Metrics for evaluating spectral matches: A quantitative comparison

Abstract

Several of the spectral match metrics considered by Imai, et al., are compared for a large number of non-metameric pairs of spectra in order to assess how accurately they track human perception as predicted by CIELAB (i.e., the extent to which the metric will exhibit a proportional relationship with CIELAB total color difference), and how precisely they do so (i.e., the relative compactness of the distribution of a metric for spectral pairs which differ by a given level of CIELAB total color difference). Both properties are important attributes of a spectral match metric. Of particular importance in optimization problems is the precision as the total color difference becomes small. We found that among the metrics considered, only unweighted RMS and Viggiano's Spectral Comparison Index provided precision for both large and small color differences. The Viggiano Spectral Comparison Index had the closest correlation to human perception, and, for the non-metameric spectral pairs examined in this study, assumed values close to 2,6 times that of CIELAB ΔE*. The paper includes a definition of non-metameric spectra, and describes a method for generating them, which include variations in Lightness, Hue, and Chroma.