
200 Variational Methods in Image Processing
using the classical finite difference operators D
+x
Φ
i
= Φ
i+1
−Φ
i
and D
−x
Φ
i
=
Φ
i
− Φ
i−1
in one dimension.
As for the topolo gical constraint, we have used the C
∞
regular ization of
the Heaviside function defined by
H
ǫ
(z) =
1
2
(1 +
2
π
arctan(
z
ǫ
)).
The partial derivatives are approximated using central finite difference
schemes. While the topologic al constraint introduced in the modeling enables
us to use the level set framework to track interfaces whose topology must
be preser ved, the main drawback is computational cost due to the nonlocal
term appearing in the evolution e quation. This limitatio n can be overcome
by slightly ...