133

(Strictly speaking, this is a question about XeLaTeX and the xcolor package but I’m assuming that the issue is actually with the color model of PDF.)

I’m having the problem that colours in my PDF don’t match. For example, I screen grabbed a colour value as RGB, declared it in my document and the resulting colour was completely different. To illustrate:

screenshot-1

Both colours were grabbed off the screen – on the left, from the PDF generated by XeLaTeX. On the right, the original colour whose RGB value I tried using in LaTeX (in case my issue isn’t clear: the colours on top are differnt, yet they have the same HSB values).

So I thought that perhaps RGB doesn’t map without loss to whatever colour model PDF used and tried the obvious alternative – CMYK, which, after all, is often used in print (right?). Again, the colours are off, but now less perceptibly.

Another example (this time using CMYK values):

screenshot-2

The colour is supposed to have the CMYK value (0.94, 0.63, 0.04, 0.07). I have declared the colour in my code as follows:

\definecolor{primary}{cmyk}{0.94,0.63,0.04,0.07}

The bottom half of the picture is created by TikZ. The top half is actually an EPS that I included via includegraphics. The EPS was created in Inkscape. Neither of the above colours has the correct CMYK value when examined with a colour picker:

The top is (0.96, 0.67, 0.08, 0.19). The bottom is (0.92, 0.61, 0.06, 0.09).

(Incidentally, when I try to enter the “correct” CMYK value in Inkscape, it “auto-corrects” my value to some other value. Wut?)

Now, this mightn’t seem like a TeX related question at all. But, long story short,

How can I reliably reproduce colours in pdfLaTeX/XeLaTeX?


I know that the issue is confounded by the fact that PNG (which is the format of the above screenshots) uses an own colour model and so the screen shots may look different on your PC than on mine. But since I pasted the colours into the same image, I think that the difference should be preserved.

