Jump to content

Talk:Lab color space/Archive 1

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia
Archive 1


Ancient move proposal

This page needs to be moved to Lab color model. --Gutza 13:42, 5 Mar 2004 (UTC)

Changed my mind. I'll just make a link. :) --Gutza 16:00, 5 Mar 2004 (UTC)

The conversions might not be accurate, instead of superscripts ("power of") there is sometimes just a factor. Someone who knows this stuff should double-check the correctness. A friend of mine working in the field just helped me get the correct potency operators in. The formulas could be correct now :-) --Mcornils 23:05, 15 Oct 2004 (UTC)

CIE Lab definitions changed?

The description of CIE 15:2004 Colorimetry, 3rd Edition http://www.cie.co.at/publ/abst/15-2004.html stated that there are even changes to the equations defining the parameters of the CIELAB colour space! What changes were those, and are the CIE 1976 equations now out of date? --Nantonos 17:06, 20 February 2007 (UTC)

I guess the answer is here (quote adapted by me for Wikipedia-style terms):

A Continuity Study of the CIE L* Function, Bruce Lindbloom, circa 2003

With constants expressed as decimal values (0.008856 and 7.787), not only is the function f(t) discontinuous, it is also non-monotonic, which makes it not invertable in "junction" region. Solving the equations shows that the values may be exactly represented as rational numbers: 216/24389 instead of 0.008856 and (24389/27)/116 instead of 7.787.

Note: I was notified by Dr. János Schanda and Todd Newman on 3 April 2003 that the CIE 15.316 and the CIE Standard on CIELAB will implement this fix (i.e. using rational rather than decimal values for these constants).

Thus, as the revision of L*a*b* specification can't be "revolutionary" (otherwise it would be a brand-new space), this is the most probable change. And, of course, as the function f(t) is used for calculating all three coordinates, the change should cover them all, not just the L* as one may think after reading the article by Bruce Lindbloom.
But, as the difference in values is very-very small (0.005% for 0.008856 and 0.0005% for 7.787), the change can be considered as practically insignificant in computer applications. Just try yourself some examples and see that no panic is required for users of 1976 formulae.
213.234.235.82 —Preceding signed but undated comment was added at 16:15, 12 October 2007 (UTC)
By the way, this Wiki article features the constant δ = 6/29, from which other constants a and t0 are derived, where a is κ×116 in Lindbloom's article and t0 is ε. So it looks like CIE originally used these constants to define the functions, but simply rounded them later to decimal values with 4 significant digits — for ease of manual calculation. So the problem is not with the concept itself, but with specifying constants as roughly-rounded decimals. That is the curse of official specifications: you ought to follow them if you want to make a standards-compliant product, even when those standards fall behind the today's technologies and computational capabilites. 213.234.235.82 16:35, 12 October 2007 (UTC)

pictures

The pictures could use numbers along each axis.

The pictures could have some sort of outline around the area which is accurate for sRGB displays. The rest probably should have markings over it to indicate that it is pretty much junk. Also draw the line showing monochromatic colors.

It would be good to choose intensity levels that show off some corners of the sRGB space, particularly the yellow and blue corners. Maybe also do an intensity level at a linear 50% grey.

AlbertCahalan 03:14, 26 Jun 2005 (UTC)

Broken images and links on this page: http://en.wikipedia.org/wiki/Lab_color_space

One is: http://upload.wikimedia.org/wikipedia/en/thumb/9/99/Lab_color_at_luminance_25%25.png/180px-Lab_color_at_luminance_25%25.png

Two is: http://upload.wikimedia.org/wikipedia/en/thumb/0/04/Lab_color_at_luminance_75%25.png/180px-Lab_color_at_luminance_75%25.png

Wondering

I'm wondering--are images made with this color system unaffected by differences in gamma correction or otherwise screen profiles?

--Michiel Sikma 2 July 2005 17:01 (UTC)

Well, the image isn't affected by what you do to your system. Its color values continue to represent the same color. Of course, if you no longer have an accurate color mapping for your monitor, it won't look the same... Notinasnaid 2 July 2005 17:44 (UTC)

In other words: the RGB values sent to your display to represent the LAB values must change if you change the profile of your display. In yet other words: To display the LAB values, software has to convert from LAB to the colorspace of your display. Changing that colorspace changes the values. Now, with perfect displays and a lot of precision, this should not be visually noticeable, because if the profile accurately describes the display, then the resulting visual impression would be the same. But there are no perfect displays, and most of them still use 8 bits/channel - so quantization effects may cause small visual differences.


hi ,
I was wondering if some body can post information on what is the formula for evalaluating normalized color distance (NCD) between two images?
Thank you,

Hari prasad 07:36, 27 December 2005 (UTC)

I believe See also MacAdam Ellipse.

