|
We are primarily interested in data visualization, computer vision and computer
graphics. We are also interested in application of high-performance and parallel
computing to problems in visualizations, computer vision and graphics.
Scientific
Visualization Computer
Graphics
In Data visualization, our interest
is feature extraction, feature presentation and data rendering, especially
of volumetric or other multidimensional data.
We are also interested in parallel feature extraction and data rendering using
high performance computing. Our work in this area has been supported by
an NSF-funded
Early Faculty Career Development (CAREER) Award. Previously,
Dr. Newman also developed visualization tools for several medical applications
under a post doctoral fellowship from the National Academy of Sciences at the
National Institutes of Health and subsequently with funding from Cray Research. We
also have interests in registration of volumetric datasets.
In this research work, our
focus is mainly on determining a volumetric density distribution of the plasma
in the Earth’s Plasmasphere. Plasmasphere
is a region in the magnetosphere. Magnetosphere is a region around the Earth
that is formed by the flow of plasma from the Sun and by the Earth’s
magnetic field and the magnetosphere has several regions of varying plasma
densities. One such region is called the Plasmasphere. Space physicists and
scientists have always been interested to know the spatial distribution of
the plasma in the Plasmasphere in order to know more about the geomagnetic
activities that occur in space. Hence this research work is an attempt to solve
this space science problem.
IMAGE is a satellite that was launched by NASA in order to externally
investigate the magnetosphere and its regions. The IMAGE has several
instruments on board but our focus is on the Extreme Ultraviolet (EUV)
imager that is used to take global snapshots of the plasmasphere. The
EUV instrument uses a photon imaging technique to image emissions from
the Helium ions in the plasmasphere. The EUV instrument has 840x900 FOV
and takes one image of the plasmasphere in every 10 minutes. Here is
picture of the plasmasphere taken by the EUV instrument.
The focus of this research work in our lab can be classified into two
main categories. One is determining the extent of the plasmapause boundary
in the equatorial plane and the other is determining a volumetric plasma
density distribution in the plasmasphere.
This webpage will soon host interactive softwares for identifying the
plasmapause boundary and for determining a volumetric density distribution
of the plasma from a series of EUV images.
A suite of tools enable discovery of terrestrial plasmasphere characteristics from NASA IMAGE Extreme Ultra Violet (EUV) images.
Features are supported by the tool suite include:
1. Magnetic Equatorial Plane Plasmapause Extraction
Semi-automatically select of the plasmapause boundary in an EUV image, and map of the selected boundary to the geomagnetic equatorial plane. The plasmapause mapping feature is achieved via the Roelof and Skinner (2000) Edge Algorithm or Miminum L Algorithm.
2. Plasmasphere Reconstruction via Tomographic Imaging Technique
Reconstruct plasmasphere plasma density distribution from a short sequence of EUV images. The plasma density reconstruction is achieved through tomographic technique that exploits physical constraints.
3. Plasmasphere Reconstruction via Latitude Variation Model Parameters Recovery
The reconstruction recovers plasmasphere plasma densities distribution through a robust fitting process that determines parameters of the Latitude Variation Model (LVM), proposed by Huang et. al. (2004). LVM is based on observation in IMAGE RPI data and requires determination of 5 parameters for each magnetic meridian plane.
access the Web-based Tool Suite at:
http://plasma.cs.uah.edu:8080/plasmasphere/page0.jsp
We have used volume growing, morphological operations, and locally
adaptive histogramming to extract organs and other structures of interest
from lower torso CT data. In our work, we addressed the renal complications
of von Hippel Lindau (VHL) disease. Von Hippel Lindau patients tend
to develop cysts and tumors in and on their kidneys. The cancers are
usually attacked surgically through removal of the diseased growths
in and on the kidney. Due to stress on the renal system, the surgical
intervention must be concluded subject to time constraints. One of
the goals of our work was the accurate and timely detection and presentation
of tumorous tissue. Through our three-dimensional visualizations, diagnosis
and surgical planning were aided.
Recently, we have begun to investigate classification of anatomical
structures in brain CT and MR images. Our work is designed to develop
strategies and techniques useful in frameless image-guided neurosurgery.
Currently, it is difficult to use volumetric data, such as CT or MR,
to guide surgery. One of the conventional approaches to neurosurgery
involves the use of unwieldy stereotactic frames to allow registration
of preoperative image data with intra-operative image data and surgical
coordinates. Although less invasive approaches have been used at some
medical centers, we aim to aid in even less invasive approaches that
exploit images in planning and guiding surgery. By classifying structures
of interest in 3D datasets, the task of registering images of different
modalities can be aided. Furthermore, the extraction and classification
of structures of interest are a critical component for visualizations
and renderings of the dataset for surgical planning and guidance.
We have developed methods for the extraction of the eyes, brain lateral
ventricles, and brain longitudinal fissure. The ultimate goal of the
feature extraction is to allow the registration of different modalities
of MR datasets of the same individual (for example, registration of a
T1-weighted dataset with MRA, T2-weighted, and proton density datasets).
We have developed vector-parallel techniques for medical visualization.
