Toward a Perceptual Science of Multidimensional Data Visualization: Bertin and Beyond

A true science of data visualizations requires both a theory of perception and of computer graphics. However, visualization designers have paid relatively little attention to perceptual issues. In this article, I outline how knowledge of human visual perception and physiology can lead to more effective visualizations. Bertin's (1983) Image Theory, the only comprehensive perceptual theory in the visualization literature, will serve as the medium for the discussion. Experimental vision research grounds Image Theory in "first-principles" and suggests corrections, modifications and extensions. The resulting updated version of Image Theory can serve as a guide to visualization design. 1.0 Data Visualization Is A Joint Function of Computer Graphics and Perception "The Purpose of computing is insight, not numbers." --Richard Hamming (1962) The power of computers to collect, store and manipulate numbers has increased dramatically since Hamming's pointed observation. Much of this increased power, however, is wasted because humans are poor at gaining insight from data presented in numerical form. As a result, visualization research takes on great significance, offering a promising technology for transforming an indigestible mass of numbers into a medium which humans can understand, interpret and explore. The transformation from numbers to insight requires two stages. As shown in Figure 1, the first maps from numbers (data/processes) to images by means of some algorithmic technique. The second maps from images to insight by means of perception. A true science of visualization must incorporate both a formal theory of computer graphics and a theory of human perception. However, visualization research, like most areas of computer graphics, has focused largely on image formalisms and has ignored issues relating to visualization and perception (Greenberg, 1988; Blake, 1990; Pun and Blake, 1990). Papers published in proceedings of visualization conferences, computer graphics journals, etc. are typically written by computer scientists who describe new software for mapping data to images but who ignore the mapping from images to perception. They offer neither a coherent rationale for the creation of a particular visualization nor any empirical validation. They employ no explicit model of perception, relying instead on intuition and introspection. In sum, they are concerned only with questions of what could be done and ignore issues of what should be done. This is unfortunate, since empirical studies repeatedly find that design evaluations based on simple introspection and preference are poor predictors of user performance.