What is data range of a* and b* ?

The article states that L ranges from 0 to 100. However, there is no mention of the data ranges for a* and b*. Perhaps this is undefined, but if this true, then this fact should at least be discussed

This is a good point, and it needs to be fixed. The assumption is that no stimulus exceeds the white point. In other words, if [1,1,1] is the white point, then X, Y, and Z of any stimulus are all less than or equal to one. For L*, you can see that its maximum is at Y=1, in which case f(Y)=1 and L=116-16=100. The minimum is at Y=0 and L=116*(16/116)-16=0. If you do that to a* you get a maximum at 500*(1-16/116) and a minimum at 500*(16/116-1). Same idea for b*. This should be checked first, I will see if I can find a reference. PAR 04:37, 24 January 2006 (UTC)


a* and b* measured values are only bounded by the realizeable colors with highest saturation: monochromatic sources (the XYZ values along the spectral locus). But in theory a* and b* are unbounded, and many color transformations can result in values that represent colors that cannot be produced in reality (or at least that humans can't see). Later calculations usually bring those values back to reality.—The preceding unsigned comment was added by 24.23.185.56 (talkcontribs).

Flourescent pigments (aka Dayglo) Might have an exceptionally large a or b value. The pigments function by catching uv light and transforming it into visible spectrum. Therefore the color returned may be larger what is present in the incident light. —The preceding unsigned comment was added by 166.102.189.10 (talkcontribs).

The a* and b* values depend on the brightness and saturation of a color relative to the white point, and are unbounded unless you bound the range of brightnesses. The Dayglo comment is correct if you consider the range of values that can achieved by illuminating a subject, relative to illuminating a white; in that case, it is possible to get a brighter more saturated color than with simple reflectance. But if you put some color LEDs into the scene, you can get arbitrarily higher a* and b* values, as well. Monochromatic sources will give bigger values than other sources for a given total luminance or randiance, I think, but there's not sensible bound that you can get by looking at the monochromatic case as special. Just as not all non-negative XYZ triples correspond to real spectra or colors, Lab space is also limited; the limit is very definite, since it's 1:1 with XYZ (given a white reference); but many of those Lab or XYZ triple that DO correspond to real colors don't correspond to any reflectance spectrum, by virtue of being too bright; it's possible to compute the envelope of Lab values that do correspond to real reflectance spectra; I leave that as an exercise for the reader. Dicklyon 17:07, 3 July 2007 (UTC)

Hunter Lab

The article seems to have been rewritten to emphasise that Lab means a space I hadn't heard of before called "Hunter Lab". I can't agree that this should be stated as unambigiously the case. In particular, people will encounter "Lab" in image editing and other software and it is likely to mean "CIELab". Even if some standards body has agreed that "Lab" must mean "Hunter lab", the article must deal with popular usage even if it notes that it is not recognised in scientific circles.

I am also forced to conclude that while Hunter Lab might have come first it is not nearly so well known. It is not mentioned in Hunt's Reproduction of Colour, for example.

So to start with I am removing this:

Without further qualification, "Lab" color space refers to that of Hunter (Richard S Hunter, JOSA, 38, p 661 (1948)), which is an Adams Chromatic Valance Space. It is not proper to refer to CIELAB as simply "Lab," not just because it is not an Adams Chromatic Valance Space, but also because it is ambiguous and confusing.

