Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!mit-eddie!uw-beaver!uoregon!tillamook!markv
From: markv@tillamook.uucp (Mark VandeWettering)
Newsgroups: comp.graphics
Subject: Re: ray tracing transparent images and color
Keywords: color transparent raytracing
Message-ID: <4806@uoregon.uoregon.edu>
Date: 2 Jun 89 18:42:15 GMT
References: <2809@scolex.sco.COM>
Sender: news@uoregon.uoregon.edu
Reply-To: markv@tillamook.UUCP (Mark VandeWettering)
Organization: University of Oregon CIS Dept.
Lines: 68

In article <2809@scolex.sco.COM> brianm@sco.COM (Brian Moffet) writes:
>Just a quick (hah!) question regarding percieved color.
>I was doing some thought experiments on how one would
>do a ray trace of a transparent object.   Ignore everything
>except tint of the object.  So, no refraction or reflection 
>involved.

	Aha.  Great (and reasonably complex) topic you have accidentally
	stumbled upon.  I recently have been thinking about precisely
	the problem you describe.

>I am going to work in a simple color scheme, where colors
>will be defined as scales of red-green-blue with 0 being
>lowest (ie off) and 16 being highest.
	
	Good ole RGB color scheme.....

>Given an object whose tint is { 2, 0, 4 } (r,g,b)  in front of
>a background whose solid color is { 14, 15, 10 }.
>So, a different color will be shown for the object due to the
>interaction of the transparent object and the solid background.
>I also assume that the color will tend toward black.

	There is actually no reason to believe that the color will tend
	toward black at all.   More on this later.

>Given this, I would say that I can take the two objects and combine them
>in the following equation:
>
>	Rfinal = Rback - Gobject - Bobject
>	Gfinal = GBack - Robject - Bobject
>	Bfinal = Bback - Gobject - Robject

	This is a strange set of equations, noticeably because you are
	using subtraction.  The operator you wish to use is
	multiplication.

	Imagine that your colors are in the range 0-1 (the math is
	easier).  Further imagine that you RGB are specific wavelengths
	of light.   The color you perceive is due to the relative
	contribution of each frequency to the overall spectrum.

	Your transparent object acts as a filter.  In order to determine
	the outgoing intensities after the filter, you MULTIPLY the
	filter at each frequency by the input frequency.

	For example,  a perfectly clear object would have RGB like 
	(1.0, 1.0, 1.0), and a red object would have (1.0, 0.0, 0.0).


	If we take your example, and normalize into the 0-1 range:

	{2, 0, 4} = (0.125, 0, .25)
	{14, 16, 10} = (0.875, .937, .625)

	output = (.1093, 0, .1562) = {2, 0, 2}

>Does this make sense?  Any good references?
	
	You method doesn't make sense, but that is because color is kind
	of hard to do properly.  The truth is, you can't really get away
	with just using RGB if you want a reasonably accurate color
	model.   Good Reference of the Day:  Roy Hall's new book,
	published by Springer Verlag.  The title... um.... help me out
	people, my copy is at home.  Something like "Color and
	Illumination Models in Computer Graphics".

Mark VandeWettering