Dr. Kupervasser Oleg 8-916-4516193 (10.00-22.00) 8-499-134-3965 (20.00-22.00)

erminerebelAI and Robotics

Nov 15, 2013 (3 years and 8 months ago)

121 views

Dr. Kupervasser Oleg

8
-
916
-
4516193 (10.00
-
22.00)

8
-
499
-
134
-
3965 (20.00
-
22.00)




OLEG KUPERVASSER

e
-
mail: olegkup@yahoo.com

address: Russia, Vavilova 54
-
1
-
71 Moscow
119296

Date of birth: 01/02/1966

Citizenship: Russia, Israel

Home page:
http://leah.haifa.ac.il/~skogan/Apache/mydata1/
Oleg_home/Oleg_home.html

Education


2004 Haifa University: Three dimensional protein structure. Postdoctorate.

(
http://leah.haifa.ac.il/~skogan/Apache/mydata1/main.html
) ; Perl, DHTML,
Fortran, C++ languages


2001
-
2002 ATLAS college: Bioinformatics for Hi
-
Tech people. Final project:
Three dimensional protein structure, under supervision of Professor Edward
N. Trifonov from Weizmann Institute of Science.

Including: Basic knowledge in Biology, Bioinformatics, Programming (C++,
SQL, JAVA, DHTML, Bio
-
PERL, UNIX, MATLAB)


Ph.D. 1992
-
1999 Weizmann Institute of Science, Faculty of Physics.
Research project: Nonlinear dynamics (Analytical methods and numerical
algorithms for the solution of nonlinear physics equations). Fortran, C++
languages


1991
-
1992 Tel
-
Aviv University, Faculty of Engineering, department of
Physical Electronics, Ph.D student, Tel Aviv, Free electron laser. Fortran
languages


M.Sc. 1983
-
1989 Moscow Institute of Radioengineering, Electronics and
Automation. (theoretical and experimental electro
-
optics, laser, fiber optics
and radars) Diploma honoris causa . Fortran language
.

Employment



2008
-
2010 Algorithm developer in
Moscow State University,
Russia, Moscow.

Algorithm
developer. C++ language

Computer drug design, nano
-
systems computer modeling; solvent influence on molecules interaction;
took part in scientific conferences; submitted six papers and one published in high impact factor
journal


2009
-
2009 Algorithm developer in
UltraSpect
, Image processing in Nuclear medicine. C++
language


2008
-
2008 Algorithm developer in
Vayar Vision
, Israel. Images search in Internet ("Google" for
images). C++ language

This company develops a
search engine

for images. For some picture this
search engine

looks for
similar pictures in the given data basis of images. This similarity is not based on objects
recognition. The similarity is based on not semantic characters (for example, a number of the
boundaries pixels over image segments and etc.).


2005
-
2008 Rafael
-
Technion: Image processing, multiple view geometry of smooth bodies,
Navigation systems; took part in scientific conferences; two published papers; Matlab language

It is creation of algorithms in the field of image processing, computer vision for navigation of rockets.
For Rafael (the leading Israeli company in the field of
rocket weapons
) programs was developed
for navigation of rockets by means of the video images and the known terrain map. This is inverse
problem with respect to the problem solved by means of Google Earth. In Google Earth for a given
trajectory and orientation the correspondent images can be found. In the developed method a
trajectory and orientation can be found on the basis of a video. From this experience expansion of
Google Earth can be developed, allowing video
-
navigation for a flying plane, a rocket or a car from
a video film.


2003
-
2004 Algorithm developer in
Intel

, Israel.
Image processing in digital TV, C++ language


2002
-
2002 Algorithm developer in
UltraSpect
, Israel. Image processing in Nuclear medicine, C++
languages


2000
-
2001 Algorithm developer in
Orbotech LTD
, Israel. Recognition of glass defects and plate
marks, Matlab, C++ languages.