I will also rearrange the article to introduce both, then present CIELab first. It would be welcome to include specific references and discussions of "Lab" as a contraction, and what the writer of that reference considers correct, but it isn't up to Wikipedia to tell everyone with Photoshop that they are, unqualified, wrong. Notinasnaid 08:00, 21 June 2006 (UTC) I've done the rebalancing, and added more introductory material. I have to say, I've reread the piece I removed several times, and I still can't understand what is meant by "It is not proper to refer to CIELAB as simply "Lab" ," ... because it is not an Adams Chromatic Valance Space..." " Why not? What bearing does this have on an abbreviation? Looking forward to learning...! Notinasnaid 09:13, 21 June 2006 (UTC)

I've also added a section "Which Lab?" which guides the reader in understanding how to interpret "Lab" in particular cases. There are no cases where Lab means "Hunter lab". This is not deliberate exclusion: I just don't know of any. If after a while none are found, I think the article will have to be rewritten to more strongly associate Lab with CIELab. Notinasnaid 09:16, 21 June 2006 (UTC)

Opponent process

This color space seems strongly related to Opponent process. CIELAB primaries seem to be the same as the opponent colors in Opponent process color theory. Shouldn't both articles mention each other somehow?

  • Sorry to say, I don't see anything indicating that they have the same primaries. I'm not sure CIELAB can even be said to have three primaries since L is a luminance channel. Do you have a source for this? Notinasnaid 17:15, 24 September 2006 (UTC)
    • What I meant to say is that CIELAB is based on the same principle as the opponent process theory: the eye converts tristimulus values into brightness and two color difference channels, which is what CIELAB also does. Also, while the color difference channels may not be exactly the same colorimetrically, they are very similar: red-green and yellow-blue.
    Opponent Recording of Tristimulus Outputs
    Opponent Process Color Wheel
    CIELAB Color Wheel
    Also, here's a quote: "CIELAB can justifiably be called an opponent color space, because its chromatic dimensions are defined as the contrast between specific opposing hues, analogous of the opponent processes hypothesized by Hering. In fact, the X–Y and Y–Z contrast dimensions have a very similar shape and peak wavelengths to the Hurvich and Jameson opponent functions." [1]

LAB question

Hey, I realize that this is not exactly the forum for this question, but I am looking for a relatively obscure answer that no one has been able to help me with so far. I'm trying to make stimulus for a psych experiment using CIE space, so I made it in photoshop with LAB color, a CIE space, but I can't save it as anything but a Tiff which most other applications won't load. I can save it as an 8-bit LAB color tiff, which will sort-of load, or a 16-bit RGB color .PNG file, which will load. The problem is that the two look very different, and I dont know which is closer to the true 16-bit LAB color. Any ideas? thanks and sorry for posting a somewhat irrelevant question here.

If you are viewing on a color calibrated monitor, and have a correct RGB profile for the monitor, then converting to RGB should not change the color. If you save with an embedded sRGB profile, and view on a calibrated monitor, the colour should be in some sense "close" to what is expected. If you don't have calibrated monitors, it isn't even worth starting. Notinasnaid 08:16, 19 October 2006 (UTC)
thank you very much, I'll look into calibration for RGB monitors

Also, most applications aren't color managed at all. Even if you have calibrated your display, they may assume sRGB, or may just throw up totally uncalibrated values (having found formulae they didn't understand on some random website and used them as-is.).

Lab demo images are just not correct!

The color images shown in the article, and described as CIELAB constant luminance, are obviously not of a constant luminance! Do a simple visual inspection, or pull out any color meter you have, and compare the L* value of the dark green (top center) to the L* value of the light blue (bottom left) of any of the example pictures. You will see a huge variance.

Just imagine seeing the image in grayscale. If the L* in one of the images was actually constant (eg. 25%), and if you showed that image on black and white TV, the grey level would be approximately constant across the a*-b* dimensions. As it is now, you would see quite different grays.

Thanks for looking into this. 72.132.235.222 06:28, 24 October 2006 (UTC)John

There is a strong argument against any visual representation that doesn't claim to depict sRGB. The process of converting Lab to sRGB, and then to your perhaps (or perhaps not) calibrated monitor may not preserve luminance. Unfortunately, the graphic does not include the technical details of how it was made so that we could check up on it. Nevertheless, we can do some rough tests. Opening in Photoshop 7, using the embedded profile, and converting to L*a*b* shows a wide variation in color. It may well be worthwhile regenerating the graphics. If you do, please include full details of the software used, color settings, profiles etc. – effectively allowing the experiment to be reproduced and the results checked. Notinasnaid 07:25, 24 October 2006 (UTC)

I suspect the problem is that (sRGB) gamma has not been taken into account in the preparation of the images, but I'm not in a position to calculate them myself at the moment. As a quick check you could try applying a gamma of 2.2 using an image-editing program then see if the resulting luminance is more uniform (though of course you might see banding artifacts). 57.66.65.38 18:07, 16 November 2006 (UTC)Andrew http://www.techmind.org/colour/


