CONTRAST
ENHANCE~mNT
AND EDGE
DETECTION TECHNIQUE
Ghalib H. Mizaal
-assistant researcher-
And
Rehab H. Alwan
-Scientific researcher-
Remote Sensing Department, Space & Astronomy Research Center
Scientific Research Council, Baghdad, P.O. Box 2441, IRAQ.
Commission number
3
ABSTRACT:
This paper proposes a technique for locally finding edges
in gray level digital images, and enhance the edge and non-edge
pixels. The technique starts by dividing the image into partitions
based on the amount of spatial activity in a neighbourhood of a
pixel, extracting the edges in the partition and enhance the edge
and non edge pixels by stretching. This method is illustrated by
application to a digital image consist of a rectangle on a lighter
surround, both with constant gray level. This technique
easily
detects and extracts the edges depending on the edge location, the
rate
of
change
of brightness and the average brightness.
Moreover it enhances the entire non edge pixels to obtain an
accurate enhanced image.
547
INTRODUCTION:
Image enhancement is one of the important operations in
image processing, where it is used for multi purposes and in many
situations. The objective of image enhancement is to improve
picture quality, more specificaly, it is employed to remove noise,
deblur edges and highlight specific features. The application of
image enhancement techniques, in general improves human viewing
ability and may increase the distance of success in picture
processing [1].
Noise in an image generally has a higher spatial frequency
spectrum than the normal image components because of its spatial
decorrelatedness. Many methods have been proposed to remove noise
and enhance the digital images such as Wiener filtering, simple
low-pass
spatial
filtering,
other
approaches
include the
constrained least square filter, optimal recursive filters, and
variations and extensions of these filters. All of the methods
evaluated are devised to preserve edge sharpness while achieving
some degree of noise cleaning, they are unweighted neighbour
averaging (AVE),K-nearest neighbour averaging (KAVE), the edge and
line weights method (EDLN), gradient inverse weighted smoothing
(GRADIN), maximum homogeneity smoothing (MAXH),slope facet model
smoothing (FACET), and median filtering (MEDIAN),each method has
its own algorithm, and as a result all mentioned algorithms
observed that are very good for preserving edges but they increase
CPU run time [2].
Concerning
the
contrast
enhancement, the
transform
processing such as the Fourier and Hadamard transforms which
provide a spectral decomposition of an image into coefficients
that tend to isolate certain features of an image [3]. The
histogram specification method which can often be improved by
amplitude rescaling of each pixel [4], and many other methods
which all perform a computational requirement.
The most significant features in any image are the edges:
Many different edge detection shemes have been advanced and
analysed, (see Refs (3,5) for surveys and Refs (6,7,8) for
comparisons
of some well
known techinques). SHIOZAKI (6)
extracted the edge using the entropy operator which calculates the
entropy of brightness in a local region of an image [6], and
yields low values in the regions where brightness is changing
rapidly. The entropy operator as well as the Laplacian is apt to
be affected by noise, so noise must be removed beforehand.
The paper presents an approach used to enhance the contrast and
detect
the
edges by partitioning the image to reduce the
computation and to do local processing which is summarized as
follows:
548
a.
Defining the partition, which means dividing the picture into
specific number of partitions.
b.
Extracting the
edge picture.
c.
Enhancing the detected edge
obtain an enhanced picture.
edge
from the defined partition and give an
pixels
and
the rest pixels to
2. ALGORITHM
The
1.
proposed technique is described as follows
Partitioning
the image:
In order to get a very accurate description of spatial
activity which defined as a rate of change of intensity from one
pixel to its neghboring pixels, meanwhile working with small number
of pixels leads to computer time reduction: For these reasons,
partioning of the image becomes necessary.
Basically, the signal model contained two regions, regions
with high spatial activity containing the edges, and regions with
low spatial activity containing the non edges.The partitioning and
the local processing on the image looks like filtering the image
by many bandpass filters depending on the spatial frequency
contents. Partition size is taken as 4x4, pixels.
2. Edges and non edges pixels discrimination:
The whole image is divided into L subsets or partitions
and each partition is represented by ki , where i
l, ••• ,L.
Discrimination of regions ki are based on the following criteria:
=
- Calculation
p(ki )
the mean value of the partition ki
1
=
N
~l
N
---... ----.... (1)
X(kij)
Where N= size of partition ki
Calculation of the maximum and minimum
partition (Max X(ki), and Min X(ki».
