VolDG - Design Galleries for Volume Graphics
Direct volume rendering is a key technology for the visualization of large 3D datasets from scientific or medical applications. Of particular importance to the quality of direct volume-rendered images are transfer functions. A transfer function assigns optical properties, such as color and opacity, to original values of the dataset being visualized. Unfortunately, exploring different transfer functions is a tedious task, often done by trial and error. Managing and organizing the exploration of transfer-function space is usually left to the user; the computer is used as a passive instrument. In the Design Gallery for Volume Graphics (VolDG) approach the parameter-setting task is divided more evenly between user and computer. VolDG interfaces present the user with the broadest selection - automatically generated and organized - of perceptually different images that can be produced by varying transfer functions. Our current system is built on top of the popular Visualization Toolkit (vtk) and Mitsubishi Real Time Visualization's VolumePro 500(tm) board. The real-time volume-rendering speed of the VolumePro 500(tm) board allows large galleries to be generated in minutes.
Background & Objective: The trick in producing compelling and useful visualizations is to find input parameters that yield desirable output values. Tweaking input parameters to this end is a tedious experience familiar to anyone who has created computer graphics and scientific visualizations. VolDG presents a viable alternative to facilitate parameter selection. Instead of asking the computer "What's best?" we ask the computer "What's possible?" The computer's task is to pick a set of input-parameter vectors that spans the space of output values as much as possible; the user's task is simply to select from among the presented possibilities.
Technical Discussion: The principal technical challenges posed by the VolDG approach are dispersion (finding a set of input-parameter vectors that optimally generates very dissimilar output values) and arrangement (arranging the resulting designs for easy browsing by the user). For dispersion we use a form of evolutionary computation. For arrangement we use multidimensional scaling. Details of our technical approach are described in a 1997 SIGGRAPH conference paper available at http://www.merl.com/projects/dg/. The dispersion process can require the rendering of hundreds or thousands of candidate images, and therefore benefits greatly from hardware acceleration by VolumePro 500(tm).
Contact: Joseph Katz
Technology Area: Graphics
Modification Date: September 14, 2007
