Visualization is the process of creating a visual representation of data. This representation is often very complex. It may include several different types of data, there may be several relationships the user wants to explore or communicate, and the standards for aesthetics in visualization are always increasing. In some cases, the visual representation helps the user interpret and understand the phenomena being represented; in some cases the visual representation obscures the meaning and prevents the user from generating hypotheses about the underlying phenomena.
In order for visualization to become more than a sophisticated craft, the principles which help the user create effective representations need to be elucidated and incorporated into the visualization tools. One area of scientific promise is in incorporating knowledge about human perception into the design of these tools, knowledge which helps the user create a visual representations which do not include visual artifacts, and which communicates the structure in the data.
The need for such tools can be easily overlooked because the tools we have today seem to be designed to fill this need. Most visualization systems, for example, allow the user to automatically assign a colormap to the data. Using the common rainbow-hue map from blue to red, a typical 2-d array of temperature values would be mapped onto a 2-d array of points whose "color" is proportional to the value of the temperature. This is mathematically correct. However, requiring that the color scale increase continuously and monotonically with temperature does not guarantee that these values will appear to do so. In the case of the rainbow colormap, perceived color increases in a stepwise manner, producing bands of color instead of a continuous gradation, and the yellow regions can appear to have higher temperature than the red regions because, although they are lower in hue value, they are brighter. If the user wants to have a visual representation which preserves the structure in the data, it is important to choose a color map in which equal steps in data value correspond to equal steps in perceptual value. Creating such colormaps involves the incorporation of guidance from the psychophysics of vision.
This, of course, is only the tip of the iceberg. In some cases we want to call attention to a particular feature in the data, segment the data into ranges, color features in the data according to semantic categories, or simply dazzle the audience with the sheer drama of our result. What, then, are the principles which describe how attention is directed, what cues visually segment data, what design principles determine which color combinations will be judged as subtle or dramatic? How do these principles interact? How do they depend on data type, domain, and the task the visual representation is trying to address? How do they depend on the individual, the context or the culture?
Research into these questions will, I believe, help us form a true science of visualization, where the use has control over the visual representations which are used to explore and describe data.