Digital BW, The Print

I just did one comparison test by upsampling an rgb image of a
landscape composite from 45 megs to 178 megs from a clients 8 bit
file. The first was done with the new Stairstep Interpolation 2
software, the second was done Bicubic Smooth in Photoshop CS2, both
set to 300 dpi. No matter how I finessed the final sharpening of the
final file, the Photoshop interpolation was always superior, both in
regard to sharpness as well as smoothness of the irregular artifacts
present from such a serious upscaling. 

Now I am going to do three or four more tests this week to investigate
this furthur. Whatever they did to improve the Bicubic mechanism in
Photoshop, it is a major improvement. I'm impressed.

John



--- In DigitalBlackandWhiteThePrint@yahoogroups.com, "Bob Frost"
<bob@f...> wrote:
> Paul,
> 
> I've always thought 'stair interpolation' is a very peculiar way to 
> interpolate an image, and I don't understand how/why it is supposed
to work. 
> When you upsample an image by 10%, what happens is that the 10th, 20th, 
> 30th, etc., lines are 'interpolated' and a new line inserted in
front of 
> each of them, based on that line and the surrounding lines
(depending on the 
> method of interpolation). When you then do the second step of
another 10%, 
> the 10th, 20th, 30th, etc lines are again interpolated, the only
difference 
> being that the original 20th, 30th, etc lines have all got pushed
along by 
> the insertion of the previous new lines. And so on. Seems a strange
way to 
> upsample an image!!
> 
> Bob Frost.
> 
> ----- Original Message ----- 
> From: "Paul D. DeRocco" <pderocco@i...>
> >
> > Stair interpolation refers to iteratively computing a spline over and
> > over again, usually in fixed proportions, until you get the output
size
> > you want. For example, if you used bicubic interpolation and increased
> > the file size by 10% each time, until you got where you wanted to be,
> > that would be stair interpolation. If you used S spline the same way,
> > that would also be stair interpolation.
> 
> It seems to me that any linear filter applied multiple times is just
another
> different linear filter, and can be precomputed as such in advance, and
> applied in the same amount of time it takes to do any other single
filter.
> Doing it iteratively seems like the dumb way to do it (if you're
writing the
> software, that is), and might also build up noise due to accumulated
> roundoff errors, especially if computation is done in 8-bit mode.