1999
-
2000 Electronics engineer in
Tower Semiconductor LTD
, Israel. Design flash
-
memory, Excel


Student physics Olympiad in Moscow (1
prize)


Student physics Olympiad in USSA (2
prize)


About 27 publications in physics, image
processing


Continuum solution model.

Moscow State University

Process of solute

solvation

in solvent

Continuum solution
model.

Three components of Gibbs
energy




Gs

=

Gcav
+

Gnp
+

Gpol
,




Gcav



Hydrophobic component






Gnp

-

Van
-
der
-
Waals component






Gpol

-

polarization component

Methods for finding of polarization
component

Gpol

.


Exact numerical methods
:



Solution of Poisson equation in 3D


Solution of equivalent equation for charge on solvent
excluded surface (PCM)


Simplified PCM
:
COSMO


water

(
ε
=78)
is changed to metal

(
ε
=∞).


Exact analytical method for spherical
cavity

полости:



Multipole moments method
,


Mirror charges method


Heuristic model:



Generalized Born,


Surface Generalized Born.

About solute surface construction in
РСМ

Protein Loop
-
Lock Structure

Haifa University



We must decompose a protein on set of
closed loops. The developed Internet site
solves this task.

Protein Loop
-
Lock Structure




We must decompose a protein on set of
closed loops. The developed Internet site
solves this task.

Images Similarity Engine.

Vayar Vision

Similarity Engine

Similarity Engine :

Finding images similar to some given image in
some image
database
.

1)
First step: basic features definition

2)
Indexing of all images in database by help
these features

3)
Indexing of some given images by help these
features

4)
Finding image (or images) with maximum
feature similarity from database

large number of investigation exists in the field of
images similarity. But the current Engine:

1) Can generate almost infinity number of features
used for images similarity.

2) The system can learned from mistakes.

3) The system can be easily adaptive to some
new
definition

of similarity

4) The set of used feature has no semantic sense.

We don’t make recognition of objects on image
and use such no sense features as number of
pixels on object boundaries and etc.

Found similar images by help our
Engine from terragaleria site

Found similar images by help our
Engine from terragaleria site

Found similar images by help our
Engine from terragaleria site

Field of use

1)
Images similarity engine in Internet (like
Google for words)

2)
Finding relevant images in film for
prod
u
ct

adverti
s
ing

setting

3)
Finding relevant information about some
building or place from photo or film.

4)
Navigation by place recognition from photo or
film (GPS equivalent) by help Google Earth

5)
Find similar images for intellectual property
rights control


Vision Based Navigation from
Image Sequence and Digital
Terrain Model.

Technion

Vision Based Navigation from Image Sequence and Digital Terrain
Model.


It is creation of algorithms in the field of image processing,
computer vision for navigation of rockets. For Rafael (the
leading Israeli company in the field of
rocket weapons
)
programs was developed for navigation of rockets by
means of the video images and the known terrain map.
This is inverse problem with respect to the problem
solved by means of Google Earth. In Google Earth for a
given trajectory and orientation the correspondent
images can be found. In the developed method a
trajectory and orientation can be found on the basis of a
video. From this experience expansion of Google Earth
can be developed, allowing video
-
navigation for a flying
plane, a rocket or a car from a video film.


Vision Based Navigation from Image Sequence
and Digital Terrain Model


Compute
n

feature correspondences

Compute pose and ego
-
motion that

best explain the features movement

Pose and ego
-
motion

Two consecutive images

2n constraints (

u,

v)

