🚀 go-pugleaf

> Why is this technique called "anti-aliasing"?

Any representation of a scene at a resolution too low to show all the
detail of the original (i.e. truly sharp edges on diagonal lines, which
no raster-scan device can do exactly) is in fact a representation of
infinitely many different scenes.  For example, a simplistic representation
of a sharp-edged diagonal line will have a stairstep appearance, and
could also represent a scene in which the edge really was a stairstep rather
than a straight line.  Hence "aliasing":  many scenes, one representation,
and no way to tell which scene just by looking at the representation.

The trouble is that the eye-brain combination will pick one of these scenes
when shown the representation, and it's not necessarily the one you want.
For example, that stairstep edge really will be seen as a stairstep unless
the steps are *really* small:  "jaggies".  For another example, small
contrasty details (e.g. streetlamps at a distance at night) will show or
not show depending on whether the sampling process happens to hit them
dead-on or not, which makes for ragged-looking images.  This problem gets
*much* worse if the image is moving, because then such details blink on
and off from one frame to the next, as the sampling grid moves:  "scin-
tillation".  Areas with lots of regularly-spaced small contrasty details
will show spurious patterns as the detail grid and the sampling grid
"beat" against each other:  "Moire patterns".  And so forth.

Note that trivial techniques like local averaging of the representation
will help jaggies somewhat but *won't* fix things like scintillation.
To do "anti-aliasing" properly, it is necessary to recognize that each
element of the representation represents a finite area of the scene, and
its color/brightness/whatever should reflect, to some degree, the color/etc.
of *everything* that is within that area of the scene, not just whatever
happens to be at the exact center of the area.  Doing this well isn't easy;
doing it well without consuming inordinate amounts of computing time is
extremely hard.  See any good graphics textbook for more discussion of the
basics; check out Siggraph proceedings for current research work.


Be warned that the above contains oversimplifications; the reality is
even worse.  Be warned also that I am not an expert in this area; consult
someone who *is* before spending lots of money or planning your PhD thesis.
-- 
				Henry Spencer @ U of Toronto Zoology
				{allegra,ihnp4,decvax,pyramid}!utzoo!henry