Path: utzoo!attcan!uunet!portal!cup.portal.com!Classic_-_Concepts From: Classic_-_Concepts@cup.portal.com Newsgroups: comp.graphics Subject: Re: Palette Optimization Message-ID: <12821@cup.portal.com> Date: 22 Dec 88 02:41:26 GMT References: <12745@cup.portal.com> <356@cs-spool.calgary.UUCP> Organization: The Portal System (TM) Lines: 18 More thoughts on reducing palettes, since I have done a great many color graphic images ... One of the problems with sampling across the spectrum in a palette reduction algorithm is that it ignores 'families of colors'. If you were creating an image of a sunset--your palette would be predominantly reds, yellows, blacks. If you imaged a china teapot (now who would want to do that!), the palette changes to whites, beiges and whatever the background is. Since image integrity is greater when you get as much mileage out of a limited palette as possible, it is preferable to maintain it. Thus, if you MUST do it algorithmically, it would be better, given a system with sufficient speed, to 'sample' the image palette BEFORE algorithmically selecting the 'best' colors. So, a possible scenario... Start from a 'database' (following the rationale in my previous post) including percep- tually distinct and pleasing colors, sample the image to be displayed to determine 'like' choices (i.e., family of colors), perform palette reduc- tion to related colors. Display image. Julie Petersen