86 Variational Methods in Image Processing
4.2.3 E ule r–La grange equations
We first consider the functional E defined by
E(u) = λ
ˆ
Ω
|f − Ku|
2
dx + |u|
NL/H
1
,
= λ
ˆ
Ω
|f − Ku|
2
dx +
ˆ
Ω
ˆ
Ω
(u(y) − u(x))
2
w(x, y) dy dx,
where Ku := k ∗u and λ > 0.
Let us a ssume that u is a minimizer of E and let us define J(ε) = E(u +
εh) for ε ∈ R and h, a test function. As u minimizes E, J
′
(0) = 0. By
differentiating J with respect to ε, we obta in
J
′
(ε) = − 2λ
ˆ
Ω
(f − Ku −εKh) Kh dx
+ 2
ˆ
Ω
ˆ
Ω
(u(y) + εh(y) − u(x) − εh(x))(h(y) − h(x))w(x, y) dy dx.
Then
J
′
(0) = − 2λ
ˆ
Ω
(f − Ku) Kh dx + 2
ˆ
Ω
ˆ
Ω
(u(y) − u(x))h(y)w(x, y) dy dx
− 2
ˆ
Ω
ˆ
Ω
(u(y) − u(x))h(x)w(x, y) dy dx.
Switching x and y in the seco nd integral and denoting by K
∗
the adjoint
operator of K (K
∗
u = k(−·) ∗ u) yields
J
′
(0) = − 2λ
ˆ
Ω
K
∗
(f −