A study of transformation of primaries and color matching function optimization

The primary purpose of the study was to further understanding of human color perception by evaluating variations in Color Matching Function (CMF) derivation. This included a test of Transformation of Primaries between two primary sets, a test of an optimization technique to derive the functions, and a comparison of the peak and centroid methods for stimulus breakdown. A secondary purpose was to compare the subjective impressions of match quality between two primary sets with predicted match quality according to a color prediction model. Ten subjects completed 260 color matches through ten experimental sessions with two different primary sets, PC (prime color) [453-533-610nm] and NP (non-prime) [487-556-642nm]. From these, a total of 13 sets of CMFs were derived per subject: 4 peak method PC CMFs, 4 centroid method PC CMFs, 4 peak method NP CMFs, and 1 set of CMFs derived to optimize the color prediction accuracy. One PC and one NP holdout sample was also collected. During two sessions, the subjective impression of match quality was obtained. The results show that the PC CMFs are slightly different in shape than the transformed NP CMFs, much less than expected. However, the PC CMFs still were statistically better at match prediction. Contrary to what was hypothesized, the centroid method of stimulus breakdown was significantly worse at predicting color matches than the peak method. On the other hand, the centroid method did give much more repeatable CMFs. In addition, the optimized CMFs greatly outperformed the PC peak method CMFs. Yet the applicability of these functions may be limited due to inconsistencies in the functional form across subject. The study also shows that there was not a statistically significant difference between PC and NP color matches in terms of color difference (i.e. brightness, red/greenness, yellow/blueness, and overall difference). The color prediction model used did not predict the same result. It is likely that the CMFs are not accurate at match prediction even within a single subject