Konrad Rudolph
  • 39,394
  • 22
  • 107
  • 160
  • 2
    I have no clue about this, but it's funny that when I try to read the color from your first image (on OS X with the DigitalColor Meter) even on the “solid” area the values flicker when you move a bit up and down. (Edit: Ah, it's a jpg, that's why) – Juan A. Navarro Jan 31 '11 at 15:48
  • @Juan: damn, so they do. I hadn’t noticed until now (I use another colour picker that doesn’t update values continuously, only on click). But for the record, these fluctuations are minimal and don’t account for the observed difference. – Konrad Rudolph Jan 31 '11 at 15:54
  • 1
    did you \usepackage[dvipdfx,cmyk]{xcolor} ? –  Jan 31 '11 at 15:59
  • 2
    According to the GIMP, the two colors in your first picture do not have the same HSV values, the first one has hue of 30, the second mostly 35 (with few spots of 34). I believe that in general, conversion between different color spaces is device dependent, so if you use different device profiles, you can get different results. – Jan Hlavacek Jan 31 '11 at 16:16
  • @Herbert: I actually didn’t load xcolor directly at all. But even if I do (and using the parameters you mentioned) nothing changes. – Konrad Rudolph Jan 31 '11 at 16:23
  • 1
    @Konrad: \usepackage[xetex,...]{xcolor} is also supported by xcolor –  Jan 31 '11 at 16:26
  • @Jan: that’s just an artifact of the PNG. I’m getting the same result as you when picking the colours off the screenshot. You’re certainly right about the device dependency. But why doesn’t even work for me on the same machine? And the colours aren’t even very similar, the difference in hue & saturation is very noticeable. – Konrad Rudolph Jan 31 '11 at 16:27
  • @Herbert: but – correct me if I’m wrong – I was under the impression that specifying a driver is only necessary (and indeed recommended) when the output format differs from the intermediate format produced by the LaTeX processor. I’m not doing any postprocessing (other than what is done by a call to xelatex itself, via dvipdfmx). (That said, using the xetex option desaturates the colours even further (i.e. they get worse)). – Konrad Rudolph Jan 31 '11 at 16:31
  • 4
    This question opens up a huge can of worms. Colour spaces are immensely complicated beasts: Even RGB and RGB be different things! There is sRGB, which is the standard colour profile for computer monitors, and there is Adobe RGB, which allows for a bigger gamut of colours (especially, more greens) and is more suited for photos. It is my understanding that one should avoid CMYK altogether unless the goal is to produce PDF for a specific printing press and paper whose characteristics are known. – Harald Hanche-Olsen Jan 31 '11 at 16:35
  • 1
    My favourite photo editing software (bibble) lists 7 different RGB colour spaces. The CMYK world is orders of magnitude more complicated. – Harald Hanche-Olsen Jan 31 '11 at 16:37
  • @Harald: I should have mentioned in the question that I am completely clueless about colour models in computers. I just vaguely remembered having heard CMYK being used for printers and thought, hey, perhaps that’s what PDF needs. – Konrad Rudolph Jan 31 '11 at 16:38
  • Like Herbert, I recommend to load xcolor using the cmyk option. If \usepackage or \RequirePackage (if required before the class) results in an option clash because xcolor is loaded by some class or package, try \PassOptionsToPackage{cmyk}{xcolor}. I had this changes in colours in my own documents because of the transparent option of TikZ. Are you using it? – Martin Scharrer Jan 31 '11 at 16:48
  • @Martin: It definitely doesn’t help. I’ve created a minimal document and tried that, still get the colour shift. – Konrad Rudolph Jan 31 '11 at 19:14
  • @Konrad: What's with transparencies in your TikZ drawing? This adds literally a whole new dimension to the color model. – Martin Scharrer Jan 31 '11 at 19:16
  • @Konrad: "the colours on top are differnt, yet they have the same HSB values"—I can reproduce the result of Jan: the hue of the right panel is 34/35, not 30. How should the same color wind up as two different colors in a PNG file (AFAIK both the screen and PNG use RGB). – Philipp Jan 31 '11 at 19:31
  • 1
    Regarding Inkscape: Since SVG doesn't support CMYK, Inkscape uses sRGB internally. When you enter a CMYK value, it gets immediately converted to sRGB and then converted back, which is lossy. (There is no bijective mapping between the color spaces.) – Philipp Jan 31 '11 at 19:47
  • @Philipp: the colors were never identical but (apparently?) got mapped to the same HSB value. Now, when I made the screenshot and saved it as PNG, its gamma was “corrected” (http://hsivonen.iki.fi/png-gamma/), thus distorting the colours further. These through PNG encoding distorted colours have different hues, not the original. This is my guess. – Konrad Rudolph Jan 31 '11 at 20:21
  • 1
    By the way, people, most of these remarks should have been answers. While they don’d solve the problem directly, they are excellent remarks and may serve as jigsaw puzzles in the explanation. Please feel free to re-post them as answers. All of them deserve to be upvoted. – Konrad Rudolph Jan 31 '11 at 20:22
  • 1
    @Konrad: I see your point, but now there is a good answer, and at least my comment basically just reflects the fact that I know enough about colours to understand that I don't know anything about them. Which is useful, but hardly deserves to be dignified with the “answer” label, methinks. – Harald Hanche-Olsen Jan 31 '11 at 21:40

2 Answers2

158

The colour model of PDF is extremely sophisticated and bewilderingly flexible. You can read all about it in section 8.6 of the pdf specification (you can download it here). There are a number of prepress tools that exist for managing a colourspace workflow with PDF. Adobe Acrobat Professional has an "output preview" tool which is what I use.

PDF supports many colour spaces: gray, cmyk, rgb, indexed, LAB, deviceN. This allows for things like spot colours (and tints thereof), special metallic or varnish inks, 6-colour printing, duotone, multitone, registration colours, and so on. You can have multiple such colour spaces within the same PDF (although for printing you'll need to convert to some common space, at least for each page). There can even be multiple variants of the same colour space family (for example sRGB and the variant of RGB that your scanner or digicam uses; cmyk with different dot gains, etc).

The situation gets even more complicated when you involve transparency (see section 11.7 of the spec): in order to overlay one object on another the PDF viewer needs to convert them to a common "blending" colour space, which can be specified in various ways in the PDF (including at the "page group" level: this explains why including transparency on a page can change the appearance of other objects on that page, because those objects now have to run through a colour space conversion). There are various restrictions and special cases, for example device colours cannot be converted reliably into CIE based spaces (such as sRGB). Since transparency groups can nest, there can be multiple rounds of colour space conversion during rendering.

