Assessment of Metrics in Color Spaces

Abstract

The perceptibility and acceptability of small color differences (ΔE* <5) were investigated using 66 textile-based color difference pairs around a CIE blue color center. Samples were polyester fabrics dyed with disperse dyes under well-controlled dyeing procedures and using precise recipe prediction methods. Each pair was assessed three times by 26 observers in three separate repetitions using gray scale and a total of 5148 assessments were obtained. A standard viewing booth was used with uniform gray surrounding corresponding to L*=74. This resulted in minimizing the crispening effect due to the large lightness difference between the surround and the samples, all of which had an L* value of approximately 35. A third-degree polynomial equation was used to convert gray scale ratings to visual differences (DV). SAS statistical analysis showed over-fitting of data when a fourth-degree polynomial equation is used at 95% confidence level. However, it was shown that a third-degree polynomial fitting was satisfactory. Third-degree polynomial fitting equations were obtained using CIELAB, CIE94(1:1:1), CMC(1:1), and CIEDE2000(1:1:1), respectively, and the results were found to be very similar for all equations. Inter and intra-group variability was assessed using paired t-test and PF⁄3, and the results indicated observer training through repeat assessments.

The performance of color difference formulae was evaluated using a correlation coefficient (r) and PF⁄3 at two lightness weights (KL or l) of 1 and 2. Correlation coefficients of 0.91, 0.92 were obtained for CIEDE2000 at KL of 1 and 2, respectively indicating that CIEDE2000 gives the best performance amongst the color difference equations examined. Using PF⁄3, the BFD equation was found to give the best results at both lightness scales giving values of 37.46, and 42.46, respectively. PF⁄3 results also indicate that at KL = 2, CIEDE2000 gives the worst performance amongst equations examined. In addition, results based on correlation coefficient and PF⁄3 evaluation do not match well especially at lightness scale of 2. The optimal KL was also obtained for various color difference formulae using a range of KL values within 0.2-3.0 and with an interval of 0.1. Using PF⁄3 as a metric the best formula performance was obtained when KL or l was set to 0.99, 1.04, 1.21, and 1.33 for CIEDE2000 (KL:KC:KH), BFD(l:c), CMC(l:c), and CIE94(KL:KC:KH) formulae, respectively.

The best-fit ellipsoid equation in the CIELAB color space was obtained for the Lee-Shamey Blue dataset developed in this study. This ellipsoid was compared with ellipsoids optimized to the RIT-DuPont dataset. The ellipses based on Lee-Shamey blue dataset have the smallest ratio of axes in the CIE a*b* as well as CIE a*L* and b*L* planes (2.73, 1.16, and 1.58, respectively), with an area of (0.93, 0.45, and 0.78, respectively). The dataset was also converted to CIExyY coordinates for comparison with other published data. In the chromaticity-discrimination ellipse, the ellipse based on Lee-Shamey Blue dataset has the smallest ratio of axes (4.27) as well as area (0.08) among the ellipses obtained from RIT-DuPont, Witt, and BFD datasets.