- Finding the rate of change R(ki)
decision
R(ki)
Max X(ki)
of
the
standard
intenSity value in the
which is
Min X(ki)
=
calulation
partition:
using:
deviation
549
used
of
the
for the region
mean
in
the
tf
==
1
N
SQRT [ ----N
j==l
L. ( X( kij )
2
f
(ki»
]
-------- (2)
Since the homogeneity is represented by the reciprocal of the
variance [8], the standard deviation is used to measure the
homogeneity and to define the dispersion of intensity in the
partition. Each region in the image has an upper and lower limit
of intensity with specific homogeneity and the whole regions of
the image are seperated by the edges, so the rate of change R(ki)
is used to decide the existance of more than one region common in
the current partition ki
H
<
R(ki)
where H is a flexible threshold depending on the user purposes,
when H increase, the detected edge decrease and homogeneity of
regions increase and vice-versa, that means H decide the height of
the detected edges. see figure-l
\L i
r mean)
------""---
-- - ---
f ( )( )
h(X}:;:f(x) -f.n(x}
h(X)
Figure 1- shows that the partition is divided into two
parts the higher which contains pixels greater than the mean
value f;and the lower one.
As a result, there are two types of partitions:
al The edged partition •
bl The non - edged partition.
550
a.
The local processing on the edged partition; In this
partition where the edge does exist, the definition of the edge
pixels and their locations are allocated by taking each pixel
the partition ki to satisfy the following
value X(ki)
in
conditions:
1. X(ki)
> P(ki)
and
2. at least one neighbour X(ktp
<
f(ki)
for J f= i
If only the first step is satisfied or If X(ki) is unique and not
located on one of the four sides of the partition then X(ki) is
considered as a noise, and substituted by the mean gray level of
the partition.
i.e ..
X(ki)
r
==
(ki)
otherwise the minimum value of the edge pixels is found -MinEand their mean value is calculated:
re( 1>.i)
D ==
l
±.
1
:::: _ __
E
E==l, ••• number of edge pixels
X(kij)
J==l
(ki) - MinE
where D is the difference to be compared with lower and upper
limit values taken to enhance the edge.
i.e.
IF D
> FU
~
fe ( ki) == re e k i) + FU
IF D
< FL
--.,..
fee ki)
==
fe( ki)
+ FL
where FU & FL are the upper and lower limit values taken to get
the proper stretching enhancement.
Finally the edge pixels are enhanced as follows:
if edge
if
if
pixel
==
==
==
==
X( ki)
> re( ki)
~
XC ki) == fee ki)
X (ki) + 51
< fee ki)
-+ X( Ki)== f( ki)
X (ki) + 52
<f
---7 X(Ki)==
(ki)
+ 51
X(ki) +
52
where
51 and 52 are an adaptive thresholds taken to enhance the
edge pixels and remove the effect of the noise on it.
The non - edge pixels in the same partition are enhanced depending
on the calculated mean value wen- of the non-edge pixels:
IF
x (ki)
>rn( ki)
X (ki)
<f
---+ X( ki)
(ki)~
== rn( ki)
X(ki) == X (ki) + t
551
--------(3)
-------... -(4)
where t is a variable threshold given by the user interactively
and un is the mean value of the non edge pixels in the partition.
b. The local processing on the non-edge partition; The non-edge
pixels are enhanced by stretching and treated just like the
non-edge pixels in the local processing on the edge partition.
Equations 3 and 4
The technique described above is in terms of gray level
distribution.
Elementary
operation, such as enhancement, noise
suppression, derivation of textures, can be performed without
blurring edges of object; On the contray edgs are extracted and
defined by their locations. Regarding the noisy pixels which
appear as edges, they are easily recognized and enhanced by
substituting the mean gray level of the partition for the gray
level of the noisy pixel.
The (4x4 pixels) partition is scanned over the image only
once, as a result the set of the values of all pixels forms the
enchanced picture with detected edges.
c. If the current and next partitions do not contain an edge
pixels, then an additional step is done by taking intermediate
4x4 partition as shown below, which is composite of last half of
2
3
4
1 '2
1
2
3
... Formgr
parti tion
3 4
4
-
Intermediat Later
partHion
partition
first partition and first half of the later partition. Then the
new partition is checked for the existence of an edge pixels*.
