Spectral Gamut Mapping Framework based on Human Color Vision

A new spectral gamut mapping framework is presented. It adjusts the reproduction, choosing spectra within the printer's gamut that satisfy colorimetric criteria across a hierarchical set of illuminants. For the most important illuminant a traditional gamut mapping is performed and for each additional considered illuminant colors are mapped into device and pixel dependent metamer mismatch gamuts. A computational separation method is proposed in order to test the framework. Utilizing this separation method on a seven channel printing system, experiments allowed a deeper view on the structure of the device and pixel dependent metamer mismatch gamuts and the possible directions in color space in which a potential metameric gamut mapping transformation could map out-of-metameric-gamut colors. Introduction In recent years spectral acquisition has become an active research eld. Today's technology is able to capture high resolution multichannel images with very small spectral estimation error. This technology is employed by museums for artwork reproduction and for archiving applications. Wide-gamut multicolorant printers are used within traditional color management to create accurate colorimetric reproductions that match originals under a single illuminant. For some applications it can be desireable for reproductions to match originals under multiple illuminants. In such cases a spectral reproduction is needed. A basic limitation of such a reproduction is the physical ability of printing devices to reproduce re ectances. The spectral printer gamut is much smaller than the space of all natural re ectances. A lower bound of the dimensionality of natural re ectances can be determined through analysis of multiple spectral databases [1]. Only by looking onto the dimensionality difference does it become obvious that the majority of spectral re ectances cannot be reproduced without spectral error on a typical printer. It becomes necessary to map the unreproducible spectra into the spectral gamut of the printer. Such a mapping is not unique and an optimal transformation strongly depends on the special application. In recent years various metrics in spectral space have been proposed [2, 3]. To show an advantage compared to traditional color management, spectral reproduction should be for one illuminant as visually correct as a colorimetric reproduction and for other illuminants superior. An approach has been proposed, combining a mapping in a perceptual color space based on one illuminant and a spectral mapping within the corresponding three dimensional device metameric black space [4, 5, 6, 7]. In this paper we are presenting a spectral gamut mapping framework that adjusts a reproduction so that it matches the original under multiple illuminants considering properties of human color vision. The Spectral Gamut Mapping Framework Terminology In order to explain the spectral gamut mapping framework we use the common terminology of discrete spectra, resulting from a sampling of the continuous spectra at N equidistant positions within the visible wavelength range from 380 nm to 730 nm. Each re ectance spectrum is a N-dimensional vector r ∈ [0,1]N and the set of illuminants for which the reproduction has to be adjusted is a set of N-dimensional vectors representing the spectral power distributions of the illuminants I1, . . . , In ∈ RN . In the following text we use the observer's CIEXYZ tristimulus X(r, I) as a function of the re ectance r = (r1, . . . ,rN) and the illuminating illuminant I = (I1, . . . , IN): X(r, I) = 1 ∑i=1 ȳiIi ( N ∑ i=1 x̄iIiri, N ∑ i=1 ȳiIiri, N ∑ i=1 z̄iIiri )T (1) where x̄, ȳ, z̄ are the CIE color matching functions for the 2◦ or 10◦ observer, respectively. The color space transformation from CIEXYZ into the nearly perceptually uniform CIELAB color space is denoted by L : CIEXYZ 7→ CIELAB and the inverse function by L−1 : CIELAB 7→ CIEXYZ. The set of all re ectances that result in the CIELAB value x for an illuminant I is called metameric re ectance set and will be denoted by M(x, I) = {r ∈ [0,1]N | L(X(r, I)) = x} (2) The spectral printer gamut, which is the space of all printable spectral re ectances of the given device, is denoted by G ⊂ [0,1]N .