[Tig] new on the TIG wiki

[Jim Houston]

On May 9, 2007, at 10:07 AM, peter swinson wrote:
> Jim seems to think the graphic is an arbitrary resolution, it is not.

No, i'm thinking that rendered graphics can have an infinite resolution
when defining what the patch applies to.  Making reductions at 1/2,  
and 1/4, and, etc...
doesn't show anything about the performance of the film system but  
merely
the effect of resampling an original image that can have any resolution
you care to assign to it.    Labeling the last patch Nyquist because  
the signal
disappears doesn't tell you much about the signal or about the  
sampling structure
it is just saying that after 5 resizes the signal gets distorted.
The combination means that the image
as shown (and partly, as I understand the meaning of it) doesn't  
prove the point
you seem to be trying to make.

> Therefore sampling at even todays 2k/4K will not give accurate  
> representation of the
> slopes on high resolution edges.

I don't disagree with this only because film has a lot of adjacency  
effects
and edge sharpening in the crystals that make the story of it's  
resolution
complicated.    The 3D nature of the distribution of image forming  
elements
in an emulsion means there are a lot of optical effects that change the
effective resolving power of the film.   At a certain point, you just
can't get anything more out of a scan of a film because of diffraction
limits.   If you set your target aliasing at the 5% level, you get
an effective target of about 5800 pixels.  Nyquist sampling of this
would suggest a scan at 5800*2 = 11,600 pixels.  (see Roger Morton's
SMPTE paper on this)   But you are already well into the noise floor
of the system by that point, so there isn't much reason to go
that high with a scan.   There is no reason to think that a
20,000 pixel scan would give you better results.

So it still seems to me that your picture doesn't
demonstrate what you are saying.

Jim H.