I am using Silverfast to make a scan and save it as CIE-Lab. I set Absolute Colorimetric rendering intent in Color Management System preferences, calibrated the scanner, then scanned some white paper with known Lab values close to [100, 0, 0]. The paper has no optical brighteners, to keep things simple. When I used Photoshop to look at the file's Lab values, they were [100, 0, 20] or light yellow! I then switched to Relative Colorimetric rendering intent, redid the scanner calibration, and re-scanned the sample. This time the Lab values were correct: [100, 0, 0]. Some questions:

  • Why is Relative the right choice, not Absolute, when saving CIE-Lab values?
  • Why does rendering intent affect Lab values, which are device independent?
  • Are there any other changes I should make to the following set of CMS preferences if I want to save correct Lab values?

enter image description here

  • Image upload is working now if you want to try again.
    – Rob
    Aug 7, 2018 at 17:53
  • 1
    Ideally, when you scan something you should get a colorimetrically accurate result. The tint of the scan should match the tint of the paper. I use an Epson V850 and it's fairly good. I make my own patch set as the IT8 set is rather tiny and use Argyll to create scanner profiles. There are a few other issues such as "spatial crosstalk" which occurs when light reflects off the brighter areas back to the surfaces of the scanning illuminant and goes back to the scanned object. It can be quite large.
    – doug
    Aug 8, 2018 at 5:59
  • Doug, do you know of a reference on scanner spatial crosstalk besides this one which is kind of expensive: webstore.iec.ch/preview/info_iec61966-8%7Bed1.0%7Den.pdf (just a preview, it's not the full document)
    – KAE
    Aug 10, 2018 at 12:49
  • 1
    Not really, but I do know that if you place an unprinted circle of, say 6mm diameter in a 2" black surrounding circle and compare the scanned L* value to that of an unprinted region there is a significant difference. My Epson scanner shows a change of between 5 and 7 L* units. This corresponds to about an additional reflected light of 15 to 20%. Also, modern scanners use LEDs with no uV component. You should get roughly the same readings on high OBA paper and no OBA paper if they have the same white point as measured with a uV cut spectrophotometer.
    – doug
    Aug 13, 2018 at 2:13
  • 1
    @KAE While 5-7 dE is quite large it occurs only in unusual situations. It's the difference between two small areas of near white when surrounded by a much larger white or black region. Profile targets are made with a mix of colors and the average reflectance is typically less than 20% which reduces the induced re-reflected light error to an average that is under 1dE. It's also not visually noticeable because you can't have two colors physically near each other that are scanned with a large dE yet have the same luminance. That's why the large effect goes unnoticed.
    – doug
    Aug 31, 2018 at 21:34

2 Answers 2


I don't think that it has anything to do with using the CIE-Lab colour space, and you should only use absolute colorimetric rendering intent if you are doing colour proofing. Relative (or perceptual) intent is nearly always a better choice, and I wonder why you want to use absolute colorimetric, or the CIE-Lab colour space, for that matter.

Fraser, Murphy and Bunting have this to say about absolute colorimetric rendering intent in Real World Colour Management, 2nd edition:

Absolute colorimetric differs from relative colorimetric in that it doesn't map source white to destination white. Absolute colorimetric rendering from a source with a bluish white to a destination with yellowish-white paper puts cyan ink in the white areas to simulate the output of one printer (including its white point) on a second device.

I don't know why you have experienced such a marked colour shift, but most white printer papers contain optical brighteners that can confuse colorimeters. White papers without optical brighteners will generally have a yellowish tint, and this will be picked up by a good colorimeter when using absolute colorimetric rendering intent. If you are scanning a fine-art book, the paper is likely not to contain optical brighteners.

Note that the conversion algorithms between CIE-Lab and other colour spaces can introduce inaccuracies, especially if you are using 8-bit channels. Always use 16-bit channels with CIE-Lab.

If you value the hair on your head, stick to relative or perceptual colorimetric rendering intent, and only use CIE-Lab if you have a real reason to. There's nothing wrong with Adobe RGB. If you are scanning for web content, keep it simple and use perceptual and Adobe RGB.

Edit: Thinking back to my colour management days, I had horrendous colour shifts if I inadvertently triggered double conversion, either on input or output. However, I was using Photoshop CS2 with Windows 2000. Neither the application nor the operating system knew what the other was doing. Maybe things have improved since then.

  • Thanks for all this info. I have to use CIE-Lab for this particular project. The white sample has no optical brighteners. A colleague pointed out that in PS, Lab is D50/2 which is yellowish, and perhaps PS converts the file Lab values to D50/2 when it imports them if they are absolute colorimetric. I will stick with relative as you recommend! The same colleague pointed out that Lab is relative to a reference white (color-image.com/2011/10/…) so that may be why relative colorimetric rendering intent is better.
    – KAE
    Aug 8, 2018 at 12:15
  • 2
    Most modern scanners use LEDs that have little to no uV. Not sure about older ones but it's easy to tell by scanning paper with high OBAs v low/no OBAs. The Epson V800/V850 does not have uV in it's illuminant. Worth checking. uV will significantly distort profiles on OBA papers.
    – doug
    Aug 13, 2018 at 2:18

Lab* = (100,0,20) does not make sense, as L*=100 is maximum intensity, and thus it is white. If it were not white (has some color cast), you could additively add another complementary color, and then L* would be >100, which is not possible per definition.

My guess is that the Lab reading cited is the result of a wrong calibration. Lab* is a calibrated relative color space, so you need white calibration. If done properly, anything else than white should have L* < 100. Another explanation could be mathematical (d)effects when converting color spaces. Then the Silverfast support should know.

  • I think the color space conversion is the issue, as you suggest. I don't recall getting clarity from SilverFast on what could have caused this, but it's been so long I may be misremembering.
    – KAE
    Jun 3, 2019 at 13:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.