Color-difference evaluation for digital images using a categorical judgment method.
Haoxue Liu,Min Huang,Guihua Cui,M Ronnier Luo,Manuel Melgosa
DOI: https://doi.org/10.1364/JOSAA.30.000616
2013-01-01
Journal of the Optical Society of America
Abstract:The CIELAB lightness and chroma values of pixels in five of the eight ISO SCID natural images were modified to produce sample images. Pairs of images were displayed on a calibrated monitor and assessed by a panel of 12 observers with normal color vision using a categorical judgment method. The experimental results showed that assuming the lightness parametric factor k(L)=1 to predict color differences in images, CIELAB performed better than CIEDE2000, CIE94, or CMC, which is a different result to the one found in color-difference literature for homogeneous color pairs. However, observers perceived CIELAB lightness and chroma differences in images in different ways. To fit current experimental data, a specific methodology is proposed to optimize k(L) in the color-difference formulas CIELAB, CIEDE2000, CIE94, and CMC. From the standardized residual sum of squares (STRESS) index, it was found that the optimized formulas, CIEDE2000(2.3:1), CIE94(3.0:1), and CMC(3.4:1), performed significantly better than their corresponding original forms with lightness parametric factor k(L)=1. Specifically, CIEDE2000(2.3:1) performed the best, with a satisfactory average STRESS value of 25.8, which is very similar to the 27.5 value that was found from the CIEDE2000(1:1) formula for the combined weighted dataset of homogeneous color samples employed at the development of this formula [J. Opt. Soc. Am. A25, 1828 (2008), Table 2]. However, fitting our experimental data, none of the four optimized formulas CIELAB(1.5:1), CIEDE2000(2.3:1), CIE94(3.0:1), and CMC(3.4:1) is significantly better than the others. Current results roughly agree with the recent CIE recommendation that color difference in images can be predicted by simply adopting a lightness parametric factor k(L)=2 in CIELAB or CIEDE2000 [CIE Publication 199:2011]. It was also found that the different contents of the five images have considerable influence on the performance of the tested color-difference formulas.