Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

2010 Third International Conference on Intelligent Networks and Intelligent Systems

The Research on the Methods of Image Edge Detection


Xiuli Zhang Wei Liu
Infoemation science and Engineering Institute Infoemation science and Engineering Institute
Shenyang Ligong University Shenyang Ligong University
Shengyang,China Shengyang,China
zhangqisong_840124@163.com zhangqisong_840124@163.com
Abstract: The methods of image edge detection based on the GX and GY can form a gradient operator by using two
differential, wavelet transform and image morphology are templates. In the digital image, we use difference instead of
introduced in the paper. Differential operator edge detection differential to calculate, adopting convolution method in
method is based on the edge of the adjacent regions' boundary that operator operation processing. Move template in the image
is not continuous with the gray, using derivative to detect the edge; and calculate the corresponding center pixel’s gradient in
Wavelet transform edge detection is by means of selecting the each location. Roberts operator is one of the most simple
appropriate wavelet function to make wavelet detail component to grads operator ,which is a 2×2 template, as shown in figure
truely reflect the image of the local gray mutations point; The 1.It adopts the difference between two adjacent pixels in
basic idea of mathematical morphology is to use some forms of diagonal direction. That is,
structural elements to measure and extract the corresponding
∇f = ( f ( x, y ) − f ( x − 1, y ), f ( x, y ) − f ( x, y − 1)) .
shape of the image to attain the objective of image analysis and
recognition. In addition, there are Sobel operator, Prewitt operator and so
Key words: edge detection; differential operator; wavelet on, which are 3×3 templates and use pixel neighboring
transform; morphology region to calculate grads. The template of Sobel operator is
shown in figure 2. The template of Prewitt operator is
INTRODUCTION
similar to Sobel operator’s.
The edge detection of digital image is the very important
basic of image segmentation, target area identification, 1.2LAPLACE OPERATOR
region shape extraction and other image analysis field, The characteristic that the second derivative of gray
which is also an important attribute of extracting image image’s knee point position is 0, which is directionless.
feature. The methods of image edge detection has a long What we need is only a template.The form of operator is
history . The methods of image edge detection based on the
∂2 f ∂2 f
differential, wavelet transform and morphology are ∇2 f = + .Differential .form. is .
∂x 2 ∂y 2
introduced, and whose advantages and disadvantages are
analyzed in the paper.
∇2 f =[ f (x+1, y) + f (x−1, y) + f (x, y+1) + f (x, y−1) −4f (x, y)]
1 DIFFERENTIAL OPERATOR EDGE DETECTION
I Obtaining common operator ∇ 2 is expressed with
1.1GRADS OPERATOR
Because edge in the image has more dramatic changes in template , it’s shown as figure 3.

gray area, the boundary between adjacent regions of edge


1 0 0 1
detection is not continuous with the gray, we can detect the
0 -1 -1 0
edge with derivative. The most common is grads operator[1
,2,3], which is the first derivative. For an image function Figure.1 roberts operator
f (x, y), there are the grads with x and y directions in (x, y)
-1 -2 -1 -1 0 1 0 1 0
∂f 2 ∂f 2 0 0 0 -2 0 2 1 -4 1
points, whose grads amplitude is G = ( ) + ( ) .
∂x ∂y 1 2 1 -1 0 1 0 1 0

978-0-7695-4249-2/10 $26.00 © 2010 IEEE 100


DOI 10.1109/ICINIS.2010.70
Laplace operator can highlight the angular line and and multi-scale analysis is more suitable for image
the isolated point in the image , but it is more sensitive to multi-scale edge detecting[7]. Smoothing signals with
Figure.2 sobel operator Figure.3 Laplace operator
noise. Marr proposed an improved method: the image is different scaling function, we can detect dramatic changing
firstly got appropriate smoothed to suppress noise and then points from their first derivative or second derivative.
derivated, which use smoothing filter to smooth, and finally Impulse response of a low-pass filter should be smooth
laplace operator is used. That will be the function. The function f(x) and a convolution of smooth
Gauss-Laplace(LOG)filter. function weaken its high frequency component, without
changing the low frequency components. So f(x) can get
1.3 CANNY OPERATOR
smoothed. Binary wavelet edge detection is one of the
Canny converted the problems of edge detection to multi-scale edge detections. Define two waves, which are
detect maximum value of the function units[4]. In 1986,
canny proposed edge detection operator based on partial derivatives ψ (1) ( x, y ) 、 ψ ( 2 ) ( x, y ) of a
optimization algorithm and provided three indexes of
evaluating advantages and disadvantages of the edge two-dimensional continuous function θ ( x, y ) along the

