EDGE DETECTION IN RADAR IMAGES USING WEIBULL DISTRIBUTION

Sundarapandian et al. (Eds) : CSE, CICS, DBDM, AIFL, SCOM - 2013
pp. 85–93, 2013. © CS & IT-CSCP 2013 DOI : 10.5121/csit.2013.3309
EDGE DETECTION IN RADAR IMAGES USING
WEIBULL DISTRIBUTION
Ali El-Zaart , Wafaa Kamel Al-Jibory
Department of Mathematics and Computer Science
Faculty of Science, Beirut Arab University
Beirut –Lebanon
dr_elzaart@yahoo.com, waf_jibory@yahoo.com
ABSTRACT
Radar images can reveal information about the shape of the surface terrain as well as its
physical and biophysical properties. Radar images have long been used in geological studies to
map structural features that are revealed by the shape of the landscape. Radar imagery also has
applications in vegetation and crop type mapping, landscape ecology, hydrology, and
volcanology. Image processing is using for detecting for objects in radar images. Edge
detection; which is a method of determining the discontinuities in gray level images; is a very
important initial step in Image processing. Many classical edge detectors have been developed
over time. Some of the well-known edge detection operators based on the first derivative of the
image are Roberts, Prewitt, Sobel which is traditionally implemented by convolving the image
with masks. Also Gaussian distribution has been used to build masks for the first and second
derivative. However, this distribution has limit to only symmetric shape. This paper will use to
construct the masks, the Weibull distribution which was more general than Gaussian because it
has symmetric and asymmetric shape. The constructed masks are applied to images and we
obtained good results.
KEYWORDS
Edge detection; Image processing; Weibull Distribution; Gradient; Radar Image.
1. INTRODUCTION
Unlike most remote sensing systems, which rely on the sun as an energy source, imaging radar
systems are active sensors that “illuminate” surface features with broadcast microwave energy
and record a returned signal. Most imaging radar systems produce microwaves with wavelengths
between 1 cm and 1 meter, longer than the wavelengths used in weather radar to detect rain and
snow. As a result imaging radar systems can be used to map surface features day or night and in
almost any weather conditions [2]. An edge is usually a step change in intensity in an image
(radar image). It corresponds to the boundary between two regions or a set of points in the image
where luminous intensity changes very sharply[10]. The decision of whether pixel is an edge
point or not bases on how much its local neighbours respond to a certain edge detector [13]. Over
the years, many methods have been proposed for detecting edges in images. Some of the earlier
methods, such as the Sobel and Prewitt detectors [11], used local gradient operators [12] to obtain
spatial filter masks. The procedure is to compute the sum of products of the mask coefficients
with the intensity values in the region encompassed by the mask [3]. Also the Canny edge
detector which depends on the Gaussian distribution for obtaining the operators for the gradient

86 Computer Science & Information Technology (CS & IT)
and Laplacian masks is a well-known edge detector [9]. In this paper we propose method that will
use Weibull Distribution instead of Gaussian distribution to obtaining edge detection operators.
The advantage of this method is that Gaussian distribution has limitation to only symmetric shape
but Weibull Distribution has symmetric and asymmetric shape.
The rest of this paper is organized as follows. Section 2 introduces Gradient Edge Detection.
Section 3 explains the Most famous Edge Detector. Gradient of Weibull Edge Detector is
displayed in Section 4. Experimental result is shown in Section 5. Finally this paper presents
conclusion and future work in Section 6.
2. GRADIENT EDGE DETECTION
Gradient is a vector which has certain magnitude and direction. In image processing, the gradient
is the change in gray level with direction. This can be calculated by taking the difference in value
of neighbouring pixels
Where is first order derivative of ),( yxf define as:
(1)
The magnitude of this vector, denoted magn(f), Where
(2)
The direction of the gradient vector, denoted dir(f), Where
)xGyG(
1
tanf)dir(
−
=∇ (3)
The magnitude of gradient provides information about the strength of the edge and the direction
of gradient is always perpendicular to the direction of the edge.
3. THE FAMOUSE EDGE DETECTOR
Before introducing the proposed algorithm, this section reviews some of the main edge detection
methods, such as the Sobel method and Gradient of Gaussian edge detector.
3.1. The Sobel Edge Detector
The Sobel method [3] utilizes two masks, Sx and Sy , shown in fig. 1, to do convolution on the
gray image and then obtain the edge intensities Gx and Gy in the vertical and horizontal directions,
respectively. The edge intensity of the mask center is defined as |Gx|+|Gy |. If the edge intensity of
each pixel is larger than an appropriate threshold T, then the pixel is regarded as an edge point.
Unfortunately, the edge line detected by Sobel method is usually thicker than the actual edge [16].
2
y
y)f(x,
2
2
x
y)f(x,
2
y)f(x,
2
∂
∂
+
∂
∂
=∇
f∇
( )2
yG
2
xG
2
y
f
2
x
f
f)magn( +=
∂
∂
+
∂
∂
=∇





















