# Joint Workshop between Tohoku and Tsing Hua

AI and Robotics

Nov 6, 2013 (4 years and 10 months ago)

342 views

A
BSTRACT

[1]
Octagonal Drawings of Plane Graphs with Prescribed Face Areas

Prof.
Takao Nishizeki

Abstract:

An orthogonal drawing of a plane graph is called an octagonal drawing if each inner face is drawn
as a rectilinear polygon of at most eight corn
ers and the contour of the outer face is drawn as a
rectangle.

A slicing graph is obtained from a rectangle by repeatedly slicing it vertically and horizontally.

A slicing graph is called a good slicing graph if either the upper subrectangle or the lower

one
obtained by any horizontal slice will never be vertically sliced.

In this talk we show that every good slicing graph has an octagonal drawing with prescribed face
areas, in which the area of each inner face is equal to a prescribed value.

Such a dra
wing has practical applications in VLSI floorplanning. We also give a linear
-
time
algorithm to find such a drawing.

[2]
On Unfolding Trees of Small Diameters

Prof.
Sheung
-
Hung Poon

Abstract:

We consider the problems of unfolding trees of small diameter
s in 2D and 3D. Such unfolding
problems have applications in motion planning

for robot arms, molecular conformation in
computational biology, and wire bending process. During the unfolding process, the length of all
graph edges are preserved and no edge cr
ossings are allowed. Let n be the number of edges of the
given graph. We present locked trees of diameter 4 in both 2D and 3D whereas trees of diameter 3
in 2D and 3D can always be easily straightened. We also present a linear locked tree of only eight
edg
es, which has been proved to be minimal. Furthermore, we show that an equilateral tree of
diameter 4 can always be straightened. Finally we propose some open problems. First, it is open
whether an equilateral tree of diameter 4 in 3D can always be straight
ened. Second, is it true that a
forest of diameter
-
3 trees in 2D or 3D can always be straightened

?

[
3
]

Introduction to Probabilistic Image Processing and Bayesian Networks

Prof.
Kazuyuki Tanaka

Abstract:

Bayesian network is one of the methods for probabilistic inferences in artificial intelligence.

Some probabilistic models for image processing are also regarded as Bayesian networks.

In this talk, the statistical aspect and the practical
schemes of Bayesian network to probabilistic
image processing are reviewed.

The first part is an introduction of probabilistic model for image processing based on the basic
framework of Bayesian networks.

The second part is a brief review of belief propaga
tion.

In the third part, we survey fundamental algorithms of belief

propagations for probabilistic image
processing.

Some recent developments of Bayesian networks and belief propagations in computer sciences are
also shown.

[4]

Computer Vision for Digita
l Photography

Prof.
Hwann
-
Tzong Chen

Abstract:

Recent progress in computer vision provides many effective techniques that are handy for digital
photography. Some of the techniques have been used to implement add
-
on functions now available
in commercial

software and consumer electronics, e.g., panoramic mosaicing, face detection, and
facial expression recognition.

In this talk, we will discuss two other vision techniques that may also be employed in digital
photography: The first one is tone reproduction

for producing and displaying high
-
dynamic
-
range
images, and the second one is feature description and visual saliency for photo composition. We
will see how the related algorithms facilitate the process of taking pictures and how they help to
improve the
quality of digital photos.

[
5
]

High
-
Accuracy Machine Vision Technology Using Phase
-
Only Correlation
-
From 3D Vision
to Biometrics

Prof.
Takafumi

Aoki

Abstract:

This talk introduces basic concepts of Phase
-
Only Correlation (POC)

technique
--

a high
-
acc
uracy
image matching technique using phase

information of discrete Fourier transform.

Since 1990s, our
research

group has developed a novel technique of phase
-
based image matching

for biometric
authentication.

Recently, we have proposed an efficient

imag
e correspondence algorithm using POC, which can find pairs of

corresponding points between
the given two images with sub
-
pixel

accuracy.

Application examples include smart image sensors,

super
-
resolution video processing, microscope image analysis,

3D mac
hine vision systems,
automotive vision systems, human interface,

and multi
-
modal biometrics.

[6]

Robust stereo vision system

Prof.
Shang
-
Hong Lai

Abstract:

In this talk, I will present our recent development of a robust and efficient

stereo vision sys
tem and
its related applications. This system mainly consists

of camera calibration, image rectification, and
stereo matching. I will focus

on automatic algorithms for robust image rectification and efficient
stereo

matching in this presentation. The propo
sed image rectification algorithm can

rectify images
based on automatic correspondence of SIFT feature points, which

ontain a considerable amount of
mis
-
matches. In addition, we also impose the

temporal consistency constraint to the rectification
transform
ations to make

the image rectification results very stable. For the stereo matching, I will

present our recent development of an efficient stereo matching algorithm based

on combining the
image over
-
segmentation with the construction of a regular

graphical

model. The stereo matching is
thus accomplished by an efficient

belief propagation procedure. I will show some experimental
results on the

reconstructed depth maps by using the proposed stereo vision system. Finally, I

will
also show some related applicat
ion in 3D TV based on this stereo

reconstruction system.

[
7
]

Intelligent integrated systems for real
-
world applications

Prof.
Michitaka Kameyama

Abstract:

Intelligent integrated systems which execute autonomously in real
-
world environment are
develop
ed for human
-
oriented applications such as intelligent robot systems to support our living
styles, highly safe intelligent systems to make secure our daily life, and information appliance
systems to support humans.

Such intelligent systems for real
-
world a
pplications need to achieve very quick response for
dynamically changing environment as well as low power
-
operations. Therefore, its emerging
tec
hnologies rely increasingly on low
-
power VLSI systems having high degree of parallelism.

discussed.

[
8
]

Wireless Sensing and Its Applications to Rehabilitati
on and Environment Monitoring

Prof.
Chung
-
Ta King

Abstract:

In this talk, we will discuss our recent progresses on developing wireless sensor platforms and the
applications of such devices to debris flow monitoring and rehabilitation behavior tracking. In
debris flow monitoring, we develop wireless se
nsors that can work in outdoor environment, operate
for a long period of time, and transmit data reliably. The sensor platform can work with rain gauge,
wire sensor, geophone, and many other sensors that are useful for debris flow monitoring. In
rehabilita
tion behavior tracking, we develop tracking systems that can monitor the gestures of
patients for correctly performing the required movements. We will discuss the challenges in
deploying wireless sensors to these applications and the research issues that n
We will also discuss some initial results.

[9]

Optimization for Power Gating Designs

Prof.
Shih
-
Chieh Chang

Abstract:

Power Gating is effective for reducing leakage power.Previously, a

Distributed Sleep Transistor
Network

(DSTN) wa
s proposed to reduce the sleep transistor areaby connecting all

the virtual
ground lines together to

minimize the Maximum Instantaneous Current (MIC) through sleep

transistors.

In this presentation, we propose

methodologies for determining the size of sle
ep transistors of the
DSTN

structure considering

charge
-
balancing effect. We also introduce a new relationship among
MIC,

IR drops and sleep transistor

networks from a temporal viewpoint and improve the sizing
results. Our

methods achieve significant bette
r

results than previous works.