Back to Taosong's Home Page
Ph.D. Computer Science, May'96
Visualization Lab,
State University of New York at
Stony Brook, New York
Thesis: "Volumetric
Virtual Environments"
Advisor: Dr. Arie
Kaufman
M.S. Computer Science, May'93
State University of New York at
Stony Brook, New York
B.S. Computer Science, May'91
FuDan University, Shanghai,
China

PostScript: thesis.ps.gz
Slide Show
We address one of the main applications of virtual reality,
three dimensional real time visualization and interaction,
primarily of volume data. We construct a virtual environment
supporting fast interaction, both volumetric and surface-based
data sets, and a variety of manipulations. A general-purpose
interaction model is proposed based on the concept of virtual
input device. Multiresolution hierarchy is applied to increase
the processing speed. Several fast volume rendering algorithms
are presented, together with the fast generation of stereo
images. Finally, virtual manipulations of data sets in the
virtual environment, such as volume morphing and hierarchical
collision detection, are presented.
Interaction Model:
- Device unified interface: We introduced the device
unified interface(DUI), which is a protocol for
communicating between applications and input devices,
based on the concept of virtual input device to
achieve input device independence. Virtual input
device converts all kinds of input device data into a
device independent representation. DUI can be used for
all kinds of 3D system, and has been implemented to
support 3D interaction of VolVis
with DataGlove, Spaceball, and mouse. New devices can
:also be easily incorporated. (Virtual
Input Devices for 3D Systems )
- General-purpose interaction model: We propose a
general-purpose 3D interaction model which can be used in
a 3D system to support real time interaction.
Specifically, this model is targeted for virtual reality
systems, where high bandwidth simultaneous input from
multiple input sources and users should be handled. In
our model, all the applications and interactions are
concurrent lightweight threads, which communicate and
synchronize fast by sharing global variables. Complex and
dynamic actions are also supported. Virtual input device
paradigm is employed.
Rendering:
- Object Simplification: We develop a practical and
robust voxel-based method for genus simplification of
objects. Our algorithms construct multiresolution object
hierarchy by sampling and low pass filtering the object
into volume buffer. It works for a wide variety of
objects and it supplements existing geometry-based object
simplification algorithms by gradually eliminating
higher-frequency features. It is also easy to implement
and is based on the robust theoretical foundation of
signal-processing theory. Controlled
Topology Simplification.
- Hardware-assisted Object Space Antialiasing: The
idea is to smooth out the transition between the boundary
of an object and empty space surrounding it by using
multiple layers of triangle meshes with increasing
translucency from the innermost layer to the outermost
one. Our object-space antialiasing approach involves
rendering of translucent polygons, which could be more
expensive than traditional image-space antialiasing on
some platforms. However, it has the advantage of higher
accuracy. Voxel-Based
Object Simplification.
- Compression-Domain Volume Rendering: We develop a compression
domain volume rendering algorithm based on the 3D
extension of JPEG algorithm and Fourier Slice projection
Theorem. We achieve high compression ratios while
maintaining high quality of attenuation and occlusion
effect.
- Wavelet-Based Volume Rendering: We prove that
there is no Wavelet
Slice Projection Theorem, we also develop algorithm
based on Wavelets to accelerate volume ray casting.
- Fast Animation: We optimize volume ray casting and
utilize the data coherence between the two consecutive
scene, both in object space and image space.
- Z-buffer assisted Visibility Determination: The
main idea is to determine the visible scenes by checking
the status of z-buffer,
when the eye-point is inside a complex and close model.
Comparing to most of the current visibility algorithms,
the new algorithm avoids the expensive preprocessing, and
fully takes advantage of the geometry properties of the
model.
Stereo Volume Rendering
- Stereo Rendering: We develop algorithm to
accelerate the volume rendering of stereoscopic pairs.
Image coherence between left-eye and right-eye images are
utilized to accelerate the generation of the second image
of a pair. Fast
Stereo Volume Rendering
Virtual Manipulation
- Volume Morphing: We develop wavelet-based
volume morphing to automatically generate a smooth
transition between the two datasets and produces morphed
volume with fewer high frequency distortions than those
obtained from spatial domain volume morphing.
- Collision Detection: We develop hierarchical
collision detection algorithm for volumetric objects.
Inter-penetration between non-solid objects is detected,
and all kinds of feedback including haptic feedback can
be accordingly provided.

Back to Taosong's Home Page