3. Evaluation of the algorithm based on a test image
To assess the performance of the above mentioned algorithm,
test images are generated containing edge signals which are
characterized numerically. Two test images wher used, one consists
of
rectangle
on
a
lighter
surround
both with constant
gray level 5 and 20 (see figure 2.a) and second consists of two
*Other probability is taken when one of the current or the next
partition contains an edge pixels, then the difference between the
mean value of the two partitions is compared with the taken rate
of change to decide if the adjacent side of the two partitions
contains an edge pixels.
552
SAftPIE ~f GRAY SHAtfS
1 2 J 4 5 II 7 t:
9 10 11 12 1l 1q 15 III 17 18 19 20 21 22 23 2Q 25 26
------....... H») 111ZZZXXXAAAIl1ln8t18COItUUUUU . . . . . . . UUU.UUIIIIII1II.
- c-
-b-
-a-
ZZZZZZZZZZZ7.ZZ2ZZZZZ
ZIIIIIIIIIIIIIIIIIIIII
IIIIIII1III11111I1111
1111I111111I1I1111111
1I111111111II11111111
11I111111I11111111111
111111111111111111111
II1I11111111111111111
1I11II11II11111111111
1II111111111111111111
I11III11111I111111111
1I111I1I11I1111111111
1I111II11111111111111
111111111.11111111111
1II1111I1I11111111111
1II111111111111111111
111111111111111111111
111111111111111111111
111111111111111.11111
III1II1I1I1I111I11111
1I11II11111I111111111
III111111111I11111111
1I111I1I1I1I111111111
EEEEIEEEE1EfEt~EEEEEE
ZIXX'~f.~,~t'~~&~!'~~=~
ZIX j(ajlJ;If.41Ujf.d~~141fj~td:::;'f
ZI~~ilflttlltillai~iXI22
ZI~&lltilittlililia'XIZ2
ZI~&III!lltlfiff~iIIXI2Z
ZI~&ltllltlliaiiilliXl2Z
Zll~lli'lliil~iiflliXIZZ
ZI~~iiii!lllillllfiIXI2Z
ZI~~'tllllif'tlttttiXIL2
E
F
F
E
E
E
E
E
E
E
E
E
F
E
E
E
E
E
E
E
E
E
ZI~~fiia!tliitliljl!XI~Z
E
Zl4&lillllfllilllltlXIZZ
E
ZI~&illjiftiittlllliXIL~
E
ZI~~lilllfifiitl~'~IXIZZ
F
1:
ZI~~llltf'flial~!tI5XI~Z
E
E
ZI&~'tf51IilltltltliXI2Z
ZI~~ltiilalifi~tiltiXI2Z
ZI!~!llf~la«ft!il'I~XI~Z
ZI~~i~~lli!t!iltfif~~IZZ
1==lXXXXXXXXXXXiXXX a
iillllllllllllllllill
E
E
E
F
F
E
E
F
E
EEEEEFEEEEEfE1EEEEEEE
z:.:Z"ZZZZZZ22ZlZZ
ZZZZlZZZ2ZZ2.2ZZZ
II~
/I
-
II
d-
II
•
II
II
II
II
/I
•
I
II
II
II
II
.
/I~~~"~H8Ht"~KIIHlllllllllln""ftr~lIn811
11 ')111111111 ~1Ir.llli 111111111111/11111 foil!' ~ 1\ !!1I1111
.-- ••••••• ++ ••••••
IInKIIIIMMII~~"~~HKnll~n811"~!~"nllnll
1Ir.1I1I.1l""IIIIIIr./lI'.II~ 1Ir.1I1!r.I!/lII/lM II I! II II II
1111 ~ IIr.H!11! r. r: r.t'./lr. ~r.t: 1Il1l1l1illlllf.1l11f1 r.!I
.-;.+++ ..................... , • .04
I--.f.' ••••••• f
IIII~n/llinl1lif~~III\II~lItnnllllllllnllllllllll
tBI\""flllll1~II •• all •• II.~KMIIIIIIIIr:1I
1I'/\III1I1r.11I1I\.UIIIIIIIUlnp.IIIIIIMIIIII
I1I1I1HhM~MII"al'I' ••• 'II~~e~rll~MII
. - - . HI 1 ••
::::::=~:::::~::::::::::::::::
/I
.
II
'--R
lrllt.III1I1/1lll11 I. t
It
I""
H..
1.
t
L
'--1
.
