Image Processing Reference
In-Depth Information
Fig. 4.15.
Examples of equidistance surfaces (surfaces for the distance values
d
s
=
r, r
= integer are shown for distance functions
d
s
,s
=
1
,
2
,...,
7
, in Table 4.6). In
d
3
of (3), slash line = double structure. In
d
4
of (4), slash line = double structure
for
r
=even.In
d
5
of (5), slash line = triple structure for
r
= even, double structure
for
r
= odd. In
d
6
of (6), slash line = double structure for
r
= even, dark = triple
structure for
r
=even.In
d
7
of (7), slash line = double except for
r
=
3
n −
2
,dark
= double structure for
r
=
n
.Here
n
-fold structure means
that
n
equidistance surfaces exist on the plane perpendicular to the coordinate axis
such as
k
n
−
1
,triplefor
r
=
3
3
−
j
plane.
Then, for an arbitrary
β
M
and arbitrary pairs of pixels, if and only if the
length of the minimal path for the variable neighborhood path between two
pixels with the neighborhood sequence
β
M
is never longer than that with
β
M
,
the length of the variable neighborhood minimal path with
β
M
becomes a
distance measure.
Furthermore, for an arbitrarily given constant
c
, the relative error of the
distance according to the above minimal path length to the Euclidean distance
can be kept smaller than
c
by finding a suitable neighborhood sequence. How-
ever, the absolute difference between such distance and the Euclidean distance
cannot necessarily become smaller than
c
. In other words, there exists a pair
of pixels such that the distance between them measured by the minimal path
length with the neighborhood sequence differs from the Euclidean distance
by the amount larger than
c
. For proof of these theorems, see [Yamashita84].
The similar results may be expected to be correct for a 3D image, but have
not been reported yet.
Search WWH ::
Custom Search