TeX support

Now for latex support via xcolor: If you load xcolor with the (default) natural option there will be no colour space conversions, and you will have access to the PDF "DeviceRGB", "DeviceCMYK" and "DeviceGray" spaces. Eg, \textcolor[rgb]{1,0,0}{DeviceRGB red}, and \textcolor[cmyk]{0,1,1,0}{DeviceCMYK Red}. Your PDF viewer (for screen) or your printer driver (for hardcopy) will have to convert colour spaces if necessary. If you load xcolor with options like rgb or cmyk then xcolor will convert to that colour space, using the formulas from section 6.3 of the xcolor manual (which are not very sophisticated -- compare them to the formulas in 10.3 of the PDF spec, with BG(k) and UCR(k) functions, etc).

If you use the dvips route to producing PDF then you may be able to access also spot colours and other colour spaces, using the named and ps models. I believe Context makes these things somewhat easier, but I don't speak from experience.

Remember that converting device colour spaces to the screen colour space is in general not well defined, and different viewers may do it slightly differently, resulting in the mismatches you've observed.

One good solution with pdflatex is to use only deviceRGB (for screen) or deviceCMYK (for most printers) and then set an "Output Intent" to define the device space as, eg, sRGB IEC61966-2.1. See section 14.11.5 of the PDF spec. The pdfx package does this, for example (although that package is pretty rough around the edges and you'll generally need to edit it directly to get it to work). A more "proper" way to use calibrated colour spaces would be the "Default Color Spaces" mechanism (section 8.6.5.6) which specifies a way to remap DeviceRGB, DeviceCMYK, DeviceGray into device-independent CIE-based colour spaces. I'm not aware of any latex package that makes use of this PDF1.1 feature, though. (Grepping through the sources suggests that again Context may have some support for this).

The case of the different coloured swatches

In your first pair of images, note that the left image has a gray triangle in the upper right corner. This indicates that it is a deviceRGB colour from a screen grab. Click on the icon to the left of the "HSB Sliders" indicator and choose the colour space as for the right swatch, and you'll see that the colours will then match.

The case of inkscape "autocorrecting" CMYK

Perhaps this article will be helpful.

Lev Bishop
  • 45,462
  • 22
    This answer is awesome on so many levels. It gives a concise explanantion of (1) PDF colour models, (2) TeX support for colour models, (3) OS X colour support and the correct usage of the “Colors” dialog, which has somehow eluded me for years, and finally (4) a pointer to the bewildering behaviour of Inkscape. What can I say? Thanks a lot – Konrad Rudolph Jan 31 '11 at 20:46
  • 6
    @Alan: I will, but first I will wait two days and make this question a bounty. Then I can award more points for the answer. – Konrad Rudolph Feb 01 '11 at 07:50
  • 2
    Do I understand it right, that for example \usepackage[cmyk]{xcolor} will convert all color definitions in the document that use another color model? sth like \definecolor{halfblack}{gray}{0.5} would in the end be CMYK? – Gambhiro May 06 '11 at 07:41
  • @Nyiti: that is correct. – Lev Bishop May 06 '11 at 15:23
  • @Lev Bishop: Does this also hold for the colors of embedded PDF images, that is, if I \includegraphics{RGB.pdf} the final PDF will have only cmyk colors? – Daniel May 26 '11 at 22:18
  • @Daniel: no, included images are not converted by xcolor. – Lev Bishop May 27 '11 at 00:07
  • @LevBishop: Great answer! This answer could lead to a wonderful article for our blog. If you would have time for that, this would be really great. – Stefan Kottwitz Oct 27 '11 at 12:18
  • @StefanKottwitz I agree that it might be nice to have a blog post on colour spaces in relation to pdf and latex. I won't have time to write one any time in the near future but I'm perfectly happy if someone else wants to convert my answer into blog format. Otherwise, maybe in the far future I'll have a go... – Lev Bishop Nov 02 '11 at 14:02
  • 1
    By 'One good solution with pdflatex is to use only deviceRGB (for screen) or deviceCMYK (for most printers) and then set an "Output Intent" to define the device space as, eg, sRGB IEC61966-2.1.' do you mean that it's not generally a good idea to use the same PDF file for the screen and the print version of a document? – Christian Feb 24 '13 at 22:51
  • 2
    @Christian: yes, having separate screen and print versions gives the most control over colour, although such a level of control is rarely needed for most simple documents, in which case I would choose CMYK or RGB depending on the primary medium of the document. You only want complete control if you're at the level of worrying about rich blacks, trapping, and such things. (There may be other reasons, unrelated to colour, to have separate print and screen versions, such as doing away with recto/verso distinction for screen, adding a thumb index for print). – Lev Bishop Feb 25 '13 at 05:47
  • @LevBishop In my document I'm using the colour lightgray which looks nice on screen, but is barely decipherable when printed (with a BW-laser-printer). Using the xcolor pkg do you think it would make sense to try and convert lightgray with \convertcolorspec to a cmyk equivalent? Could that (possibly) result in a more legible paper copy? If so, how to code it? – nutty about natty Jul 09 '13 at 22:30
  • @LevBishop Posted the question here. – nutty about natty Jul 10 '13 at 07:34
2

