Path: utzoo!attcan!uunet!husc6!purdue!decwrl!granite!allen
From: allen@granite.dec.com (Allen Akin)
Newsgroups: comp.graphics
Subject: Re: How do I handle this kind of model??
Message-ID: <229@granite.dec.com>
Date: 21 May 88 01:04:08 GMT
References: <1131@ndmath.UUCP>
Reply-To: allen@decwrl.dec.com (Allen Akin)
Organization: DEC Technology Development, Palo Alto, CA
Lines: 44
Keywords:

In article <1131@ndmath.UUCP> milo@ndmath.UUCP (Greg Corson) writes:
>
>I need to model various types of smooth surfaces such as rolling
>hills, car bodies, human faces etc...
>       ... The odd part is that the rendering routine will need 
>to look for sudden changes in the objects contours (or places
>where one object overlaps itself or another of the same color)
>and draw "feature lines" at those places.  For example, if you
>were modeling a piece of metal with a very gentle bend, only the
>outer edges would be drawn in as black feature lines....however
>if you had a piece of metal with a sharp 45 degree bend in it
>there would be a black like at the location of the bend.

(I've replied to Greg directly, but I'm posting this to see if
anyone has a better solution.)


I had a similar problem once.  Maybe the technique I used will be of
help.

I rendered the scene, buffering up a scanline at a time.  For each
pixel I recorded both the intensity (with or without shading, as you
prefer) and the surface normal.  In a postpass I then computed the
angles between adjacent normals (in both the X and Y directions) and
changed the pixel color if the normals diverged by more than a
threshold.  I also changed the pixel color if the distance from the eye
to the surface changed more than a threshold, even if the normals were
roughly the same.

The results were fairly pleasing, and didn't require much threshold
tuning.  Sharp object edges are displayed correctly, and false edges
(such as those generated by interpenetrating objects) are handled
automatically.

I used a ray tracer because I had analytically-defined objects and
extremely tight constraints on memory.  (In fact, I had so little memory
available that I couldn't tessellate the objects I was rendering.)  You
can use a similar approach with a scanline rendering algorithm, or
with Z buffering if you also have a place to store the normals for the
postpass.  If you use a ray tracer, be careful about your numerical
methods; if the intersection points aren't fairly accurate, you get
jagged edges.

Allen