[comp.graphics] Aliasing question

shirley@uicsrd.csrd.uiuc.edu (11/15/89)

I have a theoretical question for you rasterheads:

How much filtering should we do on our images?  Given a VERY high
resolution display, we should certainly kill all freqencies greater
than 2 pixel widths (especially in the vertical if you have
an interlaced display).  

In the real world, we can still see the individual pixels, and sometimes
this is useful.  As an example suppose we have a function

               f(x) = black  floor(x) is even
                      white  floor(x) is odd 

Further suppose we have a 1000**2 pixel display, and x is in pixel coordinates.
This would have 1000 alternating vertical stripes of white and black, with
each stripe one pixel wide.  If we filter out the high freqency 
components then we'd have a grey screen.  Intuitively it seems that
we shouldn't filter.

Now suppose we have the function

               f(x) = black  floor((1.001*x) is even
                      white  floor((1.001*x) is odd 

This will give us 999 vertical stripes.  If we area sample each pixel,
then we'll have stripes on the left of the screen, fading into grey in the
middle of the screen, going back to stripes on the right.  This gives us
a classic aliasing effect, and the party line is that we should filter.

My question is whether this filtering is worth the trouble.  Is the
aliasing effect described above that bad in real world scenes?  How often
does this come up in real pictures, and how important is it?


Thanks,

pete shirley
shirley@m.cs.uiuc.edu

gg10@prism.gatech.EDU (Galloway, Greg) (11/16/89)

In article <44100007@uicsrd.csrd.uiuc.edu>, shirley@uicsrd.csrd.uiuc.edu writes:
> I have a theoretical question for you rasterheads:
>  ... 
> My question is whether this filtering is worth the trouble.  Is the
> aliasing effect described above that bad in real world scenes?  How often
> does this come up in real pictures, and how important is it?

Rob Cook's "Stochastic Sampling in Computer Graphics" from ACM Transactions
on Graphics, January 1986, is about the best reference on aliasing artifacts
in point sampling systems.  It discusses using a Monte Carlo method of
randomizing your samples to substitute aliasing artifacts for noise which is
less objectionable to the eye.  It also discusses the pros and cons of
analytic methods (filtering) versus discrete (sampling) methods.  It should
have the information you need.

Greg Galloway
Georgia Institute of Technology
gg10@prism.gatech.edu