
6 Variational Methods in Image Processing
FIGURE 1.2: Top: an original image u. Bottom left: its blurry version Ku
obtained using a convolution operator Ku = k ∗ u with k a Gauss ian kernel.
Bottom r ight: its blurry and noisy version f = Ku + n obtained from Ku by
adding Ga ussian noise of zero mean.
´
Ω
|Du|
2
dx is a regularization imposing a priori information on the unknown
minimizer u, while the second term
´
Ω
(Ku − f )
2
dx is called a data fidelity
term and takes into account the image formation model. If u ∈ H
1
(Ω) is a
minimizer of (1.1), then it must solve in the distributional sense the assoc iated
Euler–Lagrange equation,
λK
∗
(Ku − f) = △u in Ω (1.2) ...