News | People | Our Github

seeDQ - A Query Language for Visualizing Probabilistically-based Features


Feature detection and display are the essential goals of the visualization process. Most volume visualization software achieves these goals by mapping properties of sampled intensity values and their derivatives to color and opacity. However, by considering samples in isolation, it is difficult to visualize features that are not surface-based. In this work, we expand the basis of classification to broader neighborhoods centered around each voxel. Our approach allows users to enter predicate-based hypotheses about relational patterns in the local neighborhoods' frequency distributions and render visualizations that show how neighborhoods match the predicates (illustrated above). We have built a simple graphical user interface for forming and testing queries interactively and we describe a volume rendering algorithm tailored for investigating statistical queries. The query framework readily applies to spatial datasets from arbitrary domains and supports queries on time variant and multifield data. Users can directly query for classes of features previously inaccessible in general feature detection tools.

The above image is the result of a query for temporal neighborhoods in a 100-year climate simulation using a leading United Nation IPCC model to quantify patterns in temperature change in relation to precipitation change. This image shows the difference between 2000-2009 and 2090-2099. The exact size and location of the temporal neighborhood, the degrees of temperature change as a query threshold and the fact that absolute amount of precipitation change should not matter in this hypothesis, were all result of interactive visualization and user controlled exploration.

'Distribution Driven Visualization of Volume Data', C. Ryan Johnson and Jian Huang, accepted (in press), IEEE Transactions on Visualization and Computer Graphics, ?(?), 2009.