12 variables

DTM

Recovering Epipolar Geometry
from Images of Smooth Surfaces


Technion

We present four methods for recovering the epipolar geometry from images of
smooth surfaces. Existing methods for recovering epipolar geometry use
corresponding feature points that cannot be found in such images. The first
method is based on finding corresponding characteristic points created by
illumination (ICPM
-

illumination characteristic points method). The second
method is based on correspondent tangency points created by tangents
from epipoles to outline of smooth bodies (OTPM
-

outline tangent points
method). These two methods are exact and give correct results for real
images, because positions of the corresponding illumination characteristic
points and corresponding outline are known with small errors. But the
second method is limited either to special type of scenes or to restricted
camera motion. We also consider two else methods, termed CCPM (curve
characteristic points method) and CTPM (curve tangent points method), for
search epipolar geometry for images of smooth bodies based on a set of
level curves with a constant illumination intensity. The CCPM method is
based on search correspondent points on isophoto curves with the help of
correlation of curvatures between these lines. The CTPM method is based
on property of the tangential to isophoto curve epipolarline to map into the
tangential to correspondent isophoto curves epipolar line. Unfortunately
these two methods give us only finite subset of solution, which usually
include "good" solution, but don't allow us to find this "good" solution among
this subset. Exception is the case of epipoles in infinity. The main reason for
such result is inexactness of constant brightness assumption for smooth
bodies. But outline and illumination characteristic points are not influenced
this inexactness. So the first pair of methods gives exact result.

Correspondent points

Epipolar geometry

Image processing in the Nuclear medicine.

UltraSpect

Diagram of parallel
-
hole collimator attached to a crystal of a
gamma camera. Obliquely incident gamma
-
rays are
absorbed by the septa.

Gamma Camera

Image Processing

Direct problem
:

To get Image on gamma camera from image of
radiated particles distribution

Inverse problem
:

To get image of radiated particles distribution
from Image in on gamma camera


Slice of Brain


Deinterlacing for video.

Intel

Transform of
interlaced video frames to progressive video
frames.


Main problems:

Motion regions, no
-
smooth
boundary

Steps of Algorithm.

1)
Motion detection

2)
Interpolation or motion compensation

3) Detection of direction

4) Detection of angles

5) Detection of “zebra”

6) Detection of “high entropy regions”

7) taking into account history and environment

8) Fuzzy boundary between motion and no motion regions

9) Median filtering, Mean filtering over space and time.

Detection and recognition of defects on
the glass


Orbotech LTD

Types of defects:
bubble
s,
scratch
s,
dirty

spot
s

Steps of Algorithm

1)
Segmentation

2)
Descriptors (features) definition

3) Finding P(X
j
|H
k
) , X
j
-
Descriptor (N), H
k
-

type of
defect (M)

4)
Naive bayes

model:

X={X
1
,…,X
N
}

P(X|H
k
) =
П
j

P(X
j
|H
k
)


P (H
k
|X) = P(H
k
)∙P(X|H
k
) / P(X)

P(X)=
Σ
k
P(H
k
)∙ P(X|H
k
)

Flash memory design

Tower Semiconductor LTD

SONOS Transistor

Flash memory design

Tower Semiconductor LTD


S
ilicon

oxide
-
nitride

oxide

silicon memory transistor, where information is
stored as
two

charges in nitride at the edges of the channel.


Silicon
-
Oxide
-
Nitride
-
Oxide
-
Silicon memmory (SONOS)


A SONOS memory cell is formed from a standard polysilicon NMOS
transistor with the addition of a small sliver of silicon nitride inserted inside
the transistor's gate oxide. The sliver of nitride is non
-
conductive but
contains a large number of charge trapping sites able to hold an
electrostatic charge. The nitride layer is electrically isolated from the
surrounding transistor, although charges stored on the nitride directly affect
the conductivity of the underlying transistor channel. The oxide/nitride
sandwich typically consists of a 2 nm thick oxide lower layer, a 5 nm thick
silicon nitride middle layer, and a 5

10 nm oxide upper layer.SONOS
promises lower programming voltages and higher program/erase cycle
endurance than polysilicon
-
based flash. SONOS distinguished from
mainstream flash by the use of silicon nitride (Si
3
N
4
) instead of polysilicon
for the charge storage material.