Due to the ever increasing of computing power in the last few decades, the size of scientific data produced by various scientific simulations has been growing rapidly. As a result, effective techniques to visualize and explore those large-scale scientific data are becoming more and more important in understanding the data. However, for data at such a large scale, effective analysis and visualization is a non-trivial task due to several reasons. First, it is often time consuming and memory intensive to perform visualization and analysis directly on the original data. Second, as the data become large and complex, visualization usually suffers from visual cluttering and occlusion, which makes it difficult for users to understand the data.;In order to address the aforementioned challenges, in this dissertation, a distribution-based query-driven framework to visualize and analyze large-scale scientific data is proposed. We propose to use statistical distributions to summarize large-scale data sets. The summarized data is then used to substitute the original data to support efficient and interactive query-driven visualization which is often free of occlusion. In this dissertation, the proposed framework is applied to flow fields and multivariate scalar fields.;We first demonstrate the application of the proposed framework to flow fields. For a flow field, the statistical data summarization is computed from geometries such as streamlines and stream surfaces computed from the flow field. Stream surfaces and streamlines are two popular methods for visualizing flow fields. When the data size is large, distributed memory parallelism usually is needed. In this dissertation, a new scalable algorithm is proposed to compute stream surfaces from large-scale flow fields efficiently on distributed memory machines. After we obtain a large number of computed streamlines or stream surfaces, a direct visualization of all the densely computed geometries is seldom useful due to visual cluttering and occlusion. To solve the visual cluttering problem, a distribution-based query-driven framework to explore those densely computed streamlines is presented.;Then, the proposed framework is applied to multivariate scalar fields. When dealing with multivariate data, in order to understand the data, it is often useful to show the regions of interest based on user specified criteria. In the presence of large-scale multivariate data, efficient techniques to summarize the data and answer users' queries are needed. In this dissertation, we first propose to use multivariate histograms to summarize the data and demonstrate how effective query-driven visualization can be achieved based on those multivariate histograms. However, storing multivariate histograms in the form of multi-dimensional arrays is very expensive. To enable efficient visualization and exploration of multivariate data sets, we present a compact structure to store multivariate histograms to reduce their huge space cost while supporting different kinds of histogram query operations efficiently. We also present an interactive system to assist users to effectively design multivariate transfer functions. Multiple regions of interest could be highlighted through multivariate volume rendering based on the user specified multivariate transfer function.
展开▼
