the window that is centered by (a0) which represents
f(z,y). The image field g(x,y) can be found by using one
of the following operators :
1)g(xy)-Max[[|ao-a2|]l , |a30-34] , |ao-ac| ,
|ao- ag 1 @)
2)gXy)--- (22 + 24 + 36 + 33 -420) (3)
8
38*y)-20.-1- Pi (4)
2.2.2 Edge Thinning
One of the algorithms was developed by Zhang and
Suen ,(Gonzalez,1987), for thinning binary image. In
this algorithm it points have value(0). The method con-
sists of successive process of two basic steps applied to
the contour points of the given region, where a contour
point is any pixel with value (1) and having one 8-
neighbor valued (0) with reference to the 8-
neighborhood definition shown in (Figure 2), the first
step flag a contour point P1 for deletion if the following
il B. ET
4) g(x,y)= Max [|ap-a2[] + |ap-a4| + |ap-as| + conditions are satisfied:
| ao - ag] (5)
Directional edge enhancement can be performed by con-
volving original image array with the compass gradient
mask .
Another operator known as walls operators (Pratt
,1978), is used. According to this scheme an edge exists
if the magnitude of logarithm of the image luminance at
a pixel exceeds the magnitude of logarithm luminance
of its four nearest neighbors by fixed threshold value;
1
g(,y) 7 log [ fGy) ] - "I [ log a; + log a3 + log a5
«logaz ] (6)
or
nin f(x,y)
arr loge) (7)
2.2.1. Thresholding Techniques
((Binarization))
In digital image processing, thresholding is a well-
known technique for image segmentation. Because of its
wide application, quite a number of thresholding meth-
ods have been proposed over the years.
Let g(x,y) be an image to be segmented and T be a
threshold. The result of thresholding an image function
g(x,y) at gray level T is a binary image function e(x,y),
such that :
bo... df, Sy «T
iun bs if g(x,y) >T
(8)
where (b 0 , b 1} is a pair of binary gray level.
In general, a threshold method is one that determines the
value of T based on a certain criterion ,(Sahoo, 1988).
Threshold selection is one of the key issues in image
segmentation ((e.g. edge detection)). In threshold selec-
tion the following two points should be considered :
1- If the threshold level is set too high, it will not permit
detection of low amplitude structural image elements.
2- If it is set too low then it will cause noise to be falsely
detected as an image edge
30
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
a)2<N(P,)<6
b)S (P1)=1 ©)
c) P2.P4.P6=0
d) P4.Pg.Pg - 0
where N(P1) is the number of non - zero neighbors of
P1, that is
N(P,) = P2+P3+P4+P5+Pg+P7+Pg+Pg (10)
and S(P1) is the number of (0-1) transitions in the order
sequence of p2,p3,......,p8,p9. for example, N(p1)= 4 and
S(p1) = 3 in Figure 3.
P9 P2 P3
Pg Pq P4
P7 Pg Ps
Figure 2 Pixels arrangement within a window
0 0 1
1 P1 0
1 0 1
Figure 3 Illustration of conditions a and b in equation
(9) In this case N(p1) = 4 and S(p1) = 3.
In the second step, condtions (a) and (b) remain the
same but conditions (c) and (d) are changed to,
(c') P2.P4.Pg = 0
(1)
(d')Pa.Ps.Ps =0
step (1) is applied to every broader pixel in the binary
under consideration.
If one or more of the conditions (a) through (d) are vio-
tep (1) is applied to every broat is not changed. If all
condition are satisfied the point flagged for deletion.
2.2.3. Edge Linking
One of the simplest approaches for linking edge points is
to analyz
borhood
image th
points tl
boundary
The two
larity of
1- The si
used to f
2- The d
The fir
Gmag[f(
ordinate
(x,y), 18 |
| Gma;
Also it i
fined ne
‚pixel at |
| o (x,
where /
and a (.
Gmag
QL =
Based o
fined ne
if both t
This prc
A deve
ments i
edge de
differer
techniq
clear e
noise.
Figure :
nique Ir
The aut
ing the
cific siz
tensity
subima
ment is
sity is |
a tuning
ed as fc