Computer Science & Information
Figure 1.
3.2. The Gradient of Gaussian edge detector
An edge detection operator can reduce noise by smoothing the image, but this
the location of the edge: or the operator can have greater sensitivity to the presence of edges, but
this will increase the sensitivity of the operator to noise.
the best compromise between noise imm
Gaussian filtering is the first derivative of a Gaussian. This operator corresponds to smoothing an
image with a Gaussian function and then computing the gradient. The gradient can be numerically
approximated by using the standard finite
in the x and y directions. The operator that is the combination of a Gaussian smoothing filter and
a gradient approximation is not rotationally symmetric. The
and antisymmetric perpendicular to the edge (along the line of the gradient).
operator is sensitive to the edge in the direction of steepest change, but is insensitive to the edge
and acts as a smoothing operator in the direction along the line
4. GRADIENT OF WEIBULL
The Gaussian distribution is the most popularly used distribution model in the field of pattern
recognition. It is used to build masks for the first and second derivative. However, it has limit to
only symmetric shape. We will propose new method that uses
general than Gaussian because it has symmetric and asymmetric shape.
In this section we will explain the characteristic of 1D
Weibull distribution from 1D. The
is given by:
The distribution is skewed with a longer tail to the right of the mean as show
µ (mean) is scale parameter and
sameσ , the function skewness increases as
very sharply in the beginning when the value of
Figure 3. Probability density function of the
)(1 xDW
),1DW(x; βα




=
0
αβ
Computer Science & Information Technology (CS & IT)
. Two convolution masks in Sobel method.
The Gradient of Gaussian edge detector
An edge detection operator can reduce noise by smoothing the image, but this adds uncertainly to
this will increase the sensitivity of the operator to noise. The type of liner operator that provides
the best compromise between noise immunity and localization, while retaining the advantages of
is the first derivative of a Gaussian. This operator corresponds to smoothing an
proximated by using the standard finite-difference approximation for the first partial derivative
y directions. The operator that is the combination of a Gaussian smoothing filter and
a gradient approximation is not rotationally symmetric. The operator is symmetric along the edge
and antisymmetric perpendicular to the edge (along the line of the gradient). This means
thing operator in the direction along the line
EIBULL EDGE DETECTOR
only symmetric shape. We will propose new method that uses Weibull Distribution which is more
general than Gaussian because it has symmetric and asymmetric shape.
ll explain the characteristic of 1D Weibull distribution and how calculate 2D
Weibull distribution from 1D. The 1D Weibull distributions have the probability density function
The distribution is skewed with a longer tail to the right of the mean as shown in Fig
(mean) is scale parameter and σ (standard deviation) shape parameters and x
, the function skewness increases as µ increases and for the same µ , the function is rises
very sharply in the beginning when the value of σ significantly greater than 1[8].
Probability density function of the Weibull distribution
Sy
-1 0 1
-2 0 2
-1 0 1
-1 -2 -1
0 0 0
1 2 1
−−
−−
)1(
2 β
αββ
βαβ
αβ x
x
ex 0x >
elsewhere
Sx
87
uncertainly to
of liner operator that provides
unity and localization, while retaining the advantages of
is the first derivative of a Gaussian. This operator corresponds to smoothing an
difference approximation for the first partial derivative
y directions. The operator that is the combination of a Gaussian smoothing filter and
operator is symmetric along the edge
This means that the
Distribution which is more
distribution and how calculate 2D
probability density function
(4)
in Fig .3. Where
0> . For the
, the function is rises

We can calculate the 2D Weibull distribution by produce of: ),;(1 σµxDW and ),;(1 σµyDW where
f is“Eq. (4)” given by:
= (5)
In the following section we will explain how perform smoothing using 2D Weibull Distribution
then we will explain the applying of edge detection using the gradient of the 2D Weibull
Distribution. After that we will present the edge detection using the Laplacian of 2D Weibull
Distribution.
4.1. Smoothing Using 2D Weibull Distribution
We can construct smoothing filter from 2D Weibull Distribution. The method is to create general
3x3 mask from 2D Weibull Distribution as show bellow:
For 5x5 mask the left corner would be )2,2(2 dydxDW −− and for 7x7 would be )3,3(2 dydxDW −−
also for 9x9 )4,4(2 dydxDW −− and so on. The sum of all values of the mask must be 1 because it
is smoothing mask. The value d is incremental and the best value of incremental we can
determine through the experiment.
4.2. Edge Detection using the Gradient of the 2D Weibull Distribution
We can construct the gradient mask of ),(2 yxDW by obtaining the first partial derivative of yx,
for ),(2 yxDW . The first x derivative for ),(2 yxDW is given by:
= (6)
The first y derivative for ),(2 yxDW is given by:
= (7)
Using the first x derivative for ),(2 yxDW we can construct xM mask:
DWMx 2 DWMx 2 DWMx 2
),( incyyincxx −− ),( yincxx − ),( incyyincxx +−
),( incyyx −



