Harnessing natural textures for multivariate visualization

n our ongoing quest to convey more information more clearly in a single image, harnessing the full potential of texture for data representation remains an elusive goal. Others have begun excellent work in this area, 1-3 and my efforts are inspired by their example. The grail that I seek is a partially ordered multidimensional palette of richly detailed and varying texture patterns that can be used—in conjunction with lightness and hue—to represent multivariate information. The goal is to facilitate the flexible visual appreciation of the correlations of various quantities across the different dimensions. The approach that I outline here departs a bit from the norm, but is motivated by a desire to proceed more directly from my vision of what I want to achieve, unrestrained by the limitations of the tools I have on hand. In the following discussion, I motivate the adoption of rich, natural textures—resembling those from photographic images 4 —as elemental primitives and sketch some of the approaches that we can take to enhance our understanding of how to effectively harness their properties. My intent here is not to present results, but to expound on the issues and conclude with the questions to which we're still seeking answers. The intricate variety and subtle richness of detail of texture patterns found in nature support possibilities for data representation far more vast and comprehensive than we could ordinarily hope to achieve from standard primi-tives. Even if we must ultimately rely on synthesized textures for data visualization, by looking to nature for inspiration we have the potential to expand our vision of what to strive for in such a synthesis. The graphic design community has long held that perfectly regular synthetic textures on a flat plane, in particular the infamous hatching patterns that Edward Tufte refers to as " chart junk, " 5 are discomforting to the eye and annoying to look at. Natural textures are not only more aesthetic, but they also put less extraneous stress on the visual system, leaving our eyes freer to observe and attend to the most intrin-sically important texture-pattern characteristics. To create a perceptually meaningful multidimen-sional texture space that can be indexed in the same fashion as a color space, we must begin by knowing what we're looking for. We need to proceed from a rigorous and experimentally supported understanding of how human observers perceive and interpret texture patterns , under the conditions in …