
140 Variational Methods in Image Processing
with
c
0
(χ
1
) =
´
Ω
g(x)(1 − χ
1
)dx
´
Ω
(1 − χ
1
)dx
,
c
j
(χ
j
, χ
j+1
) =
´
Ω
g(x)χ
j
(1 − χ
j+1
)dx
´
Ω
χ
j
(1 − χ
j+1
)dx
, 1 ≤ j ≤ m − 1 ,
c
m
(χ
m
) =
´
Ω
g(x)χ
m
dx
´
Ω
χ
m
dx
.
Theorem 32 The minimization problem (8.10), with E defined in (8.11),
has a minimizer (χ
1
, ..., χ
m
) ∈ BV (Ω)
m
, with χ
j
(x) ∈ {0, 1} dx-a.e. for 1 ≤
j ≤ m, and χ
j
(x) ≥ χ
j+1
(x) dx-a.e. in Ω.
Proof. The energy E from (8.11) satisfies E ≥ 0. Also, it is easy to find
characteristic functions χ
j
= χ
E
j
, 1 ≤ j ≤ m, with E
j+1
⊂ E
j
, E
j
with finite
perimeter in Ω or finite total variation in Ω, such that E(χ
1
, χ
2
, ..., χ
m
) <
∞. These two conditions on E will guarantee that the infimum is finite