The Hubble Space Telescope has provided the world with incredible imagery of the surrounding universe. The aesthetic quality of this imagery is limited by production resources; by creating a method to harness the highly refined detail and quality of CG elements in live-action films, we can inspire and educate at a much greater level. In this thesis, I create a rendering approach that allows camera movement around and through elements such as nebulae and galaxies, creating a more cinematic experience. The solution will also allow for reasonable scientific accuracy, visual appeal, efficiency, and extendability to other astronomical visualizations. 3D meshes are constructed and textured using telescopic images as reference. Splats are volumetrically generated using a voxelized bounding box around the mesh. Valid splats within a user specified maximum distance receive initial color and alpha values from the texture map. Probability density functions are used to create a density falloff along the edges of the object, and modifications to the RGBA values are made to achieve the desired cloud-like appearance. The data sets are rendered using a C program developed at the Space Telescope Science Institute by Dr. Frank Summers. The methodology is applied to the test cases of a nebula, star-forming region Sharpless 2-106, and a galaxy, Messier 51, or the Whirlpool Galaxy. The results of this thesis demonstrate the visual, scientific, and technical success of this solution. The code developed during this project generates the desired imagery with reasonable efficiency. A short animation moving from outside the galaxy to a close up of the nebula exhibits the flexibility in scale and camera movement. A careful balance between scientific accuracy and visual appeal were maintained through consultation with astronomers at the Space Telescope Science Institute. The favorable efficient, flexible, visual, and scientific results presented by this work make this process extendable to most other cases of nebula and galaxy visualizations.

This work presents the new Synthesized Cell Texture (SCT) algorithm for visualizing related multiple scalar value fields within the same 3D space. The SCT method is particularly well suited to scalar quantities that could be represented in the physical domain as size fractionated particles, such as in the study of sedimentation, atmospheric aerosols, or precipitation. There are two components to this contribution. First a Scaling and Distribution (SAD) algorithm provides a means of specifying a multi-scalar field in terms of a maximum cell resolution (or density of represented values). This information is used to scale the multi-scalar field values for each 3D cell to the maximum values found throughout the data set, and then randomly distributes those values as particles varying in number, size, color, and opacity within a 2D cell slice. This approach facilitates viewing of closely spaced layers commonly found in sigma-coordinate grids. The SAD algorithm can be applied regardless of how the particles are rendered. The second contribution provides the Synthesized Cell Texture (SCT) algorithm to render the multi-scalar values. In this approach, a texture is synthesized from the location information computed by the SAD algorithm, which is then applied to each cell as a 2D slice within the volume. The SCT method trades off computation time (to synthesize the texture) and texture memory against the number of geometric primitives that must be sent through the graphics pipeline of the host system. Analysis results from a user study prove the effectiveness of the algorithm as a browsing method for multiple related scalar fields. The interactive rendering performance of the SCT method is compared with two common basic particle representations: flat-shaded color-mapped OpenGL points and quadrilaterals. Frame rate statistics show the SCT method to be up to 44 times faster, depending on the volume to be displayed and the host system. The SCT method has been successfully applied to oceanographic sedimentation data, and can be applied to other problem domains as well. Future enhancements include the extension to time-varying data and parallelization of the texture synthesis component to reduce startup time.