Concerning David Bourgin's colorspace-faq

In 1994 and 1994, David Bourgin (now deceased) posted to the Internet a document "colorspace-faq." It contained many errors. Bourgin repeatedly refused to make corrections. Eventually, he agreed to withdraw his entire document, and have it replaced as an Internet FAQ by my GammaFAQ and ColorFAQ documents.

This note is a sequence of quotes from David Bourgin's "colorspace-faq". Each quote is accompanied by an explanation of why that quote is incorrect. The entries are written in the first person, exactly as I e-mailed them to Bourgin.

Many copies of Bourgin's documents are archived on the Internet. I hope that the documentation of the errors in the Bourgin document will encourage the holders of obsolete copies to withdraw or replace them.

Timo Autiokari periodically declares among several Internet newsgroups that the Bourgin document is superior to my GammaFAQ and ColorFAQ. Bourgin is dead; he cannot make any declaration on this matter. If you are tempted to believe Timo, then I suggest that you read the information here, in my GammaFAQ and ColorFAQ, and my page Concerning Timo Autiokari.

Dear David,

In your colorspace-faq, you write,

A color is simply a specific Spectral Power Distribution (SPD), the radient ...

The first substantive statement in your post is completely wrong. A colour is NOT a specific SPD. An SPD maps to a colour, but it is a many-to-one map. A given colour can be produced by many SPDs: no specific SPD corresponds to a colour.

... rod cells (which are not sensitive to color) ...

The rod cells certainly have a particular spectral sensitivity. If you want to know the colour implication of the difference between the rod sensitivities and cone sensitivities, you should read about the Purkinje Effect.

For real colors ...

What other kinds of colour are there?

... these attributes have been defined by Hunt, from the CIE recommendations, ...

You give CIE's recommendations. You may have taken them from Hunt, but they weren't defined by him.

[Brightness] ... You can blurry or enhance an image by modifying this component.

Blurring is conventionally accomplished by filtering all three primary components, not just brightness.

The previous sentence is appended to a quote from CIE recommendations, but this sentence isn't. It needs to be made clear what's the quote from Hunt (and/or CIE) and what isn't.

... a color is a visual sensation produced by a stimulus which is a specific SPD.

This is absolutely incorrect. It's not a specific SPD.

So, a color space is a mathematical representation of our perceptions.

No, this is incorrect. The CIE has standardized a system whereby physical analysis of light can be described numerically as a colour. Knowledge of perception underlies the definition of the system, but the colour does not represent perception.

It's useful to think so because computers are in fond of numbers and equations

The CIE did their seminal work long before computers.

Some color spaces are perceptually linear ...

This is in conflict with the terminology of the CIE, which is very careful to distinguish between the term "linear", applied to physical quantities, and "uniform", applied to perception. There is no way to directly measure the "linearity" of a perceptual system, so we don't use that word.

A color gamut is the boundary of the color space.

No, it's the extent of colour reproduceable by a particular device or medium.

... YIQ and YUV analogue ... (NTSC and PAL) and YCbCr digital). ... They separate luminance from chrominance ...

Please read my ColorFAQ. Video does NOT use "luminance".

SML (Short Medium Long) A perceptual color space based on the response functions of the cones in the retina of the eye. Good for psychometric research.

This does not qualify as a "color space" because no researcher expresses colour as numerical quantities using these bases.

... CIELuv and CIELab.

This is incorrect nomenclature. Use either CIE L*u*v*, the formal nomenclature, or CIELUV if you want something easier to type. Similarly for CIE L*a*b* and CIELAB.

... They are near linear ...

No, they're not. They are intended to be fairly perceptually uniform.

"Linear" refers to a system like CIE XYZ or linear tristimulus (e.g. RGB) that satisfy the principle of superposition, f(a+b)=f(a)+f(b). Such is not the case for LAB and LUV. This is a elementary mathematics, first year college.

... (unless your in the habit of swapping your eye balls with aliens) ...

Eyeballs have to do with spectral sensitivity functions, not the CIE L*u*v* and CIE L*a*b* formulations.

... CIELuv you can derive CIELhs or CIELhc ...

The CIE does not define either "CIELhs" or "CIELhc".

E.g. RGB -> CIE XYZccir601-1 (D65) provides the following matrix of numbers ...

Here you introduce a matrix equation before making any mention of the linearity of the quantities that you're going to compute. To understand color in computing or video you must first understand intensity reproduction, the subject of my GammaFAQ. David, please read it.

You say CIE XYZccir601-1 and give a matrix from RGB to XYZ. But the CCIR 601-1 document, now renamed ITU-R BT.601-1, specifies no chromaticity, and no such matrix "is provided by" that standard! In practice, Rec. 601 is used with two different sets of chromaticities, that used in North America and standardized by SMPTE, and that used in Europe and standardized by the European Broadcastng Union (EBU). Please consult my ColorFAQ for details.

All of the pictures don't use the full color space. That's why we often use another scheme to improve the encoding of the picture (especially to get a file which takes less space). ...

Most pictures are degraded, not improved, by improved by coding into pseudocolor. The pictures that are not degraded take the same space they did to begin with.

4 - What is this gamma component? Many image processing operations ... linear luminance domain.

An RGB system is properly described as "tristimulus" domain, or more loosely "linear light". It is decidedly not "linear luminance". Luminance is only one component, not three, and luminance is by definition linear.

In most cases the CRT will have a non-linear response.

In ALL cases the CRT will have a nonlinear response.

... gamma, somewhere between 2.2 [NTSC and SMPTE specifications] and 2.8 [as given by Hunt and Sproson]. ...

You omit mention of the subjective effect that led the NTSC to choose this value. This is very imprtant in computing, as explained in my GammaFAQ, because of ambient (viewing) conditions.

... (Richard Salmon and Alan Roberts) have shown that the actual value of gamma is very dependent upon the accurate setting of the CRTs black level.

This is not the result of Salmon and Roberts' work; it has been well known fifty years or so.

New_Voltage = Old_Voltage ^ (1/gamma)

No, the input qualtity is not voltage, it's intensity.

Skipping forward to the very end, in PhotoYCC,

* as a note -> do the phosphors need to match with regard to their chromaticities or is a spectral match required?

Well even though it has the letter "Q" an FAQ usually provides answers, not questions. But this question should be answered by the first statement in the FAQ.

a) this results in RGB values from 0 to 346 (instead of the more usual 0 to 255) if this is simply ignored the resulting clipping will cause severe loss of highlight information in the displayed image

The entire purpose in the PhotoYCC scaling is to PRESERVE highlights, not to clip. Whoever wrote this has a poor understanding of the way PhotoYCC works. Reference white is coded to 189 or so -- higher values, from there to 255, are devoted to conveying highlight information.

OK. Enough.

See also: Timo's technical argument - Timo and linear coding - Linear and nonlinear coding - Gamma FAQ

Charles Poynton
Copyright © 1998-07-27