3 Eyes contour detection/tracking

connectionviewAI and Robotics

Nov 17, 2013 (3 years and 9 months ago)

84 views

Projects

1
S
imulate pictures

from a photograph

in a

s
tyle of your
favorite painter
.


In this project the goal is
to convert a regular photograph to a painting as it would be
painted by a specified artist, for example, Van Gogh. The basic approach is based
on
replacing each fragment in the photograph with a fragment from a collection of paintings
of that artist. The problem is formulated as global optimization of a function that
integrates local similarity and spatial connections.


Suggested method: Crea
te a

database of paintings, typical to the painter.

For each patch
in the photograph replace it with a most

similar patch in the database.

As

the first step do
it in gray
-
scale only. As a second step try incorporating colors. You should suggest

a
method for s
titching patches

while preserving the style features.


If the patch sizes are too
small the photograph will still looks very much the same.

If the patch size is too big, the
image will be unrecognizable.

You
are welcome to consider other methods.



2
Even
t classification using state
-
of
-
the
-
art approach

Event classification is one of the central problems in computer vision. The task is to
automatically classify a video sequence (or its fragment) to a predefined events, for
instance
Kicking

(a ball),

Lift
ing
, Horseback riding, Running, etc..




In this project you should i
mplement event
classification

using
SVM classifier applied
on

Bag of Words
representation
constructed from HOG3D features using dense sampling
in space and time
. The method is described

in


"Eval uat i on of l ocal spat i o
-
t empor al f eat ur es f or act i on r ecogni t i on"

( 2009),

H. Wang, M. M. Ul l ah, A. Kl äser, I. Lapt ev and C. Schmi d;
i n Pr oc. BMVC'09
, London,
UK.


You should use
the code for feature extraction and descriptors provided in
http://lear.inrialpes.fr/people/klaeser/software_3d_video_descriptor

.


The data to be used is
USF50
provided i
n
http://www.cs.ucf.edu/vision/public_html/



3
Eyes contour detection/tracking

You can use a method of your choice. The system should be tested on a large number of
images and should
be
efficient.

4

Mouth contour detection/tracking


You can use a method of your choice. The system should be tested on a large number of
images and should

be

efficient.


5
. Tracking of a
Dorsophilia

(fly) movement

(in
collaboration with Shmulik Raz
)


The

project

is
on
st
udding
movement asymmetries of biological specimens

(flies). This is part of a multi
-
disciplinary widespread evolutionary at

Nahal Oren. The
project will require pattern detection, object

extraction and movement detection analysis
and modeling. It require
s

algorithm development, MATLAB coding, and running tests on
biological
data

(to be provided)
.


6. Secure face recognition


Nowadays, there is an emerging interest in the application of biometric authentication
and identification. Faces are obviously the

most convenient biometric templates.

Biometric eliminates the problem of forgotten or poorly chosen passwords. Biometrics,
however, pose another problem: if stolen it cannot be replaced


thus it should be
protected. In many operating systems, a given p
assword P is not stored explicitly.
Instead, a commitment of P is stored in the form of indivertible function of P. Thus it is
possible to verify that a user has entered the password correctly, without storing the
password itself. Existing techniques for
password protection are not applicable to
biometrics, because two readings of the same biometric are not identical.


This project
focuses on designing effective representations for face recognition th
at can
be used in cryptography. The project consists of

implementing and integrating several
methods and testing the recognition cryptographic strength of the representation.


The relevant papers and data set will be provided.



7. Learning to predict
the

highlights.


This project is

a

part of a larger proje
ct that involves pose estimation and recognition of
shiny (specular
) objects
.
This is a very challenging task, since the appearance of the
highlights they produce changes drastically with the viewing conditions. This problem is
also important, since most o
f the objects around us are made from materials that produce
highlights, like ceramic, plastic, wood, glass etc. Recognition systems have generally
treated specular highlights as noise.
In this project the
highlights
are used
as a positive
source of inform
ation

for recognition
. We have developed a method that could recognize
specular object with known pose and unknown illumination
, but it needs a better method
for extraction of highlights. This involves two tasks: 1) The detection
of the highlight
(there i
s a

previous work on the topic), 2) Our method assumes a simple model for
highlight formation. The rendered highlights
differ from the true ones in some r
espect.
The goal of the project is to study the
discrepancy

and
develop a method that
automatically re
duces it.


The relevant papers and data set will be provided.