A NEW EDGE TRACING METHOD DEVELOPED FOR
OBJECT RECOGNITION
Fatih TALU
1
Department of Computer Engineering,
University of Firat, 23119 Elazig, Turkey
fatihtalu@firat.edu.tr
Yetkin TATAR
2
Department
of Computer Engineering,
University of Firat, 23119 Elazig, Turkey
ytatar@firat.edu.tr
ABSTRACT
In this paper, we present a new efficient edge tracing method, which was
developed to find characteristic vector
for the black

and

white images. The previously
proposed edge tracing methods have two main disadvantages. These are: The edge
detection of only one object in case of more than one object in a numerical image and
investigating only outer contours not inner
contours. By the proposed method, these
disadvantages have been eliminated and all of the contour lines of all objects in a
numerical image have been rapidly detected.
Keywords:
object recognition, characteristic vector, edge tracing, feature detection.
1.
INTRODUCTION
The aim of the edge detection process is to evaluate the information of the
image and reduce it to
adequate
contour line by eliminating the unnecessary inform
ation
that takes time for recogni
tion
process. The edge detection process is one
of the most
basic topics in image processing [1,2] and object recognition [3]. It has been an
interesting research field because of
possessing
very important information in image
analysis and the object classification.
The task of edge tracing plays an i
mportant role in object recognition. We can
explain it as follows: Human seeing system looks at any object firstly during the
recognition period. When we apply this approach to the artificial seeing systems, it is
essential to trace the edge of object effe
ctively to reach a good recognition [
4
]. In other
words, the edge tracing plays an important role in recognition of the edge, too. Many
images do not include concrete objects and to understand these objects depends on their
structural features. The detecti
on of these features depends on edge tracing [
5
].
The recognition can be achieved with the aid of direction changes of edge. For
this, any edge line of the object
is accepted as the starting point and the edge is traced.
Consequently, the number of di
rection changes is compared with the changes in the
edge of the model objects in data

base and the type of the object is determined. For
example, suppose that the shape is a triangle. Any edge of this triangle is taken as the
starting point. There are tot
ally three direction changes from this point.
These changes
help
to understand that the object is a triangle. Tracing continues as far as the beginning
point [
6
].
Edge tracing algorithms are expected to achieve a very high success rate besides
realizing a
rapid, two dimensional seeing
processes
in automation systems. The most
important thing is to be independent of the changes in the dimensions, brightness, place
and posture. In addition, there must not be any information loss during image
processing. On t
he other hand, whole image processing takes a very long time. In the
task of tracing, only the edges of the objects are detected. Thus desired quickness and
performance can be achieved [
7
].
Separately, there are also edge detection algorithms using image
analysis. For
example, Prewitt edge detection, Sobel edge detection[
8
], etc.
However, none of these
methods cannot detect the edges of object in one pixel thickness.
Besides, recognition
process takes a long time because the edges of the object define more
than one pixel. In
this study, we can detect desired counters thanks to developed new edge tracing
algorithm.
2.
EDGE TRACING METHOD
The most important information from the edge tracing is the boundaries of the
object. Statistical or geometrical character
istic vectors are obtained with the information
about boundaries [
9
].
In order to follow the edge of an object in lines, there are some methods like
Hough [
10
] and Rotational Transformations [1
1
], but here, a new algorithm developed
by us is given. The
algorithm can follow the edge of any object as dots. The image of an
object on a data

base and the directions for edge tracing algorithm are given in
Figure 1.
Figure 1.
Directions for edge tracing algorithm on matrix data

base.
2.1. Function edge tra
cing
(x, y, α)
Step 1
:
If α=1, start the tracing process before a pixel.
Step 2
:
The subsequent points are traced like in
F
igure 2
.
a. The first point found
during this ordered tracing is accepted as the subsequent point of the object.
Step 3
:
The coordin
ates of the edge point found in the second step are processed as
follows;
If the point is a point in
)
y
,
x
(
direction, the tracing order is taken like
F
igure 2.a
.
If it is in
)
y
,
x
(
direction, the tracing order in
F
igu
re
2.
b
is used.
If it is in
)
y
,
x
(
direction order, the tracing order in
F
igure
.2.c is used.
If it is in
)
y
,
x
(
direction, the tracing order in
F
igure
.2.b is used.
Step 4
:
If there is no point following in any directi
on, this shows a broken off.
Step 5
:
The edge tracing process is completed when the beginning point is reached by
following the steps given above.
Step 6
:
The detected lines and color information are pushed to
object box
.
Figure 2
.
The edge order of any side points.
The edge order to be used is decided according to the
neighboring
positions given in
Figure
.3.
Figure 3.
The
neighboring
positions of any (x,y) point.
The mathematical expressions to determine
the advancing direction are as follow;
)
(cos
f
x
x
(1)
)
(sin
f
y
y
(2)
0
1
1
other
5
.
0
z
5
.
0
z
)
z
(
f
(3)
The change in
angle can only be in an interval of 45
o
and
360
a+ maximum. In
other words;
for the order in
F
igure
.2.a.
5
4
;
360
;
0
for the order in
F
igure
.2.b.
5
4
;
450
;
90
for the order in
F
igure
.2.c.
5
4
;
540
;
180
2.2. Main algorithm
Step 1 :
Th
e first edge of the object is found by tracing the matrix data

