, Canon Information Systems Research Australia
, Fraunhofer Institute for Computer Graphics, Germany
Pages: pp. 18-19
It's 20 years or more since the first papers specifically addressing color in computer graphics appeared, primarily on color transforms and reproduction. Since then, a small number of important contributions have addressed both these and broader aspects of color, including color coordinate conversions, perceptual color spaces, spectral color calculations, and color gamut mapping. Color as a focus in graphics, however, has not really drawn sustained attention from researchers.
For this special issue of IEEE CG& A we have drawn together some of the main color issues facing computer graphics practitioners. The development of digital cameras, the demand for increasingly realistic (or at least plausible) rendering, and the integration of data acquired by cameras with synthesized data have all emphasized the need for careful control of color in both synthesis and reproduction. Improved physical modeling of light—and of reflection and transmission—has aided the development of color management tools. Color management systems have emerged for the more straightforward color matching problems. But we still experience color reproduction or synthesis problems that prove surprisingly difficult to remedy. These problems expose both the limitations of our current tools and systems, and our understanding of these limitations.
So how do graphics practitioners decide on their color management or synthesis approach? How do we assess, for example, whether to perform color computation based on the primaries of different color presentation devices or systems, or based on perceptually defined primaries? How do we assess or predict the impact of this on the end result or even measure this impact? How do we determine the most suitable spectral sampling schemes for new acquisition or presentation devices? And how do we make even the most basic decisions about what colors to use for different presentation purposes?
Several factors repeatedly emerge:
The articles chosen for this issue address these main problem areas, with perspectives ranging from theory through heuristics to practice. They provide insight not only into the causes of problems experienced, but also into new approaches that may help redress the limitations of color synthesis and reproduction systems.
In a broad and pragmatic tutorial, MacDonald covers basic perceptual and device issues. He also provides appropriate heuristics for use of color based on physics and psychophysics.
Hall addresses the impact of the choice of color sampling representations on rendering, providing objective comparisons to help understand this impact. These results show the importance of carefully considering the choice of coordinate system and the need to have a choice of coordinate systems available for different applications.
In the past researchers at times have used, or proposed using, "full" spectral calculations. In some cases they used approximations with a number of samples larger than three across the spectrum, for example at 20-nm intervals, to achieve better accuracy. They sometimes targeted a given level of perceptual accuracy, although usually on the basis of heuristics. Johnson and Fairchild suggest moving towards spectra-based color synthesis to reduce errors and metamerism problems by using a multi-sample system. They compellingly illustrate some possible consequences of not doing this, including for synthesis of fluorescence.
Sometimes simple solutions work. Sun, Fracchia, Calvert, and Drew show how effectively a physically rationaled heuristic can improve standard three-sample approximations. Simply changing the width of a Gaussian used to model spectra, based on the degree of saturation, can achieve a surprising improvement in modeling physical phenomena such as interference. It would be interesting to apply this technique more generally to help with color management systems, for example for reproduction between devices that have primaries with different spectral characteristics.
Finally, Emmel and Hersch move toward developing a unified model for physical effects in printing. This is a very difficult problem, but the benefits of achieving integration of measurement and modeling could be substantial in the digital printing industry. Their article takes the first step toward achieving this goal, but it will need further research before it can be applied in practice.
Overall, the basic message from these collected works is that for future rendering involving both synthetic and acquired data, we need to consider the approximations arising from three-component sampling much more carefully. Until now we have largely allowed computational convenience to outweigh quality of result. But the combined demands of increasing realism, and requirements for reproduction across a greater range of devices and media, force us to redress this. The authors of the articles in this issue have shown clearly the major areas where problems can arise and the directions toward solving them.
What can we infer for the future? Clearly we need graphics systems to support both greater flexibility in choosing sampling representations and fuller spectral computations. This shouldn't prove too difficult to achieve, provided that emerging standards for color representation can accommodate this flexibility. And we need to bring greater rigor into our perceptual selections of color, choice of coordinate system, and assessment of the impact of computational approximations.
Further into the future we might challenge device makers to work on low-cost, multispectral sampling systems or advanced materials for greater control over spectral characteristics. The graphics community occupies a unique position for influencing device makers because it generates many of the most demanding applications. Let's make sure we understand, illustrate, and champion the importance of such advances.