Raph's color management page

Color management is a tough problem, still only partly solved in the proprietary world, and there only through an enormous investment of time and money.

Free tools for color management are slowly coming into existence, but are primarily at the level of low-level infrastructure. These include Graeme Gill's libicc, the GCMS project, Karl Heinz Kremer's freecolormanagement.com, and Martí Maria Saguer's LCMS. That adds up to about two under active development, with a bunch more in non-development (including the moribund gcmm codebase).


One of the serious problems with color management is the patent situation. EFI owns a number of patents and is not shy about enforcing them. I think we should be thinking about patent liability issues both for individual efforts and systemically.

There are three patents to be careful of:

According to the November 16, 1998 issue of the Seybold Report on Publishing Systems, p. 44, EFI is suing Harlequin over these three patents. I'm not sure why the Schreiber patent is in the mix, because a press release seems to indicate that Harlequin has already licensed it. (I later learned that they licensed Schreiber for some products, but apparently were not paying royalties on all products EFI claimed used the technology)

Slightly more information is contained in a EFI press release about the lawsuit. Neither source actually names the patents; the three listed above are merely my best guess.

The July 26, 1999 Seybold Report has an article stating that EFI won a similar patent lawsuit against the PhotoScript group. I can't find any relevant documents online, but this info seems to identify the suit itself:

4,500,919; 5,212,546; 5,424,754

This info was found at Intellectual Property Today.

This suit has since been settled. Overall, I think that's bad news for us, as it leaves no legal precedent that the patents are anything other than fully valid.

Thanks to Karl Heinz Kremer for alerting me about another web page devoted to color management patents.

Gamut mapping

Color management has a number of sub-issues of which mapping between device color values and colorimetrically based color values (ie, the job of ICC) is only one part. A really big part of the problem is gamut compression, ie mapping the colors in the "ideal" source image to the range of colors that can actually be produced by the device. And, while colorimetric calibration of a device can in theory be done with total rigor, gamut compression has much more of the flavor of a black art.

Choosing a gamut compression strategy depends on many different factors, depending in large part on the user's goals. In many cases, the goal is to produce the highest quality color images given a set of source images. In this scenario, choosing a "best" gamut compression map depends on characteristics of the source image. For example, if the source image contains exceedingly bright colors, then a "best" gamut compression may comprise toning down the overall color saturation of the image, so that relative color saturation is well preserved at the expense of absolute color matching. Conversely, if the source image contains only subdued colors (ie within the gamut of the device), then the gamut map will resemble absolute color matching more closely.

In some applications (including presentation of corporate identity and clothing / fabric / product colors), color match of particular colors may override overall color rendition.

One important application of color printing is "proofing," or creating a color print intended to match the appearance of a final product created with a different device (for example, web offset printing). In these applications, the gamut of this second device is actually more relevant than the gamut of the proofing device, assuming that the latter is sufficient to contain the former.

Lastly, graphic arts applications need feedback from the "color management system" about gamuts available. In this way, the color can be tuned manually for optimum results within the appropriate context. This context may be (a) producing the highest quality results on a particular device (for example, if you're making signs), (b) producing consistently high quality on a wide range of devices (the usual electronic publishing scenario - you're expecting your images to be printed out users' individual printers), or (c) optimizing results for graphic arts production (which would be the same as (a) if you could count on the proof being an exact appearance match for the final product).

Add to this the fact that most pages contain multiple different images and spot colors, so you may get better overall results by taking into account "ensemble" effects rather than optimizing each individual image separately.

Ján Morovic has a very interesting PhD thesis on gamut mapping.

While it's important to provide lots of control over the gamut mapping process for high end applications, it's even more important to provide a good set of defaults so that people who just want to print web pages and digital camera shots and have them look good.

Color profiles

Producing ICC profiles (or some other technically similar form of calibration data) is another hard problem. There are a lot of very expensive proprietary tools for this, but basically nothing on the free side. In some cases, we may be able to rely on printer manufacturers for this data, but I am concerned, as the calibration data is highly sensitive to such things as paper, inks, resolution, dithering algorithms, and the phase of the moon. We want to be able to tinker with this stuff, right?

Producing good profiles demands the right tools, a good eye, and expertise in color management. To do it quickly and consistently also requires a spectrophotometer, which costs around $2.5k for a decent scanning model.

One approach to doing profiles on the cheap is to use a color scanner rather than a spectrophotometer. Unfortunately, EFI also owns a patent on this process. In addition, it is vulnerable to serious quality problems, as the color response of scanners rarely matches human visual response very well. My tests with photographic reflective media and an Epson Expression 636 indicate worst case delta-E of about 9 for a best-fit linear model. (summary for color-science impaired: looks like shit)

Related links

Recommended Reading

Edward J. Giorgianni and Thomas E. Madden. Digital Color Management: Encoding Solutions. Addison-Wesley, 1998. ISBN 0201634260.

gimp home
* www.levien.com