The other problem is that some of the values toward the edges represent colors that are not realizeable in nature, and not visible to humans. Those values get clipped or gamut mapped into the visible colors by whatever process converted the L*a*b* values to RGB. Even in Photoshop, you can't create a map like that and get perfect luminance uniformity (try it in their color picker by clicking on the L radio button).

yes, non-realizable (in sRGB) colors should be shown as middle grey, or similar, to avoid giving people the wrong idea. --jacobolus (t) 11:15, 16 April 2007 (UTC)
Okay, I've created an image which should fix this problem! --jacobolus (t) 22:52, 2 May 2007 (UTC)

CIE 1976 (L*a*b*) conversion

Could someone help providing the CIE 1976 (L*a*b*) self-luminous and surface color specs for 620nm Red?

You can easily convert the 620nm into XYZ. Add a white point definition, then you can convert to RGB, CMYK, L*a*b*, etc.

Question regarding the CIE 1976 (L*a*b*) color scale

I was wondering how this system is able to treat value (lightness to darkness) as an equal partner in the determination of a specific color with the actual colors in the system (magenta/green and blue/yellow). Accorind to the Munsell color wheel, which is in the form of a cylinder, with the circumference of the cylinder being the hue, the radius being the chroma and the height of the cylinder being the value, lightness to darkness (i.e. value) is independant of the color (hue) or saturation of color (chroma), and thus, one can travel a distance of "x" (x being the radius of the cylinder) from the vertical axis at any height on the cylinder (i.e. at any value, one can have a saturation (chroma) level of "x". However, on the CIE color scale, the farther away from the equator (i.e., closer to the N + S poles of the 3D sphere), the distance one may travel from the vertical axis decreases at the lines of longitude head towards the poles. In other words, horizontal sections of the sphere, let's say, taken at 25% lightness, 50% lightness and 75% lightness don't give congruent circles. This is because, unlike Munsell's color wheel, value (lghtness vs. darkness) is treated just like any other color (i.e. green or blue) and not as an independant entity known as value. If anyone can please respond to this concern, please email me at shatnes551@yahoo.com, or my Wikipedia name is User:DRosenbach.


L* is as independent as they could make at the time. It is pretty well independent of a* and b* changes for all visible colors (outside the visible colors, who knows?). The small differences lead to improved color difference formula that warped the L*a*b* space a bit: DeltaE94, CMC, CIEDE2000, etc.

More recent work has gotten better colorspaces with more independence and better matches to human vision: CIECAM97s and it's successor CIECAM02. There were many other attempts along the way: RLab, LLab, etc. You can find historical details in Hunt's "Reproduction of Colour". However, to get the better match, these colorspaces/models require more information about the illumination, surrounding colors, etc.

Is Lab really intuitive?

The article claims: Since the Lab model is a three dimensional model, it can only be represented properly in a three dimensional space. A useful feature of the model however is that the first parameter is extremely intuitive: changing its value is like changing the brightness setting in a TV set. Therefore only a few representations of some horizontal "slices" in the model are enough to conceptually visualize the whole gamut, assuming that the luminance would be represented on the vertical axis.

However, changing the L* value is *NOT* the same as changing the brightness setting in a TV set. Lab is a luminance-chrominance color space, as opposed to a luminance-chromaticity color space. There's a subtle and perhaps unintuitive difference between chrominance and luminance. In a luminance-chrominance color space, the following would happen: Suppose you had a camera, and you put a neutral density (light blocking) filter in front the camera. Suppose the luminance is halved. In a luminance-chrominance color space like Lab, the chrominance will be halved. In a luminance-chromaticity color space (HSL and HSV behave like this), the chromaticity will stay constant. The article [2] explains this difference in more detail.

A- This behaviour of Lab color space is pretty unintuitive to me. B- Turning down the brightness on a television behaves much like placing a neutral density gel in front of it; it does not behave like halving L*.

While it would be intuitive that Lab is an intuitive color space, I don't think it really is. And while it would make sense that turning down the brightness on a television wouldn't affect chrominance, it does. I think I've made this mistake before. Unintuitive? Yes.

Glennchan 03:37, 30 November 2006 (UTC)


A neutral density filter linearly scales XYZ values. But L* is nonlinearly related to XYZ values and is linearly related to perceived lightness.

