Raw converter that supports CIELab?

[MirekElsner]

I noticed this soon after posting, but the histogram was the same after refresh.

[ablankertz]

Raw Therapee allows you to use _any_ RGB profile you want as working and output spaces. If you create a super wide color profile in PS using Color Settings and use that as your working and export profiles, you should be able to get what you want.

[Jim Kasson]

There is no practical reason. I just want to see what colors is my camera (after post processing) able to produce.

If you'll humor me, I'd like to back up a little bit. Cameras don't see colors. Color is a human-centric construction. Cameras see intensity in three spectral sets, the spectra being determined by the wavelength-by-wavelength multiplication of the sensitivity of the camera's sensor, the color filter array's spectral response, and the spectral response of the lens/filter. Raw processing converts the data in camera file to colors. The process is decidedly imperfect. The camera doesn't see spectra like a color-normal (or, actually, any) human. The camera responds to wavelengths to which we are insensitive. The camera produces RGB triples that are identical for spectral stimuli that produce different colors when viewed by a human. The camera produces RGB triples that are different for spectral stimuli that produce identical colors when viewed by a human.

You said "(after post processing)", so maybe you understand all that. However, since post processing can cover just about any operation, the answer to your question is dependent entirely on the color space of the raw processor. Unless the raw processor attempts to save you from yourself, you should be able to produce any triple in the gamut of the raw processor's color space. If that color space includes triples that represent stimuli invisible to humans, you should be able to produce those triples. However, I wouldn't call them "colors".

Jim

[digitaldog]

Well stated Jim! There's a lot more going on under the hood, cameras record what we can't see and are not colors, we can see colors it can't record. And ProPhoto RGB has numeric definitions that are not colors.

[Jim Kasson]

Mirek, if you're looking for a color space in which to examine the output of a raw converter, I suggest that CIEL*u*v* is more appropriate than CIEL*a*b*. Although Luv shares a deficiency of Lab -- you can represent stimuli that aren't visible colors -- at least with Luv there's a clear boundary -- the conical horseshoe -- that separates colors from abstract triples. Luv has the big advantage for your purposes that it can represent any visible color. In addition, Luv has the concept of saturation, which is defined as how much chroma is in a particular triple as a percentage of the maximum visible chroma at that luminance and hue angle. Also, we're used to looking at chromaticity diagrams, and a horizontal slice through a Luv solid is the u'v' chromaticity plane.

Unfortunately (at least I think it's unfortunate), Luv is not commonly used by photographers. There are historical reasons for this that no longer apply. In 1976, when Lab and Luv were standardized, the print community had used Lab for years, and didn't want to change. The video folks had been using Luv, and felt the same way. They couldn't compromise, and standardized both. You can see the history in the default white points: D50 for Lab, and D65 for Luv. Then along came Photoshop. Since it was aimed initially at the print community, it supported Lab. Twenty-five years later, when a photograph is as likely to find its final destination on a LCD screen as a printed page, Photoshop still doesn't support Luv. It's a pity.

This doesn't really have much to do with the original post, but I couldn't help myself.

Jim

[MirekElsner]

You said "(after post processing)", so maybe you understand all that.

Jim

Yes, I said "(after post processing)" on purpose. I am interested in typically processed end result where camera is only one link in the chain.

However, since post processing can cover just about any operation, the answer to your question is dependent entirely on the color space of the raw processor. Unless the raw processor attempts to save you from yourself, you should be able to produce any triple in the gamut of the raw processor's color space. If that color space includes triples that represent stimuli invisible to humans, you should be able to produce those triples. However, I wouldn't call them "colors".

I am mainly interested in seeing which visible colors are clipped in the processing chain based on ProPhoto RGB - that would be the blues and purples. I am interested in a "standard" pictorial processing workflow. Most people do not crank up the saturation all the way to the right and if they do, I don't care about these extremes... This is just curiosity, I don't see any practical use for it.

[MirekElsner]

Mirek, if you're looking for a color space in which to examine the output of a raw converter, I suggest that CIEL*u*v* is more appropriate than CIEL*a*b*. Although Luv shares a deficiency of Lab -- you can represent stimuli that aren't visible colors -- at least with Luv there's a clear boundary -- the conical horseshoe -- that separates colors from abstract triples. Luv has the big advantage for your purposes that it can represent any visible color. In addition, Luv has the concept of saturation, which is defined as how much chroma is in a particular triple as a percentage of the maximum visible chroma at that luminance and hue angle. Also, we're used to looking at chromaticity diagrams, and a horizontal slice through a Luv solid is the u'v' chromaticity plane.

