[Tig] new on the TIG wiki

[Bob Friesenhahn]

On Wed, 9 May 2007, Jim Houston wrote:
>
> Unless I'm missing something, this picture doesn't show scans
> at all.  It seems to be a simulation of 'scanning an RGB rendered image'
> at an arbitrary resolution.  I could also use this method to 'prove'
> that you need
> a 1000K scan.     This looks like the visual equivalent of a false
> syllogism.

I agree.  With actual scanning there are many factors at work.  A bit 
of filtering (antialiasing) in the scanning system (or after the fact) 
can do wonders to eliminate the annoyances that Peter's image shows. 
A responsible image capture system should be intentionally filtering 
out frequencies above the nyquist limit.  Of course, many image 
capture systems leave this as an exercise for the user.

The filtering does not work so well with film grain if the grain is 
sampled at a lower frequency than the minimum frequency required to 
represent it (something that Peter is usually emphasizing).

A scanner which scans a beam over the film can easily employ 
horizontal filtering (low pass filter) in front of its A/D to minimize 
aliasing and improve the actual sampling resolution since it considers 
more than just the sample points.  A CCD-based system does not provide 
a built-in filtering mechanism since it uses point sampling.

As a little test, I took Peter's "original" pattern and used 
GraphicsMagick to reduce it to 25% of its original size and then 
expand back to the original size.  The blocky artifacts from the 
original were removed but the contours in the final image are still 
pretty close to the original.  There is more visible loss when 
converting to 10% of the original size and back.

Here are typical command lines using GraphicsMagick

gm convert original.png -resize '50%' -resize '200%' final_50.png
gm convert original.png -resize '25%' -resize '400%' final_25.png
gm convert original.png -resize '10%' -resize '1000%' final_10.png

In order to use point sampling to obtain the smaller image (closer to 
Peter's approach), change the initial '-resize' to '-sample'.  To 
obtain blocky images similar to Peter's approach, change the second 
'-resize' to '-sample' as well.

Bob
======================================
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/