I ................... f .--.
------------------------
------------------------
II
•
II
II II
II
•
I,
E
I:
f
f
!:
E
fL'ltl~t2FE
k
I:
E
E
f
f
F.
f.
f
E
f
l
J:
£
•
!
!
First original test image with gray level (5, 20)
Enhanoed image
Enhanced edge image of (a) with rate=15 STD= 2.4
Second original test image with gray level (5,7,10)
Enhanced image
Enhanced edge image of (d) with rate =2 and STD = 0.4
553
~
e
f
f
f
k
t
l:
f
II
II
f
t
E
I:
I:
E
II
II
r.
HtHHHEF.
!
E
F
E
r:
HHHHffHr~eHFHHfHi!HU
•
1/
•
II
a.
b.
c.
d.
e.
f.
f.
E
t
IlUII . . . . IIII . . . U . . . . . . . UIIIIIU . .
/I
Figure 2,
E
l:
f
1:
E
f
f
.-- . . . . . . . . . . . . . . . . . . . . . . . . --111
11--' " ' 1 " • • • • • • • • , •• If< ••
II
"
.f." +...... --11
I--t ... ++ ............. ,t H I - - .
11.- ••••••••• f t . I t . " l l i •••• --1
II
II
t
f
... - - .
.--............. ,...........--.
.--.i .................. ,.,.--.
1--"
I
l
f
\11--" +••••
.--H ......
"KnI\E~/1I\"hI!MREMMII~I\III1""8I1fMII~1I
~""""nl\/lHr~~~IIIIHllllnanllr.IIIIMnllr.1I
IIIHIMI! r.llllllt!1 nt.!1nnlil1/1I1I1I1I1I1r.IIII!I~1I
IIM~lIlInlllll!l1l1r./1118KlIlIl\lIl1ftllll~MII~lIn
1I111111r.l11l 'III ~ ~ III! 11 !'I I! II t./I ""/lIIM/I /1/111/1
"""XnH/I"II~~II~l1nHIIIIH"~1I113nnnKIIII
II~ nr.lI11 1111 PlMi "II~ II 1'1 11/1 11'/1 II M ~, 11/1/111/1
k
E
E
11--11++ .. 'IM~~r'.!'I:Jllt; 18"" "--II
1I-- •• H •••• r.n;I1I1l':III11S •••••• --.
11I;1r.t:i1IMl1S
,--1
,1 n 1111.11'
+--11
11111111111\/1 hI! ~II ••• I. I •• IUIII ~ II IIH/I 1111
"tlll!~~Mlllld •• " I ' •••• ~lIrl\~~~1I1\1I
llro II lit 1\1111111'.1.1 1.11 ••• III /111/1 /111/11111
I': II II II 1I111111 11 MI II .ac ..... I1I1I1/l r.1I1I1l1I
III\HII/1~MKII~ ••••• I •• I.IMIII\I'.IIIIII"ft
1\ N1111111111 nMII ~r. '\ (111 nti 11M/! Ii !I II e1111 roll /j II
II I'! II MI! ~ nil/! 1\11 nil f.11I111111111 nllll ~II el111 til
f
f
£
~
•• - - .
.-- •••••• H ••• a •••••••••••• --1
. - - •••• t ••• M~r~~"~Mll1 •••• +t--.
.-- •••••• fllit,~/I/lilll!'\ 1.i.f .. t--I
.-- •••••• +1I~.HI1I'I1r.lIl1.'. t • • - - .
.-- ••••••• In!\!1~Ml1l\lIlII'.' ••• --.
II II 1111 1Ii111 r.IH r. ... ~, ~ r. 11 1111 nil" r:1I1!1I II 1111 Il II
1111 ~1I1'Ir.1\1I11 ~;a. '111I"1I"!1~ 111111 IHI II II
II
~t~tk~~E&LElttct£t~tE~E£f!E£EE
.. - - . . . . . . . . . . . . . . . . . . . . 1 • • • • - - .
r.lIlIlIl\rft""lInK~II'"~~MM"~II'l\lill~1I1I
E
~
E
1-- ••• " " ••••••••••••••• H--II
II-- ••••• fl • • • • • • • • • • • • • • • • • - - .
