
226 Variational Methods in Image Processing
maximum likeliho od estimation is commonly use d:
Φ(f − k ∗u) =
ˆ
Ω
X
i
|f
i
− k ∗ u
i
|
2
dx.
However, the qua dratic data fidelity ter m considers the impulse noise, which
might be caused by bit error s in transmis sions or wrong pixels, as a n outlier.
So, for the impulse noise model (or the additive Laplace noise mode l), the L
1
-
fidelity term is more appropriate, due to its ability to remove outlier effects
[12], [251]; more over, we consider the ca se of independent channel noise [38]:
Φ(f − k ∗ u) =
ˆ
Ω
X
i
|f
i
− k ∗u
i
|dx.
Thus we present two types of total energies for color image deblurring-
denoising (to be minimized ...