detection performance, that is: (1)Good noise-signal ratio, x and y direction separately. Therefore, there are:
that is, the probability of non-edge points being sentenced to
∂θ ( x, y ) ∂θ ( x, y )
edge points or edge points being sentenced to non-edge ψ (1) ( x, y ) = ψ ( 2 ) ( x, y ) =
points is low;(2) Good localization performance, that is, ∂x ∂y 。
detected edge points should be to the actual edge center as The two components of wavelet transform in scale for S are:
far as possible;(3) Unique response to a single edge, and
make the best suppression to the response of false edges. In wS(1) [ f ( x, y)] = f *ψ S(1) ( x, y)
essence, candy edge detection operator is the first differential
operator with smooth function. It has a very good wS(2) [ f (x, y)] = f *ψ S(2) (x, y)
noise-signal ratio and detection accuracy. The basic idea is: For the binary wavelet (S = 2i), there are:
firstly, some Gauss filter is selected to smooth filtrate;
secondly, non-maxima suppression technology is used to
⎧⎪ w 2(1i ) [ f ( x , y )] ⎫⎪ ⎢ f *ψ (1 )
2i
( x, y) ⎥
⎨ (2) ⎬ = s⎢ ⎥
⎢⎣ f * ψ
(2)
process the smoothed image.,and then finally edge image ⎪⎩ w 2 i [ f ( x , y )] ⎪⎭ 2i
( x , y ) ⎥⎦
can be obtained. →
= s ∇ ( f * θ )( x , y )
1.4JUN SHEN OPERATOR
Calculating the modulus and amplitude angle of wavelet
Jun Shen[5] operator is similar to Canny operator and
transform in the scale2i, respectively is:
LOG operator, which is also firstly to smooth and then
2 2
derivate. The difference is that what is selected and used to M 2 i f ( x, y ) = w2(1i ) f ( x, y ) + w2( 2i ) f ( x, y )
smooth is the symmetrical exponential filter.
A2i f ( x, y ) = arg(wS( 2) [ f ( x, y )] / wS(1) [ f ( x, y )])
2 WAVELET TRANSFORM EDGE DETECTION
Wavelet transform provides a new means for the image The modulus of wavelet transform M 2 f ( x, y ) is
edge analysis, by selecting the appropriate wavelet function,
which can make the detail component decomposed of proportional to the modulus of grads
G
wavelet truly reflect the local gray mutations of the vector ∇( f ∗ θ )( x, y ) . The amplitude angle is the angle
image[6]. Wavelet edge detection theory provides a
multi-scale approach for image edge detection, so the between grads vector and the horizontal direction, and is

wavelet transform and analysis as a kind of multi-channel also the direction of the edge. Therefore, to detect edges
,only need to find the maximum grads vector of the modulus
point.

101
line
row CAj+1
Corrosion type MEO2= f ( n ) − f ( n ) ΘB
Fj Low
Fj
row
CD(h)j+1
Dilate-erode MEO3= f (n) ⊕ B − f (n)ΘB
CA Gj
row
CD(h)j+1
line
Fj Opened operation type MEO4= f − f DB
Gj row
(h)
CD
f •B− f DB
j+1
Gj Closed operation type MEO6=