Only a small number of research efforts to date have explored the possibility
of
using vector-parallel computation for visualization. We have developed a vector-parallel
realization of the Marching Cubes algorithm that we have used for isosurface
extraction in tomographic images. Extraction of isosurfaces is one of the tools
a visualization system that we developed for lower torso diagnosis and intervention
planning. We use Marching Cubes to generate a display of structures of interest--particularly
the spine, ribs, and several organs. The spine and ribs help orient the surgeon
as he or she looks at the three-dimensional renderings to determine the positions
of kidney cysts and tumors.
We have also developed vector-parallel realizations of the popular surface
and volume ray-tracing techniques.
See the technical
report
In computer vision, our past work has been in several areas, including the
development of efficient techniques for model-based recognition and inspection
of 3D objects. We have focused on processing using range or other 3D data.
Application areas include medical, manufacturing, military and earth and space
data.
We have developed methods for fitting quadric surfaces in range images,
for detecting military vehicles in laser radar images and for inspecting
indutrial castings in range images. We have also developed a parallel
technique for 3D object recognition in range images.
We have developed a suite of methods for model-driven identification
of quadric surfaces in segmented range imagery. The model-driven parameter
estimation is advantageous because it reduces the number and/or range
of the parameters to be estimated. The methods we developed seem reasonably
immune to the effects of noise and produce good classifications of the
images tested. We are currently investigating parallel algorithms to
speed up the classification techniques.
Automatic identification and localization of 3D objects in images is a major
research issue in computer vision. This problem has been tackled using many
approaches, most of which use some form of search technique for matching objects
to models. While many parallel algorithms for low-level vi sual tasks have
been presented, relatively few parallel algorithms for high-level visual tasks
are in the literature. One of the popular high-level vision paradigms is the
interpretation tree (IT) search for matching observed scene features to known
model features. We have developed and analyzed parallel implementations of
an interpretation tree (IT) search algorithm for recognition of three-dimensional
objects in range images.
Currently, many manufactured parts and assemblies are designed using computer-aided
design (CAD) tools. As manufacturing enterprises become integrated, CAD data
will be used in downstream functions such as inspection. We have developed
a system that utilizes the CAD model for inspection of iron castings at several
points during the assembly sequence. The techniques that were developed were
aimed at detecting defects that commonly occur in the casting process. Inspection
of castings is a challenging problem because the unmachined castings contain
natural roughness from the sand molds into which the metal was poured during
casting and because of scratches from collisions with other castings during
manufacture. Some of the most common surface defects in castings are pits,
excess material, and gross insufficient material. The system that was developed
inspected for these defects in castings composed of planar and cylindrical
surfaces (which included most castings produced at the plant). It also inspected
for the presence of several design features and for their adherence to dimensional
tolerances specified in the CAD models of the castings. The casting inspection
techniques used template-matching and model-based surface classification methods.
The techniques exhibited robust behavior and correctly classified approximately
95% of the castings inspected.
See the technical
report.
Recognition of Structures (esp. Vehicles) in SAR and LADAR Images
One of the important measures used in assessing heart function is the
ejection fraction (EF). EF is the ratio of the volume of blood ejected
from the Left Ventricle (LV) during contraction (systole) to the volume
of blood in the LV at the end of dilation (diastole). Other useful
measures involve the characterization of the size and shape of the
LV and LV wall motion and thickening. Clinically, it is particularly
useful to be able to accurately estimate size and shape characteristics
throughout the systo lic cycle. We have recently developed a 3 step
automatic extraction for the Left Ventricle, on a sequence of gated
blood pool (GBP) single photon emission computed topography (SPECT)
images. The first step is the edge detection performed using a combined
Isotropic a nd Marr Hildreth edge detection. The second one is to extract
ellipse shaped hypothesis in each longitudinal slice. The last step
is to group the ellipse hypotheses. Each ellipse is a potential cross-section
of a 3D ellipsoid. The goal is to group relate d ellipse parameter
vectors into classes which represent the ellipsoids in the data. We
use the agglomerative hierarchical clustering to collect the ellipse
vectors into ellipsoids. The two first steps are automatic, and do
not require any human interaction. The last one has highlighted the
need for the designer to control, interact, and understand how the
single-link clustering properties are mapped on the set of ellipses.
We have d eveloped a visualization tool that uses a one-to-one equivalence
between a graph or tree and a single link cluster generation processing.
It provides the operator with a large scope of commands to understand
and to arrange the most efficient segmentation he can get from a single
clustering based algorithm.
Check out the sample
images about part of our cardiac functional visualization
work!
In graphics, we are currently investigating parallel image synthesis. We are
also interested in terrain visualization, atmospheric effect generation, natural
feature synthesis and surface synthesis.
This project involves synthesizing models of damaged objects through interpolation
of input models.
We are currently beginning an investigation involving
animation of the motion of the heart over time.
See under visualization.
We are investigating parallel data mining with Professors Graves and
Hinke of the Information Technology and Systems Lab. We are currently
parallelizing the miner using coarse granularity parallelism on networks
of workstations and finer granularity parallism on an IBM POWER parallel
machine.
|
