Digital BW, The Print

Steve,
growing curiosity was pushed over the edge into action by your post. I
did  not have time to do a comprehensive test but was curious to see
the magnitude of any effects.  

My mono printer is a 1270 with some MIS UT FSN inks - not the whole
set and used with my own curves (which are OK but not perfect).  This
is clearly non-standard so I also carried out the same experiment in
colour on a 1290 with MIS archival inks, and also in BO with eboni on
the 1290).  

Normally I feed the printers with 200 ~ 360 ppi images.  I expected
that the 200 dpi end would be the best test.  

Question: is there a significant difference in the appearance of a 200
ppi image sent to the 1270/1290 EPSON drivers compared to the same
image printed after first interpolating by a known algorithm.
  
The most convenient pre-print interpolation I had available was to use
Picture Window Pro. In a pre-test I looked at both bicubic with 16.7%
sharpening and Lanczos 6x6 and, given how close the results were in
that test decided to use the former method for the other tests. (I do
not posses PS.)  Interpolation was done to 720 ppi. Prints were all to
 the same size (about 10"x15").

The test image was of some tree canopies in generally high contrast
lighting, with sharp, unsharp, high-contrast and low-contrast areas. 
There was some image noise in the lower zones, from the camera, I
suspected that such noise might be a good way to distinguish the
methods. A wider range of test images would, naturally, be needed to
form a firm conclusion.

Answer: the interim conclusion is that differences due to the method
of resampling were very minor.  There were just-perceptible
differences in the background "noise" in some dark-mid-tone
out-of-focus areas, but nothing else I could distinguish from 10 inch
viewing distance in good light.  I suspect that very small
changes to sharpening explain the difference (on second thoughts this
is pretty obvious: the processes involved are linear and include
interpolation and sharpening, with what comes down to, probably, tiny
differnces in the "effective" kernel for the whole process if it were
to be reduced to a single step).  Perhaps the "smart" approach in
Qimage is different - I downloaded it to try but had some problems and
do not now have time to test it properly. 

What now? At the moment this aspect is not going to be my highest
priority in terms of workflow improvement, so no more tests in the
immediate future, and I'll trust the drivers to do the (modest)
resampling. 

If someone is surprised by this, I'd love to hear the details of the
contrary evidence, as I could quite easily have missed an important
point in what was a cursory test.  I find it quite probable that the
methods in the printer drivers break down when the resampling ratio
gets too large, but that is not happening in the regime in which I
work.

This post has grown too long.

Ken



--- In DigitalBlackandWhiteThePrint@yahoogroups.com, "Steve Kale"
<stevekale@b...> wrote:
> Given all the Genuine Fractals etc discussion I was wondering what
people's thoughts were on 
> this.  Is it best to lower the dpi and gain size (say 360 down to
300) or interpolate up 
> (through whatever method) and maintain the dpi?
> 
> Cheers
> 
> Steve