lated using
neralization
] erosion is
ed by this
discussion
ize (within
the natural
| in the case
cess called
)w such a
nplified
S
ze simplified
s well for
ogy
veloped in
jpicting the
on process.
| discussion
nent B
Qooooo oo
Q:oo0cioo:o0 ooo
0B)
ion and
n and e»
form and
two French
geostatistical scientists -- G. Matheron and J. Serra -- in the
1960s (Matheron, 1975; Serra, 1982). It has since then found
increasing application in digital image processing. The basic
morphological operators are dilation and erosion. They are
defined as follows (see Serra, 1982; Haralick et al, 1987, Li
and Chen, 1991):
Dilation: À ® B = {a + b: ae À, beB} = Up eBAb (1)
Erosion: À © B = {a: a+ beA, beB} =NpeBAp (2)
where A is the original image and B is called the structuring
element, which can be considered to be an analogy to the
kernel in a convolution operation. In Eq.(1) it is called
"dilation of A by B" and in Eq.(2) "erosion of A by B".
Examples of these two operators are given in Fig.8, where the
features are represented by pixels of “1”s and the origin of the
structuring element is marked with a circle. The structuring
element is a critical one in these operations. More discussion
regarding this element will be conducted at a later stage. To
show how these operators work clearly and exactly, “1” and
“0” are used to represent the binary images used in this
discussion. In this diagram, “+” means those becoming 1 after
dilation and “-” means those becoming 0 after erosion. This
convention will be used throughout this paper.
If a symmetric structuring element with origin at the centre is
used for dilation, then the shape of the original image will be
expanded uniformly along all directions, thus the dilation in
this particular case is called expansion. Similarly, the erosion
in this case is called shrink. These two special operations are
illustrated in Fig.9.
0.0.0. 0:0.0:0..0. 0
0,000000 00
07O 111000 0
039/173-3 10 0^0 L Lal
00:21 L'o 00 6 120 4
0.0:0/1.0.0 0.0/0 111
0.-.0,0 0:0 0,0.0.0
000000000
(a) Original image A (b) The structuring
element H
9000. 000000 0.00 -0.0.0 0 0.0
O0 + ++++000 0000900000
0 +111+ +00 00---0000
0. 1 11r 0 0 00-1--000
0+1211+4+ +40 0 0.0 .- -- 0.0.0.0
0 ++ 1+ +000 0 0:°07* 0.00 0..0
00+++0000 0° 0.0.0:0°0 0 0 0
0.000,00 0 0,60 005010 0 0.0 D 0
(c) A expanded by H (d) Ashrunken by H
(A ®H) (A ® H)
Fig.9 Special cases of dilation and erosion
expansion and shrink
Another two very important operators are opening and closing.
They are defined as follows:
Opening: AoB=(AOB)®B (3)
Closing: AeB=(A®B) © B 4)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
where, A is the original image and B is called the structuring
element.
Based on the two basic operators, i.e. dilation and erosion, a
number of other new operators have also been developed, such
as thinning, thickening, hit or miss, conditional dilation,
conditional erosion, conditional thinning, conditional
thickening, sequential dilation, and conditional sequential
dilation, and so on. However, it is not the purpose of this paper
to discuss all of them. More detailed information can be found
from the book by Serra (1982).
Structuring element is the key element in a morphological
operator. Structuring elements could take any shape. Fig.10
shows some of commonly used structuring elements. Indeed, it
is through the proper manipulation of structuring elements that
the morphological operators alter the shape, form and structure
of spatial objects.
B x =
(a) Circular (b) Diagnonal (c) Linear
(d) Squared (e) Cross
Fig.10 Some possible structuring elements
6. DISCUSSION
After the introduction of so many new concepts such as scale
dimension, natural principle and morphological operators, it
seems appropriate to usev an example to illustrate how
morphological operators can be used to depict the
transformation of spatial representation in scale dimension.
Fig.11 is one of those example (Su et al, 1996). In this
diagram, the size of the structuring element B is determined by
the natural principle. After applying some morphological
operators, the representation shown in Fig.11(a) is transformed
into that shown in Fig.11(c). Features are smoothed and
combined. The reduced image is shown in Fig.11(d).
8* .
(a) Original features A (b) Structuring element B
de
(c) Areas combined (c)Combined area reduced
C=AeB C=AeB
Fig.11 Transformation of the representation of area features in
scale dimension: area aggregation (Su et al, 1996)
457