Unfortunately (at least I think it's unfortunate), Luv is not commonly used by photographers. There are historical reasons for this that no longer apply. In 1976, when Lab and Luv were standardized, the print community had used Lab for years, and didn't want to change. The video folks had been using Luv, and felt the same way. They couldn't compromise, and standardized both. You can see the history in the default white points: D50 for Lab, and D65 for Luv. Then along came Photoshop. Since it was aimed initially at the print community, it supported Lab. Twenty-five years later, when a photograph is as likely to find its final destination on a LCD screen as a printed page, Photoshop still doesn't support Luv. It's a pity.

This doesn't really have much to do with the original post, but I couldn't help myself.

Jim

Excellent information. I will check to see if I can find formulas for converting between XYZ and Luv so that I can use it.

[Jim Kasson]

Excellent information. I will check to see if I can find formulas for converting between XYZ and Luv so that I can use it.

Here you go:
 
http://en.wikipedia.org/wiki/CIELUV#XYZ_.E2.86.92_CIELUV_and_CIELUV_.E2.86.92_XYZ_conversions

[MirekElsner]

Well stated Jim! There's a lot more going on under the hood, cameras record what we can't see and are not colors, we can see colors it can't record. And ProPhoto RGB has numeric definitions that are not colors.

I think though that they did a very good job at specifying the primaries in a way that the color space covers most visible colors and not so much garbage.

[bjanes]

Unfortunately (at least I think it's unfortunate), Luv is not commonly used by photographers. There are historical reasons for this that no longer apply. In 1976, when Lab and Luv were standardized, the print community had used Lab for years, and didn't want to change. The video folks had been using Luv, and felt the same way. They couldn't compromise, and standardized both. You can see the history in the default white points: D50 for Lab, and D65 for Luv. Then along came Photoshop. Since it was aimed initially at the print community, it supported Lab. Twenty-five years later, when a photograph is as likely to find its final destination on a LCD screen as a printed page, Photoshop still doesn't support Luv. It's a pity.

Jim,

What photo processing programs that use Luv are available?

Bill

[Jim Kasson]

What photo processing programs that use Luv are available?

For processing, none, AFAIK.

There's an ImageJ visualization tool that supports Luv, but I haven't used it:

http://rsb.info.nih.gov/ij/plugins/color-inspector.html

[added later: I'm not suggesting that CIELuv is a good space in which to perform photographic manipulations, although it's probably no better and no worse than Lab for that purpose; I'm saying it's a good space to look at how colors in an image relate to the gamut of human vision.]

Jim

[bjanes]

For processing, none, AFAIK.

There's an ImageJ visualization tool that supports Luv, but I haven't used it:

http://rsb.info.nih.gov/ij/plugins/color-inspector.html

[added later: I'm not suggesting that CIELuv is a good space in which to perform photographic manipulations, although it's probably no better and no worse than Lab for that purpose; I'm saying it's a good space to look at how colors in an image relate to the gamut of human vision.]

Jim

Jim,

I downloaded and installed the color-inspector plugin into ImageJ. It is a very nice and fast application, but it assumes that a RGB image is in sRGB, so it is not useful for evaluating "colors" that are outside the gamut of human vision.

Here is a Colorthink plot of the image that I previously used to show ProPhotoRGB and Lab renderings of a shot of flowers with saturated colors. The image was rendered into ProPhotoRGB with ACR. There is a cliff at the right edge of the plot which I interpret as indicating the presence of imaginary "colors". Is this correct?

Bill

[Jim Kasson]

I downloaded and installed the color-inspector plugin into ImageJ. It is a very nice and fast application, but it assumes that a RGB image is in sRGB, so it is not useful for evaluating "colors" that are outside the gamut of human vision.

Sorry about that.

Here is a Colorthink plot of the image that I previously used to show ProPhotoRGB and Lab renderings of a shot of flowers with saturated colors. The image was rendered into ProPhotoRGB with ACR. There is a cliff at the right edge of the plot which I interpret as indicating the presence of imaginary "colors". Is this correct?

I don't think so, Bill. Take a look at this chromaticity plot:



You can see that the red ProPhotoRGB primary is visible, and the green almost so. The blue primary is the one that's wildly out there. The cliff that you're talking about, I think, is along the line that connects the red and green primaries, which is right at the edge of what we can see, but for the most part, not outside it.

Jim

[MirekElsner]

Here is a Colorthink plot of the image that I previously used to show ProPhotoRGB and Lab renderings of a shot of flowers with saturated colors. The image was rendered into ProPhotoRGB with ACR. There is a cliff at the right edge of the plot which I interpret as indicating the presence of imaginary "colors". Is this correct?

Bill

How does it look like if you rotate the diagram so that the spectral locus in at the bottom and the pixels on top? Are there any pixels outside of the spectral locus?

[Jim Kasson]

How does it look like if you rotate the diagram so that the spectral locus in at the bottom and the pixels on top? Are there any pixels outside of the spectral locus?

