HR: 1340h
AN: IN43B-0339    [Abstracts]
TI: Exploring Tensor Fields Using a Fabric Like Texture on Arbitrary Surfaces
AU: * Hotz, I
EM: ihotz@ucdavis.edu
AF: Institute for Data Analysis and Visualization, (IDAV), University of California, One Shields Avenue, Davis, CA 95616 United States
AU: Feng, Z L
EM: zfeng@ucdavis.edu
AF: Institute for Data Analysis and Visualization, (IDAV), University of California, One Shields Avenue, Davis, CA 95616 United States
AU: Hamann, B
EM: hamann@cs.ucdavis.edu
AF: Institute for Data Analysis and Visualization, (IDAV), University of California, One Shields Avenue, Davis, CA 95616 United States
AU: Manaker, D M
EM: manaker@email.geology.ucdavis.edu
AF: Geology Department, University of California, One Shields Avenue, Davis, CA 95616 United States
AU: Conjeepuram, N S
EM: nsconjeepuram@ucdavis.edu
AF: Geology Department, University of California, One Shields Avenue, Davis, CA 95616 United States
AU: Kellogg, L H
EM: kellogg@geology.ucdavis.edu
AF: Geology Department, University of California, One Shields Avenue, Davis, CA 95616 United States
AB: Many applications in geophysics involve computing not only scalar and vector fields but also tensor fields. Examples include projecting the postseismic stress change from one fault onto nearby faults to assess seismic hazard, and calculation of stress in geodynamical models of time-dependent damage. The graphical representation of data and models, however, is typically restricted to either scalar fields, such as a damage parameter, vector fields, such as velocity, or to representation of tensors at isolated points, such as earthquake focal mechanism. But there is much more information contained in the tensor fields. For continuous three-dimensional complex datasets it may not be obvious which locations to choose and thus the exploration of the entire data set is very cumbersome. A visualization method giving an intuitive overview of the tensor field allows an effective detection of interesting regions, which than can be investigated with classical methods. We have developed a tensor field visualization method providing a continuous representation. The basic idea is to bend, stretch, and compress a fabric-like texture, that is dense in regions of compression and sparse in regions of expansion. The texture parameters, i.e., fiber density and fiber direction, are controlled by the tensor field. The texture is composed of two families of fibers, generated using line integral convolution (LIC). This method uses a random noise image as input that is blurred along a direction field, in our case the principal directions of the tensor field. A regular, homogeneous input image with circular spots results in fibers with constant density and fiber width. The impression of stretching or compressing can be achieved by changing density, shape, and size of the spots. Like most texture-based methods, this approach is basically limited to two dimensions due to occlusion problems. To be able to explore the entire three-dimensional volume, we define a one-parameter family of surfaces such that the union of the surfaces covers the entire domain. In our implementation, we support two ways to define such surfaces. The first possibility is to define the surfaces explicitly, based on a geometric property or symmetry inherent to the structure of the data. Simple possibilities are "moving planes", cylinders, or parts of spheres. More complex surfaces defined by faults or moving objects like hot buoyant fluid sphere rising in a colder matrix, can also be used. The second option is to define the surfaces implicitly using isosurfaces of an additional connected scalar field, e.g., pressure, temperature, or damage. In the final visualization we show the deformed fabric-texture on a surface moving through the volume. To compute the textures on the surfaces we project the tensor field onto the surfaces. We provide two ways of looking at these surfaces. The first uses volume rendering based on a transfer function illustrating the volume only in a small neighborhood of the chosen isovalue. The second approach is an explicit extraction and visualization of the surfaces. The most expensive part of the texture generation is the computation of the input spot image for the convolution step. This computation can be done in a pre-processing step. The three-dimensional domain can thus be examined easily using different sets of surfaces. This method supports an intuitive representation of a tensor field and supports the additional visualization of a connected scalar field.
DE: 0520 Data analysis: algorithms and implementation
DE: 0530 Data presentation and visualization
SC: Earth and Space Science Informatics [IN]
MN: Fall Meeting 2005