The p-Code Machine

One of the big criticisms of open source raw converters is that they are not capable of producing the same colors as the in-camera generated JPEGs, or manufacturer supplied software. The whole problem with this is that these color transformations are usually a secret of the company that produces the camera. The color transform takes into account several factors: (1) the spectral properties of the image sensor, (2) linearisation and an (3) artistic streak, to make things look pretty.

In the past I’ve written articles about how to color profile your camera, to get accurate colors when combined with a profiled display. This method only encompasses the first two factors, since accurate color does not involve any artistry. And while accurate color is a good starting point for image processing, it does make images look a bit clinical (without any further work).

Some vendors actually have color profiles of their own, but these often do not work well, as they might include vendor specific technology. And even if these vendor profiles would work, they still couldn’t be freely redistributed, because they are copyrighted by the respective vendors.

A while back Hub suggested (on IRC) that it would be interesting to generate new color profiles between linear RGB output from a raw converter (like UFRaw or Niepce) and fully processed images produced either by the camera itself, or the raw processing software that came with the camera.  I took interest in the concept, but my knowledge of color management wasn’t sufficient to actually get things done. I cried out for help on the ArgyllCMS mailing list, and Klaus Karcher decided to help me out. The procedure isn’t trivial, but it is manageable. To speed things along I scripted the process.

So far, early results look quite promising. The generated profiles aren’t exactly like the original images, but they are quite acceptably close.