Dissertation/Thesis Abstract

Deferred Visualization and Interaction with Explorable Images
by Tikhonova, Anna, Ph.D., University of California, Davis, 2012, 159; 3555434
Abstract (Summary)

Interactivity enhances the perception of 3D shape and spatial relationships between features in the data. Thus, it is key to insightful data visualization and exploration. Although recent advances in graphics hardware and visualization technology have enabled interactive exploration of relatively small time-invariant data sets, the data generated by extreme-scale scientific simulations still cannot be conveniently examined due to its size and the costs associated with its transfer, storage, and processing. Many existing large data visualization solutions address this challenge at the expense of losing either accuracy (lossy compression), completeness (temporal and spatial downsampling), coherence (using statistics computed for a small set of time steps rather than statistics computed for an entire time series), or many exploratory capabilities (layer-based and image-based rendering). Another solution is to simply use more computational power, i.e. employ parallel processing, which requires a sufficiently powerful parallel computer, generally shared by many users.

The goal of this dissertation research is to develop a new visualization paradigm that offers interactive visualization of very large data sets at the desired resolution and fidelity. We introduce the term explorable image, which describes a type of media that is compact, efficient to transfer and store, and is designed to facilitate the exploration of specific aspects of the data. The notion of visualization using explorable images relies on the ability to defer operations traditionally used for 3D data exploration to a more suitable intermediate representation for interaction. These operations are performed without access to the original 3D data, re-rendering, and the need for high-end graphics hardware. Such exploratory operations include spatial navigation (rotation, zoom, and translation), transfer function space exploration (modification of opacity and color mappings), and application of ghosted views, cutaways, and various lighting and enhancement effects.

It is not only feasible, but also advantageous to compute explorable images in-situ with a parallel scientific simulation. Scientists often skip tens or even hundreds of time steps due to storage space constraints. Our solution enables them to store more of the data they would otherwise have to discard. Specifically, they can store a combination of raw data and explorable images to facilitate data validation and analysis. The computation and storage of explorable images requires only a small fraction of the total simulation time, I/O time, and storage space.

Deferred visualization and interaction with explorable images is more efficient than existing methods that require access to 3D data or to a large collection of rendered 2D images of the data. Scientists can employ explorable images to examine many aspects of their simulation output on a desktop or laptop computer, mobile device, or within a browser. This suggests a new way of performing in-situ and distance visualization of large-scale data sets. Explorable images, as a type of media, will potentially change the way scientists visualize and explore their data.

Indexing (document details)
Advisor: Ma, Kwan-Liu
Commitee: Amenta, Annamaria B., Carmichael, Owen T., Ma, Kwan-Liu
School: University of California, Davis
Department: Computer Science
School Location: United States -- California
Source: DAI-B 74/07(E), Dissertation Abstracts International
Subjects: Computer science
Keywords: Deferred visualization, Explorable images, Image-based rendering, Proxy images, Visualization by proxy, Volume visualization
Publication Number: 3555434
ISBN: 9781267969910