It can be seen that: the above morphological edge


Figure.4The static 2d discrete wavelet transform algorithm
detection operator is a nonlinear differential operator, which
3MATHEMATICAL MORPHOLOGY EDGE DETECTION is also essentially the promotion of a traditional linear
differential operator in a certain sense. Intumescent
Mathematical morphology is a new method in the field
extracted is outer edge of image, Inner edge of image is
of image processing and model identification. The basic idea
corrosion type, and Expansion of corrosion type is extracted
is to use structural elements with a certain form to measure
image in the actual riding the edge of the Euclidean
and extract the corresponding shape of the image to attain
boundary. These three grads operators are sensitive to noise
the objective of image analysis and recognition[8]. There are
and are not good at anti-noise performance. Opening
a variety of grads in the process of edge detection, whose
operation of the peak-type image detection; closing
basic principles based on the following considerations: If the
operation of the trough-type image detection; opening and
grads value of a point in the image is large, it shows that the
closing operation type can be a test image in the peaks and
lightness and darkness of the point change rapidly in the
valleys, but a larger error. These three morphological grads
image, which may have the edge passing. Thus, several
are all better in anti-noise performance than the first three
morphological grads was proposed in morphological image
,but have deviation. According to the above different
processing. According to the definition of morphological
characteristics, we can construct different forms of edge
dilation, erosion, opening and closing operation and
detection operators. Meanwhile, considering analytical
morphological distensibility, the following formulas are
characteristics of multi-structuring elements ,multi-Scale
established f (n) is image, B is structuring element). and omnidirectional characteristics, we can construct good
edge detection algorithm, such as morphological edge
When B containing the coordinate origin, there is:
detection algorithm[9], morphological edge detector with
f ⊕ B ≥ f ≥ fΘB f •B≥ f ≥ f DB multiscale adaptive weighting[10], omnidirectional
morphological edge detection algorithm of alternate
Among these, ⊕ 、 Θ 、 • 、 D are respectively dilation, filter[11] and morphological edge detection method [12]
based on multi-structuring elements and multi-scale and so
erosion, open and close morphological operators. (The
on.
following is the same)
Considering above nature and the various operations, we can
get the following inequality:

f ⊕ B ≥ f • B ≥ f ≥ f D B ≥ fΘB

The basic morphological grads is structured, as following:


Leaps edge The roof edge
Intumescent MEO1= f ( n ) ⊕ B − f ( n)

102
Lugs edge 14(7):710-732

Figure.5 Mathematical.morphology.edge.detection [8] Zhang Yujin. Image Engineering - Image Processing and Analysis [M].

classification Beijing:Tsinghua University


Press.1999
4 CONCLUSION
[9] Yang Shubin, Peng Fuyuan. Application of Morphological Edge
The differential method, wavelet transform method and Detectors in Image Corrupted by Noise [J ]. Computer Engineering
method based on morphology and other detection methods and Applications, 2002, 38 (17) : 91-92
are researched in the field of edge detection. However, none [10] Yang Shubin , Peng Fuyuan, Zhang Zengchang. Morphological edge
of them is a method with absolute advantage. The traditional detector with multiscale adaptive weighting[J ]. Journal of Huazhong
grads operator algorithm is simple, but the detected edge University of Science and Technology, 2002, 30 (10) :42-45
is incomplete ,poor continuity ,breakpoints and the detected [11] Yang Shubin. The Application Research of mathematical morphology
edge line is thick. These detection methods are much in Image Processing [D ]. Wuhan: Journal of Huazhong University of
influenced by the noise. LOG operator is to use second-order Science and Technology, 2002.
differential operators to detect, which can not only detect [12] HU Yuanyuan,CAI Guang-cheng. Morphological Edge Detection
more edge, but also have higher positioning accuracy. Based on Multi-Structuring Elements and Multi-Scale [J]. Computer
However, we have seen some false edges produced and it is Technology and Development.2008,18(11):97-99
also more affected by noise. Canny operator and Jun Shen
operator have better positioning accuracy and the marginal
seal. But no matter which algorithms are a little sensitive to
noise, so we propose to filter the noise before extracting the
edge of being collected images. Wavelet transform edge
detection can effectively suppress noise and provide a higher
accuracy of edge location. Morphological methods are
simple algorithm and can keep the minutiae of image better
in image edge detecting. The selection of structuring element
is flexibility, but the different forms of structuring element
must be selected for different images and the purpose of test,
so the adaptability of Algorithm is poor.

REFERENCES
[1] Gonzalez. Digital Image Processing(The Second Edition)[M]. Beijing
:Electronic Industry Press.2008
[2] Luo Junhui. The Application of MATLAB7.0 in Image Processing [M].
Beijing:Machinery Industry Press.2006
[3] P ratt W R. D igital image p rocessing [M ]. N EWYORK: W iley,
1977.
[4] Canny J.A computational approach to edge detection[J] .
IEEE-PAMI,1986, 8:679-698
[5] Shen Jun,Castan S. An optimal linear operator for step edge detection.
GVGIP-GMIP.1992, 54:112-133
[6] Ding Yan, Liu Liudi, Guo Hong. The Application of Wavelet
Transform to Edge Detection Image [J]. Journal of Beijing Institute of
Technology. 1998,18(5):621-624
[7] Mallat S, Zhong S. Characterization of signals from multiscale
edges.IEEE-PAMI.1992,

103

You might also like