Path: utzoo!attcan!uunet!lll-winken!lll-tis!helios.ee.lbl.gov!pasteur!ucbvax!bloom-beacon!dawn.UUCP!stpeters From: stpeters@dawn.UUCP Newsgroups: comp.windows.x Subject: Re: Luminance from RGB (was "intensity" from RGB Message-ID: <8811080030.AA14681@EXPO.LCS.MIT.EDU> Date: 8 Nov 88 00:30:22 GMT Sender: daemon@bloom-beacon.MIT.EDU Organization: The Internet Lines: 100 Received: by ATHENA.MIT.EDU (5.45/4.7) id AA13335; Mon, 7 Nov 88 18:22:34 EST Received: from steinmetz.UUCP by uunet.UU.NET (5.59/1.14) with UUCP id AA18816; Mon, 7 Nov 88 18:20:39 EST From: steinmetz!dawn!stpeters@uunet.UU.NET Received: from dawn.steinmetz.GE.COM (dawn.ARPA) by kbsvax.steinmetz (1.2/1.1x Steinmetz) id AA10218; Mon, 7 Nov 88 17:58:04 est Received: by dawn.steinmetz.GE.COM (4.0/SMI-4.0) id AA23325; Mon, 7 Nov 88 17:57:55 EST Date: Mon, 7 Nov 88 17:57:55 EST Message-Id: <8811072257.AA23325@dawn.steinmetz.GE.COM> To: xpert@athena.mit.edu Cc: stpeters@dawn.steinmetz, poynton@sun.com Subject: Re: Luminance from RGB (was "intensity" from RGB Responding to my posting on RGB, Charles Poynton (poynton@sun.com) writes: > ... except that modulation onto an RF carrier for > transmission involves no inherent compromises beyond those already made in > formation of baseband NTSC. This isn't an appropriate forum for RF discussion, but one paragraph won't hurt. In baseband NTSC, the chroma signal is an oscillation about the luma level. The NTSC scaling of chroma relative to luma is such that when the composite signal is (AM) modulated onto RF, certain combinations of luma and chroma nominally correspond to more than 100% modulation. That's just one example: there *is* a substantial set of inherent compromises involved going to/from RF[1]. On RGB weighting: I argued that the NTSC intensity formula shouldn't be considered carved in stone for purposes beyond its original intent, because compatible color TV required identification of an RGB combination orthogonal to chroma, a luma that had to work with B&W televisions. Mr. Poynton correctly stated that the formula also describes the human spectral response. That the best RGB weightings for a luma orthogonal to chroma turned out to be those for the human spectral response should hardly surprise anyone. However, that has little relevance[2]. If you display RGB in the ratios matching the human spectral response, you do not get a good white, you get a bright but dreary grey. The logarithmic response of the eye masks the roughly 2:1 green/red ratio, but the 11% blue in the formula is just not enough. The CIE has a name that I can't remember, a qualified white of some sort, for this bright grey. The mixture lies in the large central region of whiteish shades in a triangular color plot, but it is well removed from what people subjectively perceive as a clean, pure white. This has been known and exploited for years. Fabric whiteners work by adding blue; they're essentially blue dyes. High-grade paper is tinted with fluorescent blue dyes to make it appear whiter; people appearing on color TV have long been advised that a blue-tinted shirt looks whiter than a white one (a problem aggravated by the low level of blue in the spectrum of studio lighting). The point is that what people perceive as pure white is not, in TV terms, a zero-chroma color. This is not unnatural: there is no reason that we should have evolved with our visual systems adjusted so that the mixture we see as white matches the weightings determined by the spectral response of our eyes. After all, the illumination spectrum in which we see things is not constant over frequency - nor is the variation constant over time. (E.g., there is more blue light on clear days than overcast ones, more blue at noon than near sunset or dawn, more blue on clear days near the equator than on clear days at higher latitudes, etc.) To restate the point, not only is there no reason we should have evolved to see NTSC luma as pure white, it is known that we did not. [I'm rather fond of summer scenes of bright white puffy clouds in a blue sky. If nature had adjusted my visual system to perceive the NTSC mixture as pure white, the clouds would seem to have a blue tint. It wouldn't be the same.] Not long from now, 24-bit systems will be as common as 8-bit ones are today, and people will want to display true-color images. If you display the NTSC RGB combination and adjust your monitor so it looks like a nice clean white, you will have problems with true-color images. If you make equal-RGB look white, your true-color images will look true-color. When you have RGB on separate cables with equal bandwidths, there is nothing relevant about the NTSC RGB combination. However, you will frequently want white window borders, background/foreground color, etc. The white you want is ((R+G+B)/3). Equal weighting and the eye's logarithmic response allow considerable monitor misadjustment to still make your white look reasonably white. -- Dick St.Peters GE Corporate R&D, Schenectady, NY stpeters@ge-crd.arpa uunet!steinmetz!stpeters 1. I'm willing to discuss RF issues further offline. 2. The NTSC/human-vision weighting is a reasonable one if you want to extract a grey-scale image from a true-color image.