Re: [Gimp-print-devel] New dither codes

[Thomas T. <tt...@bi...>]

Robert L Krawitz wrote:

> For some reason, the 360 dpi prints seem to be much lighter on every
> printer.  Try bumping up the density to 1.5 at 360 dpi and see if it
> helps.  How fast was this mode?

360 DPI on the Stylus Color 600 seems to produce smaller dots than 720
or for that matter 1440 HQ.

I think 360 is supposed to use larger dots. The dot size selections vary
between models, so it seems easy enough to get mixed up:

Model:  600             IIs,ProXL  800,850
Value:
0x00    Default         Default    Default
0x01    N/A             Small      Micro
0x02    Micro           Standard   Normal(single)
0x03    Normal(single)  N/A        Normal(double)
0x04    Normal(double)  N/A        N/A

I think this is really strange: the code that selects the smallest dots
on 800 and IIs (and more) does not do so on the 600. I would expect more
discrepancies with other printers.

Other results: I went from rotating and inverting the matrix to just
shidting it in X and Y directions. For ordered dither, this made no big
difference, but the splotches are not colored any more. I think I know
where the splotches originate from. If I print a white-to-black wedgeit
really prints as a white-to-grey, then grey-to-darkgreen, then
green-to-darkgrey (that dark grey is a bit lighter than the green) and
then to black. This is on Canon coated paper, but plain paper is
similar. It seems black has to be added earlier than it is now - at
least it has to be added in larger amounts.

If CMY print all at 70 % and you add 30% black, the black has very
little effect on overall density, because it overlaps most of the CMY. I
guess this is what is happening.

1440x720 is very ugly in ordered dither mode, as only half the dots are
not printed. 1440 mode thus is only useable with some kind of error
diffusion or other mechanism that can use half dot spacings. Ordered
dither inherently cannot do this because a dot position that is printed
at level X will also be printed at a level higher than X.

The new matrix that I built is useless. I'll have to try something else
- I'm thinking about using blue noise now. Generating white noise, and
filtering it to remove the low frequencies. Actually, Floyd-Steinberg is
pretty good at removing low frequencies: that is why floyd-steinberg
with a random threshold does very well.



Thomas