Mirek, the way I view Bill's image is that the spectral locus is at the bottom (it's the curved part of the periphery of the white, flat thingie), and we're looking at the plot from outside the red end of the magenta line that's the flat part of the horse-shoe.

Jim

[bjanes]

Mirek, the way I view Bill's image is that the spectral locus is at the bottom (it's the curved part of the periphery of the white, flat thingie), and we're looking at the plot from outside the red end of the magenta line that's the flat part of the horse-shoe.

Jim

Jim's interpretation is correct: the horseshoe spectral locus is on the bottom.

Here is the original plot:


And here is the same image looking at the horseshoe from directly above:


The ImageJ plots are of some interest even though RGB is only for sRGB. Here is the ImageJ plot of the image converted to sRGB, also as viewed from above:


The HSB plot is revealing. There are many fully saturated colors.


Finally, I don't see what is so special about Luv--it is another opponency space like Lab that is just as unintuitive. Perhaps Jim can enlighten us.

Bill

[MirekElsner]

Mirek, the way I view Bill's image is that the spectral locus is at the bottom (it's the curved part of the periphery of the white, flat thingie), and we're looking at the plot from outside the red end of the magenta line that's the flat part of the horse-shoe.

Jim

I meant directly from the top, so that we could see which points are directly above the locus (visible colors) and which are not (not visible colors). Sorry for the confusion. Bill actually did it later and it seems the red, yellows and blues are outside of the locus and possibly not visible?

[Jim Kasson]

Finally, I don't see what is so special about Luv--it is another opponency space like Lab that is just as unintuitive. Perhaps Jim can enlighten us.

Bill, Luv is not the greatest thing since bottled beer. You're right, it is an opponent-color space like Lab. Is is about as perceptually-uniform as Lab, exhibiting about a 6:1 worst-case ratio on the MacAdam ellipses, but the errors occur in different places. The luminance axis is the same in the two color spaces. What it has that's useful in looking at image gamuts:

• You can represent all visible colors in Luv

• You can see the boundary where the visible colors stop and the mathematical constructions begin

• You can measure how close a color is to the most chromatic a color of that hue angle and luminance can be, IOW, how saturated that color is

Lab doesn't have any of those things.

Jim

[bjanes]

Bill, Luv is not the greatest thing since bottled beer. You're right, it is an opponent-color space like Lab. What it has that's useful in looking at image gamuts.

• You can represent all visible colors in Luv

• You can see the boundary where the visible colors stop and the mathematical constructions begin

• You can measure how close a color is to the most chromatic a color of that hue angle and luminance can be, IOW, how saturated that color is

Lab doesn't have any of those things.

Jim

Jim,

Personally, I prefer draft beer, but bottled beer is more convenient.

I'm afraid I don't know how to interpret Luv. How does one determine if a triplet is imaginary?

Thanks,

Bill

[Jim Kasson]

I'm afraid I don't know how to interpret Luv. How does one determine if a triplet is imaginary?

Bill,

I've been ignoring the asterisks and the single quotes, but I'm afraid I'm going to have to get into that to answer your question. If this turns out to be TMI, I apologize.

The plot that you posted is not actually CIEL*u*v*, usually abbreviated CIELuv or just Luv (pronounced "love"). It looks to me to be L in the vertical axis, and u' and v' in the horizontal axis. u'v' is a chromaticity space like xy. It's derived from xy as follows:

u' = 4x/(-2x+12y+3)

v' = 9y/(-2x+12y+3)

u'v' is more perceptually uniform than xy, which, among other things, gives undue prominence to the greens.

In the Lu'v' space, to see if the triplet is visible, ignore the L value and just look at u' and v'. If that pair is inside the horseshoe, the triplet is a visible color. If it's not, it's not. Pretty simple, huh?

In CIEL*u*v*, the chromaticity components are reduced by those of the white point, multiplied by a constant, and multiplied further by L*, the same luminance component as in CIELab. This has the effect of making the 3D L*u*v* visible color gamut roughly conical, with the only possible u* value for a color with an L* of zero being zero, and likewise for v*.

That makes it a little harder to look as a L*u*v* 3D plot and see the limits of visible colors. If the tool you're using doesn't provide the option of plotting the spectral locus (and it looks like the Lu'v' plot did have that option, the best thing to do is include a bunch of points that define the locus. If you want, I can whip up an Excel spreadsheet with that data; you'll also be able to see the calculations. PM me if you want that.

In my own work, I used Excel or Smalltalk to generate the data and commercial 3D graphing tools (Mathematica works fine) to plot the graphs. Now, when I have to, which is not often, I do the whole thing in Matlab.
'
Now that I've said all that, I think there's something wrong with your Lu'v' plots, if that's what they are. Before I possibly waste my time, is that what they are?

Jim