We present a flexible method by which large unstructured scalar fields can be represented in a simplified form. Using a parallelizable classification algorithm to build a cluster hierarchy, we generate a multiresolution representation of the original data. The method uses principal component analysis (PCA) for cluster classification and a fitting technique based on a set of radial basis functions. Once the cluster hierarchy has been generated, we utilize a variety of techniques for extracting different levels of resolution.

A revolution in sensor technology is introducing new problems that must be addressed by visualization research. These technologies will enable earthquake engineers to monitor tectonic plate activity, park rangers to respond more effectively to wildfires, and marine biologists to uncover mysteries from the ocean depths using a network of small, lightweight sensors collecting data in real-time. Visualizing data from these sensor networks is crucial to understanding the information the data represents. However, data collected from such networks are scattered through space and time without explicit connectivity information, posing difficult challenges to researchers and engineers. While past research on such meshless data has focused primarily on interpolation schemes, we propose the development of visualization algorithms and processing frameworks in which existing interpolation schemes can be applied. This work not only plays a significant role in supporting the continued development of sensor network technology; it fills a major gap in visualization research, since effective direct methods for meshless data visualization did not previously exist. Meshless methods can be applied to existing problems encountered in visualizing data with a mesh, giving a set of common techniques that can be applied to a broad class of data sets.

We present a streaming compression algorithm for huge time-varying aerial imagery. New airborne optical sensors are capable of collecting billion-pixel images at multiple frames per second. These images must be transmitted through a low-bandwidth pipe requiring aggressive compression techniques. We achieve such compression by treating foreground portions of the imagery separately from background portions. Foreground information consists of moving objects, which form a tiny fraction of the total pixels. Background areas are compressed effectively over time using streaming wavelet analysis to compute a compact video texture map that represents several frames of raw input images. This map can be rendered efficiently using an algorithm amenable to GPU implementation. The core algorithmic contributions of this work are methods for fast, low-memory streaming wavelet compression and efficient display of wavelet video textures resulting from such compression.

We present a new approach to isosurface visualization that we call ''iso-splatting.'' We use point primitives for representing and rendering isosurfaces. The method consists of two steps. In the first step, point samples are generated throughout the volumetric domain of a scalar function. In the second step, these points are projected onto the isosurface of interest. We render the resulting point set using a surface splatting algorithm. The method can be extended to out-of-core or parallel environments.Our results show that this method can offer much greater time and space efficiency when compared with standard triangle-based methods, thereby supporting higher levels of interactivity. Parts of the algorithm can be accelerated using graphics hardware. One key advantage of this approach is that, since extraction computations are divided into two smaller phases, work can be distributed to exploit all available resources.

We present a method to represent unstructured scalar fields at multiple levels of detail. Using a parallelizable classification algorithm to build a cluster hierarchy,we generate a multiresolution representation of a given volumetric scalar data set. The method uses principal component analysis (PCA) for cluster generation and a fitting technique based on radial basis functions (RBFs). Once the cluster hierarchy has been generated, we utilize a variety of techniques for extracting different levels of detail. The main strength of this work is its generality. Regardless of grid type, this method can be applied to any discrete scalar field representation, even one given as a ''point cloud.''

We present a simple and output optimal algorithm for accelerated isosurface extraction from volumetric data sets. Output optimal extraction algorithms perform an amount of work dominated by the size of the (output) isosurface rather than the size of the (input) data set. While several optimal methods have been proposed to accelerate isosurface extraction, these algorithms are relatively complicated to implement or require quantized values as input. Our method is based on a straightforward array data structure that only requires an auxiliary sorting routine for construction. The method works equally well for floating point data as it does for quantized data sets. We demonstrate how the data structure can exploit coherence between isosurfaces by performing searches incrementally. We show results for real application data validating the method's optimality.

We present an isosurface generation algorithm for large-scale volumetric scattered data. Rather than construct a global tessellation of the data points, we dene a set of local tetrahedrizations that are guaranteed to cover the domain. Inside each local tetrahedrization, a point-based isosurface contouring algorithm is applied to generate isosurface geometry. Our work differs from previous work in that we make very few assumptions about the density of the point distribution, and we construct visual representations of the data directly without global tessellation or resampling. The merit of such an approach is a fairly general method that produces faithful renderings of large-scale scientic data.

We propose a meshless method for the extraction of high-quality continuous isosurfaces from volumetric data represented by multiple grids, also called ''multiblock'' data sets. Multiblock data sets are commonplace in computational mechanics applications. Relatively little research has been performed on contouring multiblock data sets, particularly when the grids overlap one another. Our algorithm proceeds in two steps. In the first step, we determine a continuous interpolant using a set of locally defined radial basis functions (RBFs) in conjunction with a partition of unity method to blend smoothly between these functions. In the second step, we extract isosurface geometry by sampling points on Marching Cubes triangles and projecting these point samples onto the isosurface defined by our interpolant. A surface splatting algorithm is employed for visualizing the resulting point set representing the isosurface. Because of our method's generality, it inherently solves the ''crack problem'' in isosurface generation. Results using a set of synthetic data sets and a discussion of practical considerations are presented. The importance of our method is that it can be applied to arbitrary grid data regardless of mesh layout or orientation.

In this work, we visualize high-dimensional particle simulation data using a suite of scatterplot-based visualizations coupled with interactive selection tools. We use traditional 2D and 3D projection scatterplots as well as a novel oriented-disk rendering style to convey various information about the data. Interactive selection tools allow physicists to manually classify interesting sets of particles that are highlighted across multiple, linked views of the data. The power of our application is the ability to correspond new visual representations of the simulation data with traditional, well understood visualizations. This approach supports the interactive exploration of the high-dimensional space while promoting discovery of new particle behavior.