So, you got proof?
Hi Jeff,
As you know, it's hard to prove something that's hardly noticeable in regular images (with photon/read/pattern noise). Nevertheless, it may be interesting to conduct an experiment that's goes a bit further than handwaving. One might even learn something.
May I propose the following, so others can also participate with their particular tools / workflow / output profiles and devices:
1. Download
this image (it's in a ZIP archive to make sure no download corruption can take place) of a 16-b/ch version of a Granger Rainbow. It's designed to have more than 6 pixels per degree of hue-change, and plenty levels in the vertical brightness/saturation axis.
2. Open the image in e.g. Photoshop.
3. In the Missing Profile warning dialog select "Leave as is (don't color manage)"
3. Make 2 additional copies, 3 copies in total.
4. On copy #1,
Assign a ProPhoto RGB colorspace profile.
5. On copy #2 and #3, Assign a Beta RGB colorspace profile.
6.
Convert the copy #3 from Beta RGB to ProPhoto RGB.
The copies #2 and #3 contain the same colors, only mapped to two different colorspaces. Copy #1 is a torture test for any system, because it contains imaginary 'colors' and Out-of-Gamut colors for all output modalities, so we are guaranteed to see issues after conversions. We could also additionally make a Copy #4 (e.g. a duplicate of Copy #2) for an 8-bit/channel output pipeline, with a mode change to 8-b/ch RGB if we stay in Photoshop.
7. Now resample all the images to 360% size (like a conversion from 300 to 360 PPI and a 3x magnification to make it easier to see the effects at 100% zoom), while using
Bicubic Smoother to reduce the risk of adding resampling artifacts.
8. Now
Convert the colorspace profile of all images to that of an output modality's profile, e.g. a display profile to evaluate on a monitor display (sRGB or Adobe RGB capable) or a paper profile of choice for the evaluation of printed output.
The posterization issues are most likely to occur in the Blue/Cyan to Yellow range, especially at the transitions between display primaries or at transitions between ink colors. Look at the brightest least saturated regions for curved posterization and the regions just below the middle of the vertical brightness/saturation scale for horizontal posterization bands where I'd expect the most prominent trouble, if any. One may prefer to crop out that area to avoid wasting too much ink when actual prints are made. The printing process, dithering/weaving, and paper structure may still hide remaining issues, so nothing beats an actual print.
Of course, these output profiles may introduce additional modality specific corrections and non-linearities on top of the colorspace conversions. So we're looking for pipeline differences, not non-linearity issues (which are almost impossible to avoid). Again, these artificial targets are much more critical than real images, but that's because we want to spot the issues and avoid them (if possible) by modifying the workflow.
Maybe you have a better suggestion, so feel free to share your thoughts. The famous "Twenty-Eight Balls.tif" target is also a useful basis for testing, but it's a bit larger, so it requires more paper.
Cheers,
Bart