Except for linear devices like CRTs, color transformations from colorimetric specifications to device coordinates are mostly
obtained by measuring a set of samples, inverting the table, and looking up values in the table (including interpolation),
and mapping the gamut from input to output device.
The accuracy of a transformation is determined by reproducing a second set of samples and measuring the reproduction
errors. Accuracy as the average predicted perceptual error is then used as a metric for quality. Accuracy and precision are
important metrics in commercial print because a print service provider can charge a higher price for more accurate color,
or can widen his tolerances when customers prefer cheap prints.
The disadvantage of determining tolerances through averaging perceptual errors is that the colors in the sample sets
are independent and this is not necessarily a good correlate of print quality as determined through psychophysics studies.
Indeed, images consist of color palettes and the main quality factor is not color fidelity but color integrity. For example, if
the divergence of the field of error vectors is zero, color constancy is likely to take over and humans will perceive the color
reproduction as being of good quality, even if the average error is relatively large. However, if the errors are small but in
random directions, the perceived image quality is poor because the relation among colors is altered.
We propose a standard practice to determine tolerance based on the Farnsworth-Munsell 100-hue test (FM-100) for
the second set and to evaluate the color transpositions-a metric for color integrity-instead of the color differences. The
quality metric is then the FM-100 score. There are industry standards for the tolerances of color judges, and the same tolerances
and classification can be use for print workflows or its components (e.g., presses, proofers, displays). We generalize
this practice to arbitrary perceptually uniform scales tailored to specific applications and present an implementation.
In essence, we propose to extend the color discrimination test procedures used to evaluate human observers, to mechanical
and electronic color reproduction devices.