Perceptual Approaches to Visualizing Data

Bernice E. Rogowitz

IBM T.J.Watson Research Center
P.O. Box 704
Yorktown Heights, NY 10598
rogowitz@watson.ibm.com
914-784-7954

The human observer is an active information-seeker. As we make our way through our environment, we recognize patterns, categorize events into categories, develop hypotheses about relationships, and use new information to modify and tune these hypotheses. In this class, I will describe some of our work designing computer-based systems which take advantage of the natural capabilities of the human observer.

In the two systems I'll describe, we model the user's task as an exploration process, where the user iteratively designs the solution to a problem. In the first system, Diamond, the user's task is to analyze tabular data; in the second system, PRAVDA, the user's task is to build a visual representation of scientific data.

In Diamond we depict data and statistics as pictures, and let users interactively manipulate these pictures to discover patterns and relationships in the data. To do so, we have used perceptual and cognitive principles to create visual representations and visual operations which help the user understand correlations, interactions and patterns in multivariate data.

In PRAVDA (Perceptual Rule-Based Architecture for Visualizing Data Accurately), we take this idea a step further by explicitly adding perceptual rules to an interactive visualization system. For example, most visualization systems offer users a default colormap. This rainbow colormap, however, produces a number of perceptual artifacts which can misrepresent the information in the data. In PRAVDA, the user is offered a set of perceptually-appropriate colormaps based on properties of human color vision, properties of the data, and the users' task.