base from the
first line that has a different color from α in the direction
x
. Send the found x, y, and
α values of pixel to the edge tracing algorithm. Before investigat
ing of inside
boundaries of detected object, all of the objects at the same level are detected.
Step 2 :
If
object box
is empty go to Step 4 .
Step 3 :
Pop the first element of the
object box
. α= α’. Go to Step 1.
Step 4 :
Process is completed.
Wi
th the edge tracing algorithm which is developed by
Türkoğlu
[1
2
], the
characteristic vectors can be realized in a short time. Because, the points on any edge of
an object are traced instead of tracing all the determined points of the matrix data

base.
Be
sides, there is no need to derive the edges of an object with this method. It is also
independent of the rotational translation movement of the object. However,
t
his method
has
two main disadvantages. These are: The edge detection of only one object in cas
e of
more than one object in a numerical image and investigating only outer contours not
inner contours. By the proposed method, these disadvantages have been eliminated and
all of the inner and outer contour lines of all objects in a numerical image have
been
rapidly detected
.
For example,
crack points which are
given in
Figure
4 are found and
these found crack points are sent to the edge tracing algorithms.
Figure 4.
Crack points of objects.
3.
CONCLUSION
According to researchers, edge tracing algor
ithms must have two properties.
These are:
Result of the edge subtraction process must be independent from the edge
directions. Recently detected edge knowledge mustn’t change when object
changes direction.
Thickness of edge detection line must be one pi
xel.
In this paper, we have presented a new edge detection method that is based on
the edge tracing method. Developed edge detection process is based on digital input
image knowledge of object. The edge detection problem can resolve at real time by
adding
digital filter to digital image given to system. Besides, developed algorithm has
two basic properties of the edge detection algorithms which are discussed above. It also
eases the task of object recognition especially from the point of view of time.
(
a) (b) (c)
Figure 5.
Some objects which is defined successfully by the new edge tracing
algorithms. a) ori
g
inal image b) only the edge tracing result c) the new edge tracing
algorithm.
4.
REFERENCES
[1]
J.C.
Russ, The Image Pro
c
essing Handbook
CRC Press LLC North Carolina State
University
,
2002.
[2]
I.
Pitas,
Digital Image Process
ing Algorithms and Applications
, John Wiley&Sons,
Inc., New York, Aristotale University of Thessalariki
, 2000.
[3]
J. D.
Irwin, Supervised And Unsupervised
Pattern Recognition. C
RC Press LLC,
Auburn University, 2000.
[4]
F. Bağcıoğulları and C.H., A Edge detection algorithm developed for object
recognition system,
The Symposium of 2. Signal Processing and Applications, s.47

52, 1994.
[5]
S. Banks, Signal Processin
g, Image Processing and Pattern Recognition. Sheffield
University, 1990.
[6]
H. Uzunalioğlu, Model

Based Recognition of Polyhedral Objects, Master Thesis,
ODTÜ Science Institute, 1992.
[7]
R.O. Duda and P.E. Hart, Pattern Classification and Scene Analysis. Stanf
ord
Research Institute, 1989.
[8]
P. Ioannis, Digital Image Processing Algorithms, Prentice hall
international series in
signal processing, 1993.
[9]
K. Marefat, Geometric reasoning for recognition of 3

D object feature, IEEE
Transactions on Pattern Analysis and
Machine Intelligence, vol.12, pp.950

964,
1990.
[10]
Y. Shirai, Three
–
Dimensional Computer Vision, Springer
–
Verlag, 1987.
[11]
C. W. Kang, etc, Extraction of Straight Line Segments Using Rotation Transform,
Generalized Hough Transform, Pattern Recognition, v
ol. 24 (7), pp. 633, 1991.
[12]
Türkoğlu, İ., ve Arslan,A., “An Edge Tracing Method Developed For Object
Recognition”, The 7

th International Conference in Central Europe on Computer
Graphics, Visualization and Interactive Digital Media'99, 9

12, Plzen, Slovak
ya,
1999.
Comments 0
Log in to post a comment