Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!brutus.cs.uiuc.edu!usc!venera.isi.edu!raveling
From: raveling@isi.edu (Paul Raveling)
Newsgroups: comp.graphics
Subject: Re: How to map 24-bit RGB to best EGA/VGA palette?
Keywords: RGB EGA VGA color
Message-ID: <9485@venera.isi.edu>
Date: 1 Sep 89 17:07:08 GMT
References: <9473@venera.isi.edu> <4379@cps3xx.UUCP> <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>
Sender: news@venera.isi.edu
Reply-To: raveling@isi.edu (Paul Raveling)
Organization: Information Sciences Institute, Univ. of So. California
Lines: 31


> 	... A further refinement that I hope
> 	to try today is to add another input knob to control a
> 	popularity threshold.  By twiddling this knob it should be
> 	possible to get something like a time-multiplexed histogram
> 	(pixel frequency over the RGB domain).

	That variant went into code late yesterday and I've looked
	at 2 images.  The results are again "illuminating" and a bit
	surprising.

	Most of the RGB space volume occupied by image colors is at
	a VERY low density.  Shift from displaying all colors to
	those which occur in only 2 or more pictures, and the
	"color clouds" shrink dramatically.  Contine up to
	show only pixels with ~1/2 dozen and the "clouds" shrink
	to modest-sized spindles or small planes -- in fact they
	look something like shreds of stratus clouds.

	Beyond ~6 shrinkage slows, leaving a "hard core" of colors.
	This suggests an obvious question:  Would quantization benefit
	by discarding the very-low frequency colors before selecting
	the final set?  I think it would depend on how the quantization
	algorithm balances popularity and geometry (RGB space partitioning),
	but mean square error measurements would be the most likely
	to show a difference.  I'm not sure about color retention measures.


----------------
Paul Raveling
Raveling@isi.edu