Path: utzoo!utgpu!watmath!att!dptg!rutgers!iuvax!uxc.cso.uiuc.edu!tank!eecae!cps3xx!flynn
From: flynn@pixel.cps.msu.edu (Patrick J. Flynn)
Newsgroups: comp.graphics
Subject: Re: How to map 24-bit RGB to best EGA/VGA palette?
Keywords: RGB EGA VGA color
Message-ID: <4379@cps3xx.UUCP>
Date: 30 Aug 89 23:54:59 GMT
References: <126@vsserv.scri.fsu.edu> <3129@cbnewsm.ATT.COM> <7743@cbmvax.UUCP> <13319@well.UUCP> <586@celit.com> <4862@eos.UUCP> <13381@well.UUCP> <9464@venera.isi.edu>
Reply-To: flynn@pixel.cps.msu.edu (Patrick J. Flynn)
Organization: Michigan State University Pixel Bashing Laboratory
Lines: 39

In article <9464@venera.isi.edu> raveling@isi.edu (Paul Raveling) writes:
>In article <126@vsserv.scri.fsu.edu>, pepke@loligo (Eric Pepke) writes:
>> 
>> One could try other color spaces, such as HSV for example....
>
>	In the last few days I've been experimenting with 3D
>	renderings of images in RGB space.  Most look generally
>	like a plume around an axis running roughly from black
>	to white; some are fairly conical, most are flattened
>	to some extent, some show definite fan-shaped planes.  In many
>	the central axis runs from slightly on the blue side of
>	black to slightly on the red/yellow side of white -- that's
>	possibly due to lighting in images digitized from outdoor
>	photos, with relatively blue ambient light and relatively
>	red/yellow specular reflections.

Now *this* is interesting! A geometric look at color space. How do
the shapes of the plumes vary as one
 + changes the ambient lighting
 + changes the basis for color space (RGB to HSV, etc.) ?

On interesting, realistic images, do well-formed clusters appear in
the 3D color space?  If so, one *could* attempt to reduce the pallette
through the application of a traditional clustering algorithm (there
was an article in ACM TOMS on this last year; they came up with a
clustering alg. for color quantization and compared it to Heckbert's
median-cut and the K-means clustering method).  I say `could' with asterisks
because most clustering methods make quite a few assumptions about the
statistical properties of the data, and run for quite a while, and I
seriously doubt that the `clusters' in color space (if any) are
(say) multivariate Gaussian; I also doubt that people are willing to wait
hours (!) for a clustering algorithm to reduce their pallette.

However, it would be interesting to compare some clustering methods
(perhaps a representative squared-error method and a graph-theoretic
approach) against the algorithms arising from the graphics community.
Patrick Flynn -- flynn@cps.msu.edu -- FLYNN@MSUEGR.BITNET -- uunet!frith!flynn

"Ah don' want no Tea; it give me a headache." -- Pete Puma