[filmscanners] RE: Suggestions for scanning 4x5 transparencies

["Paul D. DeRocco" <pderocco@ix.netcom.com>]

>Laurie Solomon wrote:
>
>> The 720dpi figure (360 for large format printers) is the exact
>> resolution that the Epson driver converts to _before_ dithering.
>
> That was not my understanding; but I am willing to stand corrected.
> My understanding was that the input resolution should be between 200
> and 300 dpi for the desk top photo inkjets where the printer would
> convert it to 720 or some higher multiple of that as a result of the
> dithering process and that the 720 dpi or higher figures should not
> be used to determine input resolutions since one does not have a
> means of factoring in the dithering factor unless one had Epson's
> proprietary algorithms.

No, it's been proven by people on one of the Epson lists (perhaps it was the
old Leben list) that the driver resamples to exactly 720ppi. This means that
printing an image with strong high spatial frequency content at something
close to but not equal to 720ppi will produce the same sort of aliasing that
you get when resampling to 720ppi in Photoshop with "nearest neighbor" mode
selected--which is why resampling manually in bicubic mode can be used to
solve the problem.

For instance, if you create an image with alternating one-pixel black and
white lines, and then set the resolution to 700ppi (pixels per inch), you've
created an image whose spatial frequency is 350lpi (lines per inch). If you
give that to the 720ppi driver, you'll get 350 black lines (and 350 white
gaps) for every 720 pixels, or 35 lines for every 72 pixels. This means that
there will necessarily be one double-width black line (and one double-width
white gap) every 72 pixels. Since 72 pixels is a tenth of an inch, this
represents a 10lpi alias component, which corresponds to the difference
between the Nyquist frequency of 360lpi (half the 720ppi resampling
frequency) and the 350lpi image.

The dithering is something separate that happens after the resampling to
720ppi. That is, the resampling to 720ppi has nothing to do with the
dithering, and is unaffected by the precise dithering algorithm. It seems to
me that it ought to be possible to invent a dithering algorithm that accepts
input at any resolution, and has the necessary resampling filtering built
right in. Epson may have not done this because they don't know how, or
perhaps because there are patents that prevent them. I don't know.

>> And since the conversion to 720dpi is an unfiltered
>> conversion, there are situations where it causes visible aliasing,
>> and you get better results by manually resampling.
>
> Here again, that is not my understanding; but I am willing to be
> corrected. Alas, in such situations, I am not sure how one determines
> if the artifacts and alaising are a product of what you suggest or
> some other factor in the process.  I do know that many suggest using
> an input resolution that is a even multiple of 360 dpi and that an
> uneven multiple of that resolution will cause some of the same
> problems as you suggest.

The only really objectional form of aliasing is moire, and is generally
pretty recongizeable. To avoid aliasing in this case, resample up to the
next higher integral submultiple of 720 (that is, 720, 360, 240, 180, 144,
120), or down to 720 if you're starting with a resolution above 720.

Aliasing can also occur on incoherent high-frequency information, like
leaves on a tree. However, in this case, since what is being resampled is
somewhat random, the aliasing will also be random (albeit coarser). Since
all randomness looks more or less the same, the resulting aliasing can
actually be pleasing. For instance, a scenic shot of a view across a forest,
when "properly" resampled, might produce smooth green tones with some
variations from tree to tree. But if the distant leaf detail is allowed to
alias, it can produce graininess in the image that suggests more detail than
is really there, but still looks good to the eye.

> Thus, in light of your post, I would suggest that the user would have
> to test for themselves to see if they needed to resample or not.
> I,however, would suggest ( and I do not know if you would agree or
> not) that the least amount of resampling one does the better in most
> cases; so use interpolation and downsampling sparingly and only when
> absolutely needed.

I agree. Since aliasing is only really objectionable when there is
_coherent_ high-frequency spatial energy (resulting in moire patterns),
doing manual resampling to avoid it is only necessary on certain images. For
instance, a distant picket fence, or a tweed pattern, or focusing on a
window screen. These images are probably rare enough that you can just
pretend this problem doesn't exist until you see some moire in a particular
print, and then just redo that print. Personally, I've yet to see it crop up
in any of my prints.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@ix.netcom.com

----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe 
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or 
body