A key component of an effective visualization involves the visual representation of the data. The representation determines how the items in the dataset are rendered on the computer display. The best representations are often domain-specific: scatterplots for statistical data, maps for spatial data, and node and link diagrams for network data, for example. Inventing a representation for a new domain is a difficult, creative, and iterative process.1 The representation should take full advantage of perceptual cues such as size. positions. color, depth, and may even use motion and sound.
Our representations are often compact, color-coded glyphs positioned spatially. By using compact glyphs that overplot gracefully we can pack a lot of information into an image and thereby display a large dataset. A high-resolution 1280×1024 workstation monitor has over 1,300,000 pixels. Our goal is to use every pixel to display data, thereby maximizing the information content. in the image.
In some cases is is possible to display an entire dataset on a single screen, thereby eliminating the difficult navigation problems associated with panning and zooming interfaces that focus on small portions of the database.
Often information-dense displays become overly cluttered with too much detail. One approach to solving the display clutter problem involves interactive filters that reduce the amount of information shown on the display. Humans have sophisticated pattern recognition capabilities, perhaps due to our evolution, and are very efficient at manipulating interactive controls to reduce visual clutter. We exploit this to effortlessly solve the complex computational problems involved with determining when a display is too busy for an easy interpretation. Our approach is to leverage people's natural abilities by designing user interface controls that parameterize the display complexity.