If I take a white piece of paper and have someone compare it with and without a neutral density filter with 50% transmission - they will NOT say that it is half as light, because their perception is not linear. If I do the same experiement with a neutral density filter with 18% transmission - THEN they will say it is half as light. And if I display L* = 100 and L* = 50, they will say that L*=50 is half as light, because L* is a good match to human vision.

And matching perception is far more intuitive than trying to match the behavior of light because people are very familiar with their own perception, and not so familar with how it maps to actual photons.

You're right- I forget about the non-linearity in L*. Do you have a Wikipedia account by the way? I think you'd make some good contributions to the Wikipedia. Glennchan 03:36, 2 December 2006 (UTC)
I find L*a*b* quite intuitive for dealing with photographs. Modifying a* and b* in different portions of an image is an incredibly powerful and easy way to remove (or add) color casts, and modifying L* of course allows change of lightness contrast without altering chromaticity, which is otherwise hard with a Curves type tool. I personally feel that Curves adjustment layers in L*a*b* mode, along with layer masks, is the most powerful tool in Photoshop. --jacobolus (t) 11:13, 16 April 2007 (UTC)

Chromatic Value vs Chromatic Valance Spaces

CIELAB is an Adams Chromatic VALUE space. (Hunter) Lab is an Adams Chromatic VALANCE space. Would whoever keeps changing the classification of CIELAB to Adams Chromatic VALANCE kindly knock it off? Thank you in advance.

Lovibond 01:39, 9 February 2007 (UTC)

Can you provide a source? This would be the best way for editors to know what is right, and in the long term the entire article must be sourced. Notinasnaid 08:54, 9 February 2007 (UTC)

Can anybody describe the difference between Chromatic Value and Chromatic Valance Spaces? It seems to me that this is identical to chrominance and chromaticity spaces (as defined in Color_Models.pdf and Chromaticity_Chrominance_DAK.pdf - though i don't know if this is reliable source). I would therefore suggest using the terms "chromaticity-valance" and "chrominance-value space". For the orgins and explanation of the terms one can then refer to the mentioned artikels and the wiki artikels Adams Chromatic Valance and Elliot Quincy Adams. Also: chrominance and chromaticity are in wiki defined in Chrominance and Chromaticity. It should be clear in colorwiki if this equals the formentioned definitions or not. BartYgor 10:52, 16 July 2007 (UTC)

