Sie sind auf Seite 1von 17

Assignment 5

NumericalGeometry of images




Avithal Elias
Std id:950459073
avithal@gmail.com





Problem1
Mean curvature flow to obtain minimal surface:
2 2
3/ 2
(1 ) 2 (1 )
uu v u v uv vv u
t
z z z z z z z
z H g g
g
+ + +
= =

Also
Explicit scheme (m-dimensional):
1
1
( )
m
k k k
l
l
u I A u u X
+
=
+
= +

' '


Where ( )
l
A u is a differential operator defined as follows:
2
( )
2
( )
0
i j
ij ik
k i
g g
j N i
h
a u a i j
else
{
+ |

= =
,


( ) g s is any diffusivity function, that meet the required conditions:
: [0, ) [0,1], (0) 1 lim ( ) 0
s
g with g and g s
pg
g p = =

Curve 1:


The explicit & mean curvature form :







Curve2:










Curve3:











Observations:
1) It is found from the above diagrams that mean curvature evolution is more
efficient & stable than explicit method;
2) It is also found that the Mean curvature flow becomes unstable wen the
derivative height 'dx' is decreased. For the above cases the value was one.
3) The 'dx' is decreased the curve which is more affected adversely making it
unstable if curve 2 as it has lot of variation.
4) As 'dx' is increased the value the time required to evolve to minimum surface is
increased.
5) It was seen as 'dx' is increased the explicit scheme is more stabilized .
Conclusions.
1) For curves with high variation in values; dx ie derivation step should not be very
small for mean curvature flow.
2) For explicit scheme dx ie derivation step can be increased for stability for all
curves .




PROBLEM2
Using Majorization we minimized the function
4 2
1 1
( )
4 2
f x x x = :
1. Guess a Majorization function ( , ) g x y , which satisfies the following:
a. ( , ) g x y is convex (in x ).
b. ( ) ( , ) f x g x y e
c. ( ) ( , ) f y g y y =

4 2
1 1
( , )
4 2
g x y x xy y = +
2. Minimize ( , ) g x y according to x :

3
1
3
min
( , )
0
( ) ( )
g x y
x y
x
x k y k
x
= =
x
=

3. For initialization,y(0)=-4 the minimum f(x)=56
The Majorization process is displayed in the figure below:

Section2
The stress function is given by:
2 2
( ) ( ) 2 ( )
ij ij ij ij ij ij ij
i j i j i j
S X d X d X [ H [ [ H
" " "
= +

Where

1/ 2
( )
T
ij i j i j
d X x x x x

=

|


,

1/ 2
1/ 2 1/ 2
( )
( ) ( )
T
ij ij ij ij ij i j i j
i j i j
T T T
ij ij ij ij
i j i j i j i j i j i j
Cauchy Shwartz i j i j ij ij
d X x x x x
x x x x y y y y x x y y
d Y d Y
[ H [ H
[ H [ H
" "
" "

=

|

= >

| |



So we can define:

1/ 2
2 2
( , ) ( ) 2
( )
T
ij ij
ij ij ij ij i j i j
i j i j i j ij
G X Y d X x x y y
d Y
[ H
[ H [
" " "

= +

|

And we just showed that ( , ) ( ) G X Y S X > . Thus, ( ) S X can be majorized by ( , ) G X Y
(clearly the other terms for Majorization function are fulfilled).
3 Section
Majorization of the Stress function:
2
( ) ( ( ) )
ij ij ij
i j
s X w d X H
"
=

yields the multidimensional scaling.


Orginal







Classical MDS:


2
1
2
B JD J = where
1
11
T
J I
N
= and D is the distance map.
Using spectral decomposition
1
2
T
N m N m m m
B V V
X V
v v v
= A
= A

And the canonical form of the bunny (using classical MDS):

SMACOF:
Iteratively update X :

( 1) ( ) ( )
1
2
k k k
X X s X
N
+
= \
Where -
1
2 2 ( )
1
( ) , ( ) 0
( ) 0 , ( ) 0
N N N N
ij ij ij ij
ij ij
ik
k i
s VX B X X
V NI
w d X if i j d X
b X if i j d X
b if i j
H
v v

{
\ =
=
| { {

= { =
,

=
|

















Problem3#
Pepper Image with initial segmentation contour:



Geodesic active Contour
The functional to be minimized is ( ) g C s ds

where s is the Euclidean arclength, and g


is some edge detection function.
The steepest decent flow is:

,
t
C g g N N O \
& &

With the level set formulation:
t
div g
J
J J
J
+
\
= \


\
' '

Adding constant velocity term ( ) ( , ) v t g x y :

( )
t
div g I g I v t
J
J J
J
+ +
\
= \ + \ \

\
' ' ' '

2
( )
1
c
v t
t

=
+

where c is a constant.


Chan Vese segmentation
The PDE for an energy-based segmentation (operating on a level-set) is:
2 2
1 0 1 2 0 2
( ) ( ) ( ) div u c u c
t
I
J J
H J Q R P P
J
+
x \
= +


x \
' ' |

1 2
& c c are the statistical means inside & outside the segmentation contour (using
Heaviside smooth function).


Observation
1)GAC-
a) when dt is increased greater than or equal to one the system becomes inefficient ; this is
mainly because the value increases & it goes out of memory
b) When selecting the velocity term the
2
( )
1
c
v t
t

=
+
the constant term c when it was 11
made segementation faster .here value iof c is 11
2)Chan vese Segmentation
a)here when mean is increased time for segmentation is increased I have taken =.01
lamda1 = lamda2=0.5 ;dt=1
b) when dt is increased the segmentation is achieved fast but for the same iteration it also
includes the neighboring parts.

Das könnte Ihnen auch gefallen