Image Fidelity

Absolute/Relative Intent, White Point

This is a digression into the issue of absolute versus relative colorometric rendering intents of the white point. It is essential to understand when reading the article IT8.7 Profiling Tests.

[Stockholm 2004-12-23]

The ICC specification

The ICC profile specification defines a profile connection space that is based on colorimetry relative to a D50 reference illuminant, from which absolute colorimetry can be derived using the specified media white point. (D50 is paper "yellowish", daylight is D60, and monitors are often calibrated to D65.)

Assigning a profile

When such a profile is assigned to an image in Photoshop, the RGB values are preserved no matter what rendering intent is used, and no matter what gamma was used. (But the gamma of the scanned image must of course be the same as the gamma that was used when the profile was made.) The image will of course look different (look well calibrated) after a profile is assigned, but that is because the RGB values are now interpreted according to the profile.

Lab values in different rendering intents

The Lab values given in the Info palette of Photoshop, however, depends on the rendering intent specified in Color Settings. (Choose Lab Mode in the Palette Options to see those values. We assume here that the image is in RGB mode.) When the rendering intent is set to Absolute Colorimetric, the Lab values of the image are exactly as corrected by the profile, preserving the white point of the target. If, on the other hand, the rendering intent is set to Relative Colorimetric, the white point of the target is mapped to the white point of your system, so that the L-value becomes close to 100 (often about 99). In both cases, the RGB values may be just about anything.

As an example, assume that the clear film patch in the IT8 target, named GS0, has the reference file Lab triple [90, 1, 1]. When the scanned image of the IT8 target is assigned the profile in Photoshop, you will thus see:

However, for some reason, actually carrying out the conversion to Lab Color in Photoshop will actually give more correct values — some gamut outliers will often differ about one or two units in either the 'a' or 'b' component between the actual conversion and what the Info palette says before the conversion.

Converting to a working space

When the image, having been assigned an input profile, but still being in RGB mode, is converted to a working space of your choice, the RGB values will change according to the new working space. The change in RGB values depends on what rendering intent is used. The Lab values, however, will not change as a result of profile conversion provided the rendering intent specified in Color Settings is the same as the one used in the profile conversion and the color patch is within the gamut of the newprofile! The profile conversion compresses out-of-gamut colors, which will change the Lab values accordingly.

As an example, assume that the GS0 patch of the scanned target gives you the RGB triple [230, 225, 220] before you color manage it in any way. Assigning the correct profile will not change those values. But when you convert it to a working space, you may get the following:

Practical uses

Since we want to preserve the look of the image, and also use as much as possible of the RGB range 0-255, the Relative Colorimetric rendering intent is a very good one. However, that also means that the Lab coordinates will be shown according to this setting, which means that a scanned IT8 target will get different values compared with the ones specified in the reference file for the target. That is usually no problem, since we normally just want a good picture, even if the picture is just a dull IT8 target. But if we would like to really see the Lab coordinates, despite the image has been converted to a standard working space, we need a procedure for that.

How to convert to a new space while preserving most of the Lab values

In my comparisons, I wanted to display images in a web friendly format, so the working space of the image had to be sRGB. So, to convert an image to a standard working space while preserving the original Lab values, at least for the in-gamut colors, one can do:

  1. First of all, make sure that the image is still in 16 bits/channel mode, since conversion at 8 bits/channel may introduce severe rounding errors.
  2. Set the rendering intent in Color Settings to Absolute Colorimetric.
  3. Convert to Lab Mode. (After this, one can return to Relative Colorimetric in Color Settings, but it's not at all necessary.) This stage changes the appearance of the image, usually making it darker — a GS0 patch with the Lab triple [90, 1, 1] will turn darker, but it will not turn yellowish, even though the values are relative to D50. This state of the image is the most accurate one for comparing the Lab values with those in the reference file.
  4. Convert to a working space of your choice, using Relative Colorimetric. The RGB values will then be around 225 for patch GS0. Those colors that are within the gamut of sRGB will retain their Lab values, whereas the outliers will be "compressed".

After the procedure above, the image may be converted to 8 bits/channel and jpg'ed for display on the web.

Further reading and references

There are zillions of sites on the internet dealing with color management and related issues. Rather than trying to list many of the essential ones, I will only list two:

Harald E Brandt
Hägersten, Stockholm
Last updated: 2011-10-27 at 15:07:10 +0200