Rescaling question

[Leonard Evens]

I regularly scan my 4 x 5 film at 3200 ppi with my Epson 3200 scanner.
I then rescale in my photoeditor to 2000 ppi in order to speed up
rocessing. I see little loss of fine detail in so doing. Recently I
scanned a test bw negative I made with included some fine line patterns.
After rescaling I noticed a slight amount of moire in the reproduced
line patterns. In principle that could be the result of aliasing.

I wonder if I might be slightly better off if I used a small gaussian
blur before rescaling. If so, how much should I use?

 

[Kennedy McEwen]

I regularly scan my 4 x 5 film at 3200 ppi with my Epson 3200 scanner.
I then rescale in my photoeditor to 2000 ppi in order to speed up
rocessing. I see little loss of fine detail in so doing. Recently I
scanned a test bw negative I made with included some fine line
patterns. After rescaling I noticed a slight amount of moire in the
reproduced line patterns. In principle that could be the result of
aliasing.

It certainly will be, information which is resolved at the original
sampling density, yet incapable of being resolved at the lower sampling
density.

An interesting question which actually proves that the Epson 3200 can
resolve more than 1000lp/in (approx. 40lp/mm).

I wonder if I might be slightly better off if I used a small gaussian
blur before rescaling. If so, how much should I use?

Some form of blur is definitely required - and resharpen after
rescaling.

How much is best determined by trial and error because it depends on how
much MTF the scanner has above the target sampling density which, as you
know, is the subject of much debate in these here parts. ;-)

It also depends critically on the application that you are using for the
resampling and the type of resampling method you choose. Photoshop, for
example, already applies a certain amount of blur before downsampling
specifically to address this issue. I believe that Paintshop Pro does
this now as well, since v8.

Gaussian blur may not be the best solution though, since the fourier
transform of a gaussian is also a gaussian. You are looking for
something that would retain as much MTF below the Nyquist of the target
sample density, yet have zero response above that limit. As you know,
gaussians extend out to infinity, so getting a reasonable attenuation
above Nyquist means significant unwanted attenuation below it. Ideally,
the brick wall spatial frequency filter you would need would be a sinc
filter. You might find a plug-in that will do this, or experiment with
some user defined blur filters to approximate a suitably smoothed sinc.

 

[Bart van der Wolf]

I regularly scan my 4 x 5 film at 3200 ppi with my Epson 3200
scanner. I then rescale in my photoeditor to 2000 ppi in order to
speed up processing. I see little loss of fine detail in so doing.
Recently I scanned a test bw negative I made with included some
fine line patterns. After rescaling I noticed a slight amount of moire
in the reproduced line patterns. In principle that could be the result
of aliasing.

It almost certainly is.

I wonder if I might be slightly better off if I used a small gaussian
blur before rescaling. If so, how much should I use?

My rule of thumb is something in the order of 0.2-0.3 radius per integer
reduction factor, but each photoeditor is a bit different in the amount of
pre-blur (if any) before downsizing, and you have the choice of different
algorithms for different types of image content.

A method that's quick and usually creates relatively few (but more than with
the above blur) artifacts is by using Photoshop's Filter|Pixelate|Mosaic...
with a cell size equal to the reduction factor (e.g. 1:2 cell size 2, 1:3
cell size 3, etc.), followed by a resize (to 50%, 33.33%, etc.). In your
case that would give you a 1600 ppi resolution, which might be too small, I
don't know if there is a special reason you chose 2000 ppi.

You could also consider ...Mosaic with a cell size of 2 and either a resize
to only 62.5% (mathematically not solid), or a resize to 50% followed by a
resize to 125% (or a stepped approach of 110% and 113.64%, or 111.8% and
111.8%). A minute preblur might help in this case as well. The stepped
resize up, already serves as a kind of re-sharpening in Photoshop.

All this is more art than science, but it may work quite well for most
continuous tone images. It does create a few visible artifacts on a critical
test target, unless you preblur from the start with a radius of 0.1,
although it all depends on the photoeditor in question.

You may also want to try the free ImageMagick Q16
(http://www.imagemagick.org/) suite of image processing functions. I have
yet to try it myself, but I have been told it does preblur.

If you need a test target that is probably more critical than average
images, you can download one at:
http://www.xs4all.nl/~bvdwolf/main/downloads/Rings.gif

Bart