Image Processing Reference
In-Depth Information
any 0 ≤i,j ≤n, the following hold: [38]
(i−j)c u ≤ (iy j −jy i + i)...........(a)
(i−j)c l ≤ (iy j −jy i + i)...........(b)
(i−j)m u ≥ (y i −y j −i)...........(c)
(i−j)m l
≥ (y i
−y j
−i)...........(d).
Lemma 3.4. The following hold:
(A) c l < c < c u if and only if m
l (c) < m
u (c)), where
m
l (c) = max
i
(y i
−c)/i and m
u (c) = min
i
(y i + 1−c)/i
l (m) < c
u (m), where
(B) m l < m < m u if and only if c
l (m) = max
i
u (m) = min
i
c
(y i −mi) and c
(y i + 1−mi)
Proof: Part(A): Let c l < c < c u . Assume m
l (c) ≥ m
u (c). Clearly, there
exists p,q > 0 such that m
l (c) = (y p −c)/p and m
u (c) = (y q + 1 −c)/q. So,
m
l (c) ≥m
u (c) implies (y p
−c)/p ≥ (y q + 1−c)/q. Rearranging terms we gets
(p−q)c ≥ (py q
−qy p + p).
Thus, from Lemma 3.3, (p−q)c ≥ (p−q)c u and (p−q)c ≥ (p−q)c l . If
p > q then c ≥ c u and if p < q, then c ≤ c l . p = q is ruled out as it implies
that 0 ≥ p > 0.
Hence c l < c < c u implies m
l (c) < m
u (c).
Conversely, let m
l (c) < m
u (c). Select any m such that m
l (c) ≤ m <
m
u (c). From the definition of m
l (c) and m
u (c) we get,
∀i,m i + c−1 < y i
≤m i + c.
That is, y i = ⌊mi + c⌋. In other words, D(L(c,m)) = D o . Hence, from
Theorem 3.7, we get c l < c < c u .
Part(B) can be proved analogously.
Theorem 3.10. (The domain theorem). The domain Domain(D o ) of a DSLS
D o is given by the following [38].
l (c),m
u (c)); or
Domain(D o )
=
[m
c l <c<c u
l (m),c
u (m))
Domain(D o )
=
[c
m l <m<m u
Search WWH ::




Custom Search