Thursday, December 10, 2009

Perceptual vs. colorimetric color spaces

In yesterday's post I noted that the authors had used a perceptual color space instead of a colorimetric one for their study. What is the difference?

It is already half a dozen years ago that I was assigned a project related to metamerism. To solve the open-ended problems, I implemented an extensive library for color science. For one specific solution I used a perceptual color space based on the McLeod's approach in making calculations using cone responses.

cone and rod responses

A principal color scientist then noted that perceptual color is used in vision research, but in engineering we use CIE based color spaces. Since my library was object oriented, all I had to do was to change a single import statement so that the CIE color matching functions (CMF) would be used as measures in the integration of the spectral data, instead of the cone fundamentals.

However, this minuscule engineering change has a big theoretical effect, because the CMF cover almost the entire visual domain, while the cone responses cover only a fraction of it. Form a signal processing point of view, the result is that that CIE color spaces have more signal modulation than perceptual spaces. Numerically the effect of this change in my code was small and irrelevant for the solution, but it is a big effect in terms of the underlaying science: when writing a paper the distinction is very important.

color matching functions


Now to my question.

I inherited a manuscript from a discouraged color scientist who gave up after years of rejection. I am now revising it in the hope of having it published, because the work has archival importance. The research is based on calculations in CIELAB. One of the reviewers writes:

"I object to […] their use of CIE modeling […], given that CIE did not design those models to serve as any sort of color appearance processing framework, but rather provided them as rules for engineers who want to reproduce chromaticities across different labs.

[…]

"Based on the history of the development and use of CIE, I believe it is suboptimal to model color appearance using CIE light mixture spaces (even the ones like CIELAB that claim to be approximately perceptually uniform). That is, using appropriate CIE chromaticities to report what stimuli are used in experiments is a fine practice, but for modeling and predicting human response to color it is best to go with psychophysically modeled spaces and the associated indices like Weber contrast, Michelson contrast or RMS contrast. To the degree that you continue to model human response with the CIE engineering emphasis, I think your results may be useful for rendering color in displays, but will tell us far less about the human response of [the studied effect]. I would suggest removing any CIE modeling of human response from your report and expanding explanations based on psychophysical modeling."

The implication here is that CIE colorimetry is not based on psychophysics. If I had the raw data, I might just have recomputed all quantities using cone responses, but unfortunately the raw data appears to be lost.

The only way out of this conundrum is to fully understand the difference between perceptual models and colorimetric models, then to provide a solid explanation. At this point I am seeking either an explanation or a good pointer to something I should read and understand.

Can you help?

conceptual CMF experiment