Thanks for the reply, Martin.

> Apart  from a handful of pro lenses that are truly

> diffraction limited at f4 this puts a lower  bound

> on the FWHM of 3.5 um or 280 lp/mm so a 36mm frame

> needs  at  most  9M  samples  assuming  a  perfect

> recording  medium. Realistically most lenses are a

> factor of two worse than this  rough  estimate  in

> practice.

9M  is  about  100 pixels/mm. How can 280 line pairs

fit into 100 pixels, if you do  consider  a  digital

image a medium close to perfect?

Also, if two systems connected is series have corre-

sponding resolutions Res1 and Res2, then  the  final

resolution is roughly:

              1/Res = 1/Res1 + 1/Res2

Doesn't  it mean that the scanning resolution should

exceed that of the image being scanned?

> Your low quality slide at high resolution consists

> of mostly grain noise and the image itself is very

> soft. He hasn't lost much  by  downsampling  it  -

> mostly edge detail around the dust.

In  the  comparisons I posted, look at the detail on

the cap, scarf, and on the texture of the coat.  The

loss is there, although not _very_ prominent.

> The  high frequency grain noise could be separated

> from the true image with a bit of cunning adaptive

> filtering.

The  addition  of  noise is an irreversible process,

and all noice-removal  tools  (for  both  sound  and

images)  operate  probablity  and  do not have exact

criteria of what it noise and what is not. They  may

partially  help,  but  always at the expense of some

real information. It is a fundamental law.

Anton