1I1\1Ir.IIII/I~p.r.r.M~rIlIlM/lr.~lIl\a/lr.nIlI1MII
II
l.tH H
f
fE
-----------------------a--I-----------------------••• I. "I.".'.' •••• 111.--1
IIIIIIIIIIIIKr.llhllll""1I1I1I1I1I1I1I1Ir.8MP.llllftll
11
-
••
•
•
.
•
l u •• I U n •••••• IIII . . . . . . IIII8I
II
/I
II
-e-
t
E
E
k
E
E
rectangles one inside the other with spcific modification to
measure the ability of technique in orientation detection, gray
level used in the second image is 5, 7 and 10 (see figure 2.d).
Many types of noise cause image degradation. A model which
is widely accepted as accurate for noise is that of additive
random gaussian noise, which is generally represented as a random
stationary sequence added to the ideal image grey levels in the
form of h = f + n, where h is the corrupted image, f is the ideal
image, and n is the noise. A set of test images are generated at
different signal to noise ratios using different formulas:
r1 - r
2
a.
SNR = (
2
)
~
where
pl
background
~
p2
&
are the
respectively
mean
value
for
the
foreground
and
is the standard deviation of additive noise.
This formula
of edge detection.
ds
b. SNR = _____
was
defined by [5] and used in the evaluation
dN
where dS is the standard deviation of the signal
6N =
=
= =
=
=
=
noise
This formula was defined by [9] and applied on the test image
for the evaluation of contrast enhancement & edge detection.
DEFINITION:
Let f= uall; Ri
be the noise - free image partitioned
into k disjoint regions Ri characterized by i pairs (Xi, Ti),
where Xi = {Xl is a set of pixel locations X € Region Ri and
where Ti is the thresholding range specifying the partitioning
criterion, i=l, ••• k. Then a set of performance indices {€i },
i=l, ••• ,k based on the mean - square error is computed as [2]:
€i
=
1
2
h(x)}- f(x)]
L
where hex) is the intensity at x of the noisy image.
Q
is a smoothing operator to clean noise & preserve edge.
L
is a normalized factor used for scaling.
These measures €i characterize the departure of the processed
image
from the ideal image at different levels of spatial
activity.
Figure 3 is a block diagram describing this evaluation method.
The smoothing algorithm is applied to the set of simulated
test images and computing their performance indices. First the
test image was partitioned into two regions regions RI, and R2, a
region with low spatial acitivity (homogeneous areas) and region
with high spatial activity (edges), respectively. The choice of
K=2 is based on the fact that the test image has two major levels
of spatial activity (see figure 4a).
To study the effects of noise, iterative processing, and
the
ability
to preserve different types of edges, several
simulation runs were made using different combinations of signal
-to noise ratios and iteration. Some of the results were plotted
as shown in fig. 4. It is worthy to note that edge detection using
this algorithm is independent of the edge slop and orientation but
edge location, so it is not necessary to take thresholding of the
image to get the edges because their locations are already
defined.
This is an example of smoothing algorithm that is designed
not only to preserve edges but also to enhance edges and making
a contrast enhancement for the whole image •
n
Q(h)
Fig. 3 Block diagram of the image evaluation system using image
partitioning
f = ideal image; n = additive noise; h = noisy
image; Q = noise cleaning operater: Q(h) = restored image; Ri =
partitioned regions of f such that UiRi = f; Pi = partitioned
regions of Q(h) such that UiPi = Q(h).
555
a
b
1. 3
1.
a-
_-*
1. k
X
,0 --
\
L
I
L
L
W
Q)
'"
.7
> .6
"7
+>
It!
* --
• ..J;-
o
.5
Q)
a:: . 4
.2~
. li0
'.
'\•..
,
"
\
.3~
1
--------
1 .3
\
I
-*I
X
1
1 .2r
1.
0
Rate=l ,STD=.41
Rate=2 ,STD=. 4!
Rate=4 , STD=. 41
"\
\
9
~L
81
Q)
•
.7
.4
\
.3
~
.2
\
e
45
10
25
Signa 1 to Noise Ratio
t1-i:,2
0'
\
2
100
I
I
0
I
I
I
I
\\
\\
\
*--
1
25
10
45
5 i gn a 1 to Noise Ratio
A
..Q
.p
-,-t
:::
H
0
H
H
i%l
Q)
~
6.
!i
til
II
~
I
\
1
.,-t
I
I
I
Q)
\
~'
\
\
~.5
H
Q)
~
~
II
\
+>
fL
't
~*
\\
> 6
......
