkla@c...> wrote:
> Roy Harrington wrote:
> 
> > Speed of algorithms is very important. The shear amount of data that has
> > to get processed is quite astounding:  consider an R800 printing an 8x10
> > at 2880x1440 dpi.    That's 8x10 x 2880x1440 locations on the page.
> > That's 330 million locations, x 8 inks, 16bit = 2byte internals.
> > Total of 5 gigabytes processed for an 8x10!
> > What about the guy doing a banner on his 9600 -- 40 in x 25 feet?
> 
> According to Robert Krawitz the next version of Gimp-print aka 
> Gutenberg can handle 32 channels with 16 bit input. So it 
> isn't getting less complex. Will there be anyting gained for 
> B&W printing if you ever base QTR on the newer versions of 
> Gimp-print ?

The issue is getting the 16 bits through all the upper levels. I've been using a 16 bit
interface with gimp-print code all along. 

> 
> On EvenTone and Ordered dithering there was something written 
> that's interesting, EvenTone usually unloads the droplets 
> often in one pass even when more passes are used:

Sounds very strange to do it in one pass.  One pass of a head puts down drops in
a very regular pattern -- a 2200 puts down an exact 360x180 set of drops in a pass.

Roy

> 
> Quote:
> 
>   Robert L Krawitz wrote:
> 
>     >Graeme Gill wrote:
>     >   Good point. So to use this type of screening in the 
> real world, it
>     >   would be desirable to have the different color planes 
> screens
>     >   de-correlated once the dot density grow high enough, 
> thereby
>     >   getting the benefits in the highlights where it 
> counts, without
>     >   triggering banding and pattern issues.
>     >
>     >Why would this *trigger* banding issues?  At worst it 
> wouldn't help
>     >solve them, but I don't see how it would actually trigger any
>     >problems.
> 
>     Because if you have two screens that are correlated, they 
> need to
>     be printed in exact alignment (to sub pixel precision) to 
> avoid
>     moiri type effects between them (assuming the pixel 
> density isn't
>     sparse). Banding effectively displaces pixels (generally 
> sub pixel
>     sized displacements) in a pseudo random "banding" fashion, 
> so moiri
>     effects amplify the banding displacements. If the screens are
>     completely uncorrellated, then banding displacements will 
> have much
>     less visible effect.
> 
> Good point.
> 
> One other thing that I've had good experiences with (mostly for a
> different reason) is to use a very good screen (Raph Levien's 
> EvenTone
> Screening) with some perturbation.  In this case, the purpose 
> of the
> perturbation is to break up screening artifacts that create
> patterning, but this slight decorrelation may also favorably 
> affect
> banding.  The EvenTone Screen algorithm suffers from banding 
> more than
> the Ordered (farthest neighbor matrix) dither algorithm that's our
> mainstay, and watching the printer actually print suggests 
> that this
> would be the case.  It's very common for something printed with
> EvenTone to print the vast majority of its drops in one pass, 
> even if
> 4 or 8 passes are being used.  Ordered dithering doesn't 
> suffer from
> this, and is faster.  The problem with ordered dithering, of 
> course,
> is the noise that's visible in the midtones.
> 
> Using the dither matrix to perturb EvenTone screening 
> partially breaks
> this up and reduces the banding a bit.
> 
> -- Robert Krawitz
> 
> End of Quote
> 
>                     --
>            Ernst Dinkla
> 
> 
> www.pigment-print.com
> (         unvollendet         )