
42 Variational Methods in Image Processing
Then
A
′
(t) = −
1
2
ˆ
b
a
h
∂C
∂t
,
~
Nidq −
1
2
ˆ
b
a
hC(q, t),
−
∂
2
C
2
∂t∂q
,
∂
2
C
1
∂t∂q
idq,
= −
1
2
ˆ
b
a
h
∂C
∂t
,
−
∂C
2
∂q
,
∂C
1
∂q
idq
+
1
2
ˆ
b
a
h
−
∂C
1
∂q
,
∂C
2
∂q
,
∂C
2
∂t
,
∂C
1
∂t
idq,
= −
ˆ
b
a
h
∂C
∂t
,
−
∂C
2
∂q
,
∂C
1
∂q
idq,
= −
ˆ
b
a
h
∂C
∂t
,
~
Nidq,
= −
ˆ
b
a
h
∂C
∂t
,
~
N
|
~
N|
i|
~
N|dq.
The last expression decreases most rapidly if C solves
∂C
∂t
=
~
N,
where in the la st line
~
N is the unit normal. We conclude that the area mini-
mizing flow is the motion in the normal direction by a constant.
2.9 Variational level set methods
The level set approach introduced and developed by Osher and Sethian
[254], [285], [286], [256] for curve and surface evolution provides a powerful
and elegant