Spostovani,
V okviru SDRV seminarja bosta v sredo 17. maja, ob 13:00 oz. ob 14:15, v
diplomski sobi na Fakulteti za elektrotehniko dve vabljeni predavanji:
13:00
****************************************************************************
*
Real-Time Medical Image Analysis: From Dreams to (Virtual) Reality
Milan Sonka, The University of Iowa, USA
Cardiovascular disease is the primary cause of death in the western world. A
variety of medical imaging methods was developed to visualize cardiovascular
system - X-ray angiography, ultrasound, MR, CT. The methodology for
automated and real-time image analysis is quickly following understanding
the needs to help treat as well as prevent further development of
cardiovascular disease.
The talk will concentrate on three cardiovascular applications in which
imaging, quantification, and visualization in close-to-real time play an
important role: Geometrically correct fusion of intravascular ultrasound and
biplane angiography, analysis of left ventricular and left-atrial function
from intracardiac ultrasound pullback sequences, and non-invasive assessment
of cardiovascular system status via brachial artery flow-mediated dilatation
approach. Interactive demonstrations as well as validation results will be
presented.
14:15
****************************************************************************
*
Fractal-Based Methods of Signal/Image Analysis and Compression
Edward R. Vrscay, University of Waterloo, Canada
Fractal image coding, normally performed in the spatial (pixel) domain, may
be viewed as a kind of "self vector quantization" procedure in which
codebook vectors are chosen from the image itself. In fractal coding, an
image I is partitioned into a set of range blocks Ri. For each range block
Ri of I, there is an associated (larger) domain block Dj(i) of I along with
a typically affine greyscale mapping phii = alphai*t + betai which optimally
maps the image supported on Ri onto the image supported on Dj(i). The domain
block indices j(i) as well as the parameters alphai, betai comprise the
fractal code of image I under the partition Ri. These parameters define a
contractive fractal transform T whose fixed point I' is an approximation to
I. The problem is to find the "best" such operator T. (Mathematically, this
is the inverse problem of approximating elements of a metric space by fixed
points of contractive operators on that space.)
Fractal-wavelet coding seeks to perform such a "self VQ" procedure in the
wavelet domain. In a matter analogous to spatial fractal coding, domain
wavelet subtrees are mapped onto lower range wavelet subtrees. The strength
of the fractal-wavelet transform lies in its ability to combine desirable
features of fractal coding (scaling, local self-similarity) with those of
wavelet transforms (multiresolution analysis and discrete wavelet
transforms). For example, it is known that the geometric decay of wavelet
coefficients in a given subtree is determined by the local regularity of the
function. This decay is naturally measured by fractal coding.
In this talk, aimed at a general audience, we shall outline the development
of fractal coding from its early days, including original hopes, failures,
advantages and disadvantages. Finally, we outline the results of more
research: a simple hybrid fractal-wavelet coder that demonstrates extremely
competitive compression characteristics in terms of rate-distortion curves
(close to the best wavelet coders) with minimal computational cost.
Admittedly, fractal compression on its own will probably never be a viable
competitor to wavelet-based methods. Nevertheless, we continue to
investigate the interesting question of whether or not fractal-based methods
can be used to enhance wavelet compression methods.
****************************************************************************
*
Dodatne informacije in trenutni program seminarja dobite na:
http://biprog.fe.uni-lj.si/Sdrv/seminar.htm
Vljudno vabljeni,
Bostjan Likar