Chromatic value spaces are based on a model of color vision consisting of three types of receptors, a non-linear transformation applied to each, and an opponent mechanism. An example is the CIELAB color space, in which the three receptors are assumed to have sensitivities which match the CIE XYZ color matching functions (yes, this is a bit naïve), the non-linearities are basically the Ladd-Pinney cube root transformation, and the opponent mechanism causes the transformed Y signal to be subtracted from the transformed X signal to form one opponent channel (a*), and the transformed Z signal to be subtracted from the transformed Y signal to form the other (b*).
Chromatic valence spaces, on the other hand, do not appear to be based on any plausible model of color vision, but on the definition of saturation as chroma relative to lightness. First, lightness is computed. Then, uniform chromaticities are multiplied by the lightness to arrive at two chrominance channels. The neurological wiring needed to do this would be intractible. Examples of chromatic valence spaces are Hunter's Lab, in which Lightness is computed from Y using Priest's square root, and the uniform chromaticity scale is basically as suggested by Adams in 1942, with some centering: and ; and CIELUV, in which Lightness is computed as in CIELAB, and the uniform chromaticity space is based on MacAdam's 1960's suggestion, centered and with the yellow-blue axis stretched 50%.
Hope this helps. Lovibond 16:33, 8 August 2007 (UTC)
Not really. So you nawadays as CIELab has been replaced by more accurate CIE models it is no longer a value space? Surely that can not be true. Or even: RGB is en non-linear transformation of CIE XYZ so it must be a valence space? And I don't see the conceptual difference between CIELab en CIEluv both start from CIEXYZ and try to capture our uniform experience of color. And yes CIEluv defines uv in terms of lightness (L) but as a, b and L in CIElab are also interdependent we can say the same of CIElab.BartYgor 16:06, 9 August 2007 (UTC)
The fact that other models may supplant CIELAB (note correct capitalization) has nothing to do with its mathematical pedigree, just as the adoptation by the CIE of the 1964 U*V*W* space, and, later, CIELUV, did nothing to change Hunter's Lab space as a Chromatic Valence space. The adoptation of the newer models did not change the definition of Hunter's Lab space, did it? RGB does not qualify as a chromatic valence space, because it is not based on a perceptually uniform lightness scale and a perceptually uniform chromaticity diagram. Nor does it qualify as a chromatic value space, because, even though there are non-linearities, none of the coordinates are good approximations of lightness, nor are there opponent mechanisms accounted for. As far as the conceptual differences between CIELAB and CIELUV, please refer to their definitions. The first is defined in terms of non-linear transformations of surrogate cone responses followed by opponent modelling (which maes it an Adams Chromatic Value space); the second interms of a uniform Lightness scale and a uniform chromaticity scale, with the chrominance coordinates determined by multiplying the chromaticities by Lightness (making it an Adams Chromatic Valence space). Lovibond 19:48, 9 August 2007 (UTC)
I, too, wonder what's behind this distinction. Lovibond, can you direct us to sources about these concepts? Dicklyon 16:21, 9 August 2007 (UTC)
The primary reference is Adams's 1942 JOSA paper, which is cited in the article to which this discussion pertains. A secondary source is Dorothy Nickerson, Munsell renotations used to study color space of Hunter and of Adams. Journal of the Optical Society of America, 40:2:85 (1950). Something which also explains this is Hunter's book, The Measurement of Appearance. Lovibond 19:52, 9 August 2007 (UTC)
Thanks, I should have looked. Is there a accessible copy any place you know of? Dicklyon 19:54, 9 August 2007 (UTC)
Thanks, so in short a chromatic value space is perceptually uniform in chromaticity and lightness. A chromatic valence space has some lightness yet is not perceptual uniform. So there can be only one chromatic value space (e.g. CIELab or it's newer versions); strange that one would talk about chromatic value spaces when there's only one. Now to answer my own question: chromaticity spaces are something cdifferent than valance or value spacesBartYgor 17:03, 14 August 2007 (UTC)
I don't think that's the right summary. CIELAB is certainly not perceptually uniform nor unique in its attempt. Dicklyon 17:10, 14 August 2007 (UTC)
I have to disagree with BartYgor and agree with Dicklyon. I've re-read my discussion several times, and nowhere did I say or suggest that Chromatic Value spaces were more/less perceptually uniform than Chromatic Valence spaces. I may have given the impression that CIELAB is solidly perceptually uniform, for which I apologize. Both CIELAB and CIELUV are approximately perceptually uniform. Also, because the only Chromatic Value space I mentioned is CIELAB, that doesn't mean it's the only one, as Dicklyon has also pointed out. Another example is Ebner and Fairchild's IPT, which is also approximately perceptually uniform after scaling the axes. Some YCbCr spaces may also be thought of as Chromatic Value spaces.Lovibond 18:34, 15 August 2007 (UTC)
I must have interpreted things incorrect (my mistake). So a colorspace can be perceptual uniform without being based on a model of colorvision. But then I would think that you can't tell by the colorspace as such that its a chromatic value or valence space. You should look at how it came to be. This seems a bit odd in a scientific sence to me that you would name two things differently not because thay have some different property but because of their origin, of how they came to be. I'm also not really familiar with YCbCr spaces but it supprizes me that they would be based on "a model of color vision consisting of three types of receptors, a non-linear transformation applied to each, and an opponent mechanism", they don't seem anymore based on colorvision than RGB spaces, apart from where the axes are put: this is indeed trying to mimic human perception; but sureky that isn't enough otherwise HSV / HSB /... would all be value spaces where indeed they are simple transformation of RGB spaces.--BartYgor 00:20, 23 August 2007 (UTC)

Visualizing LAB space

Figure 4.9 on page 73 of the book Photoshop Lab Color: The Canyon Conundrum and Other Adventures in the Most Powerful Colorspace, ISBN 032135 illustrates an excellent way to visualize the LAB space. I wonder if this diagram, or something similar to it can be made available here. --Lbeaumont 21:49, 20 March 2007 (UTC)

You'll have to explain for those of us without that book. --jacobolus (t) 22:51, 2 May 2007 (UTC)

New image which only shows colors in sRGB space

Okay, I got tired of looking at the deeply misleading planes of "constant lightness" in the previous images. So I made a new image. Unfortunately, the svg (Image:Lab color space.svg) is not rendering correctly. So in the mean time, there is a png version (Image:Lab color space.png). Dark grey areas represent out-of-gamut colors. --jacobolus (t) 22:55, 2 May 2007 (UTC)

Origin of the square (or cubic) root

In the second paragraph, the articel states: "However, CIELAB is calculated using cube roots, and Hunter Lab is calculated using square roots." There is no explanation, however, as to the origin of these square of cubic roots. In other words, how did one came to a conclusion that Lightness is percived as , k=2 or 3 ? Nor there is an explanation as to why CIE chose to take square-root, instead of Hunter's (earlier) cubic-root.

