Another way is the dot product with the vector to the 100% colorant value to white (or some other choice related to the underlying colorant vector direction) in a particular colorspace (XYZ, L*a*b* etc.). This could be viewed as creating an "optimal" mix of the co-ordinates as a measure, rather than just picking one of the co-ordinates.
Sounds neat... will take a look at the idea. I wonder how will it interact with the 'hook' shape.
XYZ isn't a good choice as absolute measure, as it is linear light, so the blacks steps will be too big. A gamma or L* curve applied to XYZ is needed at least.
Of course not raw XYZ.... more like XYZ emulating a reference curve shape. For example, take offset printing, and try to fit the same shape (scaling the endpoints).
Density is a log scale, which is why it's not too bad, although in the Colorbus calibration system I set a target density curve corresponding to a typical 20% dot gain press curve.
Really the problem is not that much the linearization (incremental dE or incremental dE00 could be good choices also); the real problem is determining the ink limits automatically. Right now, the way I find ink limits is by plotting (in 3D) the linearization curves, and superimposing that over the target gamut shape (for example, target being ISOCoated). Then I can clearly see, by rotating the graph and looking at it, which limits do I have to set in each ink channel to be able to encompass the target gamut. I can go that much, or higher; but not lower. I've seen some systems (EFI/Best, for example, starting with v5) trying to determine ink limits automatically, for example by chroma. This fails in too many cases. -- Roberto Michelena Infinitek Lima, Peru