2020-05-21 update: My remarks about Evince rendering pdf's too bright are not true. See the comments section for details.

Expanding on the following remark of Lev Bishop:

Remember that converting device colour spaces to the screen colour space is in general not well defined, and different viewers may do it slightly differently, resulting in the mismatches you've observed.

I noticed pdf viewer Evince does not render colors as specified. E.g., take the following document with grayscale boxes:

\documentclass[10pt]{article}

\usepackage[pdftex,rgb]{xcolor}
\definecolor{gray0}{gray}{0}
\definecolor{gray2}{gray}{0.2}
\definecolor{gray5}{gray}{0.5}
\definecolor{gray8}{gray}{0.8}
\definecolor{gray10}{gray}{1}

\begin{document}
%Draw grayscale boxes
\frame{\color{gray0}\rule{2cm}{2cm}}
\frame{\color{gray2}\rule{2cm}{2cm}}
\frame{\color{gray5}\rule{2cm}{2cm}}
\frame{\color{gray8}\rule{2cm}{2cm}}
\frame{\color{gray10}\rule{2cm}{2cm}}
\end{document}

Open the pdf with Evince and pick with Gpick yields:
0.09 (indeed, black text is not rendered as true black as in rgb(0, 0, 0))
0.27
0.54
0.81
0.99

Make a screenshot of the pdf, open with Gnome's Image Viewer: The picked colors match the colors specified in Latex document.

Open the pdf with Adobe Reader and pick with Photoshop (on Windows) yields: The picked colors match the colors specified in Latex document.

Open in Evince a pdf-document generated with LibreOffice and pick the text: 0.09 (again, black text is not rendered as true black as in rgb(0, 0, 0))

Conclusion: Evince or its pdf backend Poppler does not render colors as specified.

Rendering inconsistencies, unfortunately, seem to be part of all pdf viewers:
https://srg.doc.ic.ac.uk/projects/pdf-errors/results.html
https://link.springer.com/article/10.1007/s10664-018-9600-2

For the interested, details about how Evince renders colors:
Plot specified gray vs picked gray (using wolframalpha.com):
Plot[{0, 0.09}, {0.2,0.27}, {0.5,0.54}, {0.8,0.81}, {1,0.99}]
Inconsistent colors Evince
Fit a curve through the dataset:
Fit[{0, 0.09}, {0.2,0.27}, {0.5,0.54}, {0.8,0.81}, {1,0.99}]
A linear approximation comes very close: grayEvince = 0.9*graySpecified + 0.09
Conclusion: Evince (or Poppler) particularly renders dark colors too light. It maps the range [darkest,lightest] to [0.09,0.99] instead of [0,1]

Final request: Can somebody pick the color black with Gpick in another Poppler-based pdf viewer? This way we can determine whether the rendering problem lies with Evince or Poppler.

Bart
  • 397
  • Not sure whether I correctly perform the experiment, but even with Evince on current Debian unstable, the black from the document given here appears as #000000. Same thing with xpdf, which uses libpoppler (package libpoppler82 version 0.71.0-6 on my system). – frougon Apr 17 '20 at 16:42
  • 1
    I later found out that Evince slightly brightens its appearance when it is not selected. So, the Evince window was slightly brightened every time I sampled it with GPick. So yes, you're right, the colors are okay if you have the Evince windows selected. – Bart Apr 20 '20 at 13:13
  • 1
    Ouch, these things are full of traps. :) – frougon Apr 20 '20 at 13:14
  • :) Indeed they are – Bart Apr 23 '20 at 18:25