Device-Independent Color via Spectral Sharpening

Color sensors in scanners and color copiers are not colorimetric | RGB values are not a linear transformation away from device{independent XYZ tristimu-lus values. For a given set of targets or dyes one can readily nd a best linear transform or use interpolation. However, when the possible targets are unknown, a data-independent transform is needed. Here, we set out a very simple linear transform for forming XYZ from RGB, developed in analogy with a well-known solution for the color constancy problem in computer vision, based on using narrow-band sensors. In a scanner, we know the illuminant. Therefore the color constancy paradigm|illumination{ independent colors|is not applicable. Instead, we change lters|from RGB to XYZ. The von Kries adaptation form of the color constancy solution can then apply if we can \sharpen" both the RGB sensors and the XYZ color{matching functions. Recently, we developed just such a \sharpening" basis transform: most of the sensitivity of the new possibly partly negative sensors is isolated in a particular wavelength interval. Here we \sharpen" both sensor sets; after dividing by sharpened white{spot values an inverse transform results in recovered XYZ values. Applying the method to 462 Munsell chips yields a median CIELAB error of only 3 units for two diierent systems.