\
SN? == (-----------)
"
L·
~Q)
Rate=2 ,STD=.2
Rate=2 ,STD=.4
Rate=2 ,STD=. 8\
X
W
\
~
til
I
I
100
§
..q
•
~
-,-t
Pee
(0
l()
l()
4. DISCUSSION AND CONCLUSION
Edges regions are the regions which involve abrupt changes of
brightness. Although there is a considerable and growing body of
literature devoted to edge detection and image enhancement. Here,
we have introduced a technique that detect the edge and enhance
the region locally by working on a partition of the image. The
technique decides first if the partition has an edge pixels or
not, depending not only on the rate of change of brightness, but
also on the average brightness in the partition and on its
standard
deviation. Then the edge pixels are extracted and
enhanced as well as the non-edge pixels. We can get a smooth Wedge
picture- with noise reduction and an edge extraction, depending on
edge locations, and independent on edge slop and orientation which
is the most important step in the known edge - preserving noise smoothing
techniques (AVE, KAVE, EDLN, GRADIN, MAXH, FACET,
MEDIAN).
The stretching contrast enhancement used for the non edge
pixels
defined
on basis of a threshold which is selected
interactively
corresponding to the user purpose. An evaluation
procedure is performed which involves:
(1) the division of an image into partitions based on the amount
of spatial activity, and (2) the measuring of mean
square
errors
for
each partition. Evaluation studies on simulated
images were made under the effects of different noise levels.
The evaluation measures were used to compare the effectiveness
of
the proposed
technique
taking
different
standard
deviations and
rate of changes. They perform well in both
homogeneous areas and edge domianated areas. Results also show
that, using iteration, the technique achieve additional noise
reduction and converge to their corresponding error measures.
Regarding the proposed algorithm with other algorithms.
First the algorithm discussed above deals with the three types
of image enhancement simultaneously which are 11 Edge detection,21
Edge enhancement and
31 Contrast enhancement, other algorithms
deal with one of the above three types. Second the use of (4x4)
partition and accessing the 16 pixels at a time to enhance an
(nxn) image, reduce the computational time by n2 divided by the
size of partition, in which the iteration is done by shifting one
partition at a time to cover the whole image; other methods use
partitionining to enhance one pixel at a time, and iteration is
done by shifting one pixel to cover the whole image. Thus to
enhance an (nxn) image using this method needs n2 step which leads
to higher computational time. Third, in trms of relative error,
the proposed algorithm gives significant noise reduction in the
edge and contrast enhancement in comparision with other methods
which use one of the mentioned functions. (see ref.2, the curves
of relative error with SNR).
557
REFRENCES:
1-
A.
DE
~~BUQUERQUE
ARAUJO,
Slliu
obsolute
grey level
differences
M1 edge
preserving smoothing
approach,
electronics letters 5th dec. 1985. vol.21 ~o. 25/26.
2-
nOLAND T. CHIl·j AND CAL2\. - LUHG YEH, Quantitative Evaluation
of some edge - preserving noise - smoothing techniques, NOTE,
computer vision, Graphics and processing, 23, 67 - 91 (1983).
3-
A ROSEnFELD AND A.C. K.A:Z, Digital
ed., Academic press New York, 1932.
4-
B. CHMIDA, B.B CHAUDHURT and D.DUTTA I-Il0U:1DER, On image
enhancement and threshold selection using the gray level
co-occurence matrix, Pattern recognition letters 3 (j uly, 1935)
243-251.
5-
H.K. PRATT, Digital Image Processing, \'lily, l1eH York, 1978.
6-
N(IRA
SHIOZAKI, Edge Extraction Using Entropy Operator,
Computer vision, Graphics and image processing 36, 1-9 (1987).
7-
N.A.
FURST
AND P.E. CAINES Edge Detection \.;ith iIJage
Enhancement via Dynamic prograa~ing, Computer vision, Graphics,
and image processing, 33,263 -279 (1986).
8-
ZBII:1GNIEH E. nOJCIK, Z'-!;. natural approach in image processing
and Pattern recognition: rotating neighbourhood technique, self
-adapting threshold, Segmentation and shape recognition, pattern
recognition, Vol 18 No.5. PP 299-326 1985.
9-
OPPE~lHEllI,
picture Processing, 2nd
A.V. (1978): Application of digital
Processing, prentice hall, INC, NevJjersey - U.S.A
558
Signal