[Gimp-print-devel] Re: Epson 870

[Jean-Marc V. <ver...@if...>]

On May 14, 10:41pm, Robert L Krawitz wrote:
> Subject: Re: Epson 870
> First off: the gray scale is much better than mine.
>
> 720 dpi looks pretty good.  1440 looks a bit distorted, and 1440
> enhanced (pseudo-1440x1440) looks positively strange.  I think that
> thaere's some kind of arithmetic overflow going on in darker regions;
> light tones are OK.  It might be the ink budget stuff triggering it, though.
>
> However, in general 1440 and 720 look noticeably different.

OK, this is kind of what I have found too, although I didn't find that
1440-enhanced whas _that_ bad. It could be a matter of density or paper quality
as well in that particular case.

>
> Try downloading http://www.tiac.net/users/rlk/colors.tif.  This is a
> color sweep pattern.
>

Got it.

>
>    2 - I made one major change to dithering. That's how we think about
>    density.  In my understanding, current gimp-print scales everything
>    by density/oversampling. That includes values to print and
>    thresholds.  My approach is to take advantage of oversampling to
>    use (more) lighter inks.  For that purpose, I'm scaling thresholds
>    only by density (not oversampling).  This results in an increase in
>    the usage of light inks as quality (oversampling) increases.
>
> I adopted the approach of scaling everything by density, and ignoring
> the oversampling issue per se (except for scaling density) between
> 3.1.2 and 3.1.3 because the behavior of the 3.1.2 code when the user
> changed density (or resolution) was too unpredictable.  I wanted the
> code to behave identically (as far as ink choice) for any choice of
> density and resolution.
>
> I'm not sure that using more light ink is the way to go.  If anything,
> I think that at high resolution you're often better off using more
> small dots, which means dark ink at some intermediate points, to avoid
> laying down too much ink.
>

I think you're right, but a little more light ink would help reduce grain.
In the version you've tried, I had basically pushed the light-ink button all
the way down and this is obviously not good. This is how I see the issue:
1 - there's no issue as long as there's no oversampling.
2 - With oversampling, any ink value between density/oversampling and density
should work in theory. The former will need to be printed several times, the
latter only once in a while. The former will give better smoothness, the latter
better color values. The former will be more sensitive to artifacts, like
additive errors or soaking the paper. Current gimp-print is on the latter side;
the safe side, whereas my test shot is the former. The issue as I see it, is to
find a good and stable compromise.

>    That's all (or almost), but was not easy with the current
>    dithering. I came across major (and unexpected) color problems. The
>    reason I believe is that for any given density, dithering will use
>    mostly one of 2 inks only (one darker one lighter). Without going
>    into the details, that means that if the settings for only one ink
>    level is not perfect, everything having a color density between the
>    immediately lighter ink and the immediately darker ink will
>    accumulate (and therefore amplify) this error, particularly in flat
>    tones. I think there is a problem with that.
>
> I think you're right.  I think that it should be possible for segments
> to overlap, so that if a particular point lies within multiple
> segments, they're each tried in turn.  I don't know if that would drop
> too much ink, though.
>

As of today, I've been trying to work something from our (mostly common)
conclusions. I'll keep you posted in a few days.

Thanks again for giving it a try and for your helpful comments.


-- 
--
Jean-Marc Verbavatz		<ver...@if...>
5, rue La Fontaine		"http://perso.cybercable.fr/verbavat"
F-75016 Paris