Unfortunately, I don't know the answers to these questions, but I'll be glad if this info will make its way into the main article.

Hunt's book "The Reproduction of Colour" and Wyszecki and Stiles "Color Science" go into some of the history. It has a lot to do with the assumptions of surrounding light, adaptation, and how they matched various test data. [the preceding unsigned comment was added by user:24.23.185.56, 1 December 2006]
Actually, CIELAB uses a cube-root formula, developed by Charles Reilly of DuPont. Also, it appears that Hunter originally used a rational function approximation (order 2 in numerator, order 1 in denominator). Both were attempting to approximate the relationship between luminance and ten times Munsell Value. I think that's the main answer to your question. Why did one use one and the other use the other? Hunter needed greater simplicity; he needed to build the analog circuit to do the conversion. Reilly and his associates made a slide rule to do most of the calculations; they weren't selling an instrument so their space didn't put so high a premium on ease of use. See "Cube Root Color Space," Journal of the Optical Society of America, 48:10:736-740 (1958), and Dorothy Nickerson, "Munsell renotation used to study color space of Hunter and Adams," JOSA, 40:1:85 (1950). Also, please sign posts here with four tildes, like this: ~~~~ -- thanks! Lovibond 02:02, 5 May 2007 (UTC)
I think by far the most useful thing to explain this would be to show a graph showing Hunter L, CIE L*, Munsell Value, 2.2 γ curve, maybe square and cube root, etc on the horizontal axis, against Y on the vertical axis. Bruce Lindbloom sent me such an image showing Munsell, square and cube root, and CIE L*, in a low-resolution gif format. But I think such an image would be very useful for informing this article. Along with the offsets in these functions, CIE L*, Hunter L, and Munsell Value aren't all that far apart. The current explanation given in this article is I think a bit misleading about the extent of that difference. --jacobolus (t) 03:57, 5 May 2007 (UTC)
Four Lightness scales plotted as functions of normalized luminance. Munsell Renotation; [1] a fifth-order polynomial in Munsell Value; Hunter's rational function; [2] Square Root; [3] and Ladd and Pinney's modified cube root[4] are shown. Ladd and Pinney's modified cube root is truncated (no negative values appear).
Something like this? [image to right →]
--Lovibond 23:00, 5 May 2007 (UTC)

Yeah, I like that! :) Though actually, can those lines be made a bit smoother? They seem a bit wobbly right now. I take it the Ladd/Pinney is CIELAB L*? I think that graph would do a good job of showing that L*, Munsell value, and Hunter Lab L aren't that far apart. The current description (in the article), which just describes them as "square root" and "cube root" gives a misleading impression, I think. --jacobolus (t) 23:46, 5 May 2007 (UTC)

Glad you like the picture; sorry you don't like the wobbliness. I've traced the jaggedness to gnuplot's svg export facility, which, for reasons unbeknownst to myself, quantizes point locations extremely coarsely. If it really bothers you, you could request that the gnuplot developers address this bug; if and when they fix it, and I'm able to upgrade, I'll happily re-draw the plot. Lovibond 15:55, 1 June 2007 (UTC)

References

  1. ^ S M Newhall, D Nickerson, D B Judd, "Final report of the OSA subcommittee on spacing of the Munsell colors." Journal of the Optical Society of America, 33:385 (1943).
  2. ^ Dorothy Nickerson, "Munsell renotation used to study color space of Hunter and Adams." Journal of the Optical Society of America, 40:1:85 (1950).
  3. ^ I G Priest, K S Gibson, H J McNichols, An examination of the Munsell color system. I: spectral and total reflection and the Munsell scale of Value. Technical paper 167, United States Bureau of Standards, 1920.
  4. ^ J H Ladd and J E Pinney, "Empirical relationships with the Munsell Value scale." Proceedings of the Institute of Radio Engineers, 43:1137 (1955).

