
180 Variational Methods in Image Processing
This is defined as the minimizer of the “interpolating” energ y
E(v) =
ˆ
Ω
h
µ(|∇v
1
|
2
+ |∇v
2
|
2
) + |∇f|
2
|v − ∇f|
2
i
dx, (9.6)
where f(x) is the negative of one of the external forces use d in the snake model:
f(x) is |∇I(x)|
2
, |∇(G
σ
∗ I)(x)|
2
, −I(x), or −(G
σ
∗ I)(x). In particular, the
vectors ∇f point toward the edges and are nor mal to the edges at the edges.
When |∇f| is large, v will be close to ∇f due to the second term, while in
regions where |∇f | ≈ 0, v will be slowly varying due to the H
1
diffusion
regular ization o f the first term.
Vector v is obtained by time-dependent gradient descent of the Euler –
Lagra ng ...