We use the term visualization metrology to refer to the process of assessing uncertainties associated with the visualization of scientific data. Creating visualizations of data produced by physical or computational experiments can involve various transformations of those data. Each of these transformations may introduce uncertainties in the final visualization. We seek to quantitatively assess these uncertainties in visualizations and in data derived from visualizations.
The visual display of virtual 3D scenes has become commonplace in modern computing systems. While many such visualizations are intended to provide a qualitative experience to the viewer, visualization is increasingly being used to convey accurate representations of spatial relationships. Applications such as computer-aided design and scientific data visualization have important quantitative components. For example, at NIST, we have implemented interactive measurement tools in the virtual world.
As we use such systems for these sorts of quantitative tasks, it is incumbent upon us to understand how the visualization process contributes to uncertainty in these tasks. This work is a first step toward this understanding.
We have focused first on assessing errors introduced by the rendering process. Rendering is the process by which we transform a 3D geometric description of a virtual scene into a set of pixels that are displayed to the user on a computer monitor. Our next effort has been the measurement of errors in polygonal representations of surfaces.
Future work will involve the assessment of uncertainty introduced by other aspects of the visualization process.
Methods for Quantifying and Characterizing Errors in Pixel-Based 3D Rendering