MacAdams or curves

Or CIELAB was based on the effort of turning the MacAdams ellipses into circles or it was based on the logarithmic response of the eye. Both would seem to me to be to much constraint on a 3D transformation. On the other hand both could be if the MacAdams colordifferences where only defind in a xy sectione of CIExy at Y=50 (xy of CIExy -> ab of CIELAB transformation) and Y of CIExy would be adjusted (because of the logarithmic response)into L of CIELAB. (Y->L). That would then be the maximum of three constraints. Which is true? BartYgor 13:34, 15 July 2007 (UTC)

I added a section to MacAdam Ellipse describing the extension of the ellipses to 3-dimensional ellipsoids in the full XYZ color space. The MacAdam ellipses were defined for a fixed luminance of 48 cd/m^2. They remain fairly constant in size and orientation (that is, their projection onto a constant-luminance subspace remains fairly constant) down to about 3 cd/m^2, where one of the receptors starts to drop out (I forget which) and a type of color blindness sets in. At even lower luminance, all cones drop out and vision becomes monochromatic. PAR 16:32, 16 July 2007 (UTC)

Forward transformation: delta cubed or squared?

As for now, the equation for f(t) looks like:

Earlier it was:

for
otherwise

I think there is an error in the new formula, because δ=6/29 is squared in the computaion of , not cubed. Simply put, the equotient in the new formula simply is not equal to the old 7.787. And, to my opinion, replacing 16/116 with 4/29 renders the formula less intuitive. —Preceding unsigned comment added by 213.234.235.82 (talk) 16:44, 18 April 2008 (UTC)

 Done It should be squared, of course. All the threes got to my head and I made a typo. How to express the fraction is a matter of preference; I have seen several variants. --Adoniscik(t, c) 19:59, 18 April 2008 (UTC)

Variability Determination

The background for this question is anlaytical chemistry. Using a laboratory instrument, based on the Hunter LAB model, the color measurements of suspended pigments are determined as L, a, b, and deCMC (a differential determination from a standardized point). How is the variability of L*a*b determined? If, given a population of ten measurements, how would the variablility of the color be presented and calculated? The determination of relative standard deviation on the deCMC value would magnify the individual variabilities of the L*a*b values, but the RSD on the indiviual values does not properly characterize the color.Felch (talk) 14:27, 4 June 2008 (UTC)

Lab vs. LAB

This wiki is the first time I have ever seen the A and B get dropped to lower case. What's up with that? —Preceding unsigned comment added by 76.185.6.18 (talk) 22:57, 19 October 2008 (UTC)

Are you referring to Hunter Lab or CIELAB? --Adoniscik(t, c) 23:02, 19 October 2008 (UTC)

Naming CIE L*a*b* as merely Lab makes it sounds too cheap

It's very annoying entering http://en.wikipedia.org/wiki/cielab and being redirected to http://en.wikipedia.org/wiki/lab instead of the reverse way - CIE L*a*b* sounds it is based on a serious research, instead of the Lab name makes it sound too cheap as it were an useless colourspace used for nothing. —Preceding unsigned comment added by 87.196.54.186 (talk) 20:31, 4 November 2008 (UTC)

The first link goes nowhere. Your proposed name does not reflect the fact that this article covers a color space distinct from CIELAB, called Hunter Lab. --Adoniscik(t, c) 21:37, 4 November 2008 (UTC)

Konica color space book

Konica Minolta "Precise color communication" [3] (198.175.166.202 (talk) 12:17, 17 March 2010 (UTC))

That seems like a great resource. It probably belongs in the external links of the article. –jacobolus (t) 23:44, 17 March 2010 (UTC)

In reverse transformation, why the f variable labels?

Not sure if whoever wrote up the reverse transformation on this page is still around, but I think it’s pretty confusing to have fx, fy, fz, given that in the forward transformation, the f function is something completely unrelated. I’d like to change these labels, because then we could write, e.g., X = Xnf-1(ϕx), where ϕx is whatever we decide to label the variable currently named fx: basically the reverse transformation uses the inverse of the same function that the forward transformation uses, but that isn’t completely clear given the way it’s currently written. This function, f, might be called the "kinked cube root" function, with its inverse the "kinked cube", or similar. –jacobolus (t) 16:16, 30 April 2010 (UTC)