2001
年
2
月
第
38
卷第
1
期
四川大学学报(自然科学版)
Journal of Sichuan University (Natural
Sc
ience Edition)
Ar
ticle
ID:0490.6756(2001 )01.0012.05
Detection
of
Step-Structure Edge ßased on
Order
Statistic
Filter
.MA.
Hong
1
, YU Yong
1
,.MA. Li2 ,
M.
Umeda
3
Feb. 2001
Vo
l.
38
No.l
(
1.
Mathematical
Co
llege , Sichuan University, Chengdu 610064 , China;
2.
De
partment of
Co
mputer
Sc
ience, McGill
University
, Montreal,
QC
,
Ca
nada; 3. Faculty of Engineering, Neyagawa, Osaka, 572, ]apan)
Abstract:
In
the
theory of edge detection of binary image, the classical method
is
to
use regular convolution
kernel to construct edge detect operator
, such as
So
bel operator, Prewitt operator, Kirsch operator and Roberts
operator etc. However
, these operators for the complex geometric structure of digital image edge lack algorithm
adaptability.
The
authors propose stochastic filtering to edge detection. Using order statistic filter to construct
stochastic convolution kernel
,
the
authors yield a new kind of stochastic edge detect operator-OSF edge operator.
The
authors carry out the edge detection for radar images and document images , and the experimental results show the
efficency of
the
OSF
edge operator.
Key words: binary image; ste
p-
structure edge; edge detection; order statistic filter;
OSF
edge operator
CLC
number:
但
11.
9
Doc
ument code: A
(2000 MSC
68U
lO)
1 Introduction
According to a psychological result of the survey and estimation, the information captured by the humain brain ,
85 % comes from the eyes , 10 % comes from
the
ears, and the remaining 5 % comes from the other sense organs.
Hence we can assert
that
in the computer multimedia, 85 % of all the data we are dealing with
is
image data.
It
fully
prov
臼
the
importance of digital image processing in the multimedia technology.
The
basic feature of image
is
edge
which
is
the mutation of gray level for gray level image. There are many geometric forms of mutation of gray
level
,
here we only consider the common and important one - step-structure edges of binary image.
The
document images of
printed matter such
as
newspapers, magazines , books etc. , generally are binary images with step-structure edges.
It
is very important
t
。但口
Y
out edge detection and extract
the
feature of image edge to pattern
re
∞
gnition
of binary
lmage.
The
classical edge detection method
is
to construct edge detect operator for original digital image according to
some neighborhood of pixel
, such as Roberts operator,
La
placian operator,
So
bel operator, Kirsch operator, Prewitt
operator
, Rosenfeld operator etc. Despite the various forms of these edge operators, they have one character in
common
that
is they are all constructed by several regular convolution kernels. However,
the
re
a1
image edge usually
has very complex geometric structure. In this case
,
the
algorithm using regt.ilar
∞
nvolution
kernel edge operators can
not be adaptively adjusted which has a lot of limitations. We shall carry out stochastic filtering for edge detection
, and
use order statistic filter to construct a new kind of edge operator.
The
essencial difference between this edge
d
巳
tect
operator and the classical edge operators is
that
the new operator is constructed of stochastic convolution keme
l.
For
the edge of binary image
with
complex geometric structure, the new stochatic edge operator can adaptively adjust the
algorithm which shows very good results of edge detection. Since here the stochastic convolution kernel
in
fact
is
an
order statistic filter
, we label the stochastic edge operator
as
OSF
edge operator. First we shall introduce order
Received
date:
2000.08.09
Foundation
tenn:
the
National
Natural
~元
ience
Foundation
of
China
(1
9971063)