 +−−−
=×=
0
)(1122
σ)µ,1DW(y;σ)µ,1DW((x;σ)µ,y;2DW(x,
ββαββ
βα
yx
eyx
Elsewhere
x>0,y=0




−−
∂
∂
=
+−−−
0
)(1222
)1(),(2),(2
β
αββ
ββαββ
βα xyxDW
x
yxDWM
yx
eyx
x
x>0,y=0




−−+−−−
0
)(1222
)1( β
αββ
ββαββ
βα yyx
exy
Elsewhere
x>0,y=0
),(2 dydxDW −−
),(2 ydxDW − ),(2 dydxDW +−
),(2 dyxDW − ),(2 yxDw ),(2 dyxDW +
),(2 dydxDW ++),(2 ydxDW +),(2 dydxDW −+
Elsewhere

Computer Science & Information Technology (CS & IT) 89
Using the first y derivative for ),(2 yxDW we can construct yM mask:
We need to calculate two increments one for x and one for y.
The sum of the gradient mask should be zero. So after constructing the masks, they should be
normalized. The positive values are added and then divided by their sum to obtain 1 and the
negative values are computed in the same way to obtain –1.
The obtained masks at alpha = 1, beta =2 are as follows:
After normalization we got these results:
The obtained masks at alpha = 1, beta =3 are as follows:
),( incyyincxx −− ),( yincxx − ),( incyyincxx +−
DWM y 2 DWM y 2 DWM y 2
),(2 yxDWM y
),( incyyx −
DWM y 2
),( incyyx +
DWM y 2
),( incyyincxx −+ ),( yincxx + ),( incyyincxx ++
DWM y 2 DWM y 2 DWM y 2
=xM =yM
),(2 yxDWM x
DWMx 2 DWMx 2
DWMx 2 DWMx 2 DWMx 2
),( incyyx +
),( incyyincxx −+ ),( yincxx + ),( incyyincxx ++
0.6951 0
-
0.3538
1.5025 0
-
0.7648
0.9850 0
-
0.5014
0.6951 1.5025 0.9850
0 0 0
-0.3538
-
0.7648
-
0.5014
0.2184 0.4721 0.3095
0 0 0
-0.2184 -0.4721 -0.3095
0.2184 0 - 0.2184
0.4721 0 - 0.4721
0.3095 0 -0.3095
=xM =yM
=xM
0.1550 1.2799 0.9149
0.1785 1.4738 1.0535
-0.2606
-
2.1526
-
1.5388
=yM
0.1550 0.1785
-
0.2606
1.2799 1.4738
-
2.1526
0.9149 1.0535
-
1.5388

After normalization we got these results:
5. EXPERIMENTAL RESULTS
We present in this section our experimental results of using Weibull Distribution in detecting
edges using Gradient of this distribution and compare this result with Sobel.
(a)
(b)
Figure 4. Original Images used in edge detection
(a)
(b)
0.0307 0.2532 0.1810
0.0353 0.2915 0.2084
-0.0660 -0.5447 -0.3894
0.0307 0.0353 -0.0660
0.2532 0.2915 -0.5447
0.1810 0.2084 -0.3894
=xM =yM
Figure 5. Sobel Results

Computer Science & Information
(a)
Figure 6
Computer Science & Information Technology (CS & IT)
6. Results of the new Gradient detector of
weibull of size 3x3
Alpha=1,Beta=2
Alpha=1,Beta=3
91

From Fig.5 and Fig.6 we can notice that the Weibull distribution is better than the gradient of
Gaussian method it produces thinner edges and is less sensitive to noise. This is because the
Gaussian distribution has limit to only symmetric shape but Weibull distribution has symmetric
and asymmetric shape.
6. CONCLUSION AND FUTURE WORK
Smoothing is a prior step in any edge detection process to suppress as much noise as possible.
Edge detection using first derivative depend on Gaussian works well when the image contains
sharp intensity transitions and low noise, while Edge detection using LOG make better
localization, especially when the edges are not very sharp. We proposed new method that uses
Weibull Distribution instead of Gaussian distribution to build masks for the first and second
derivative and for smoothing image.
ACKNOWLEDGMENT
We thank the Dr. Toufic El Arwadi for his help on the mathematical calculation
REFERENCES
[1] J.K.Kim, J.M.Park, K.S.Song and H.W.Park ” Adaptive Mammographic Image Enhancement Using
First Derivative and Local Statistics” IEEE Transactions on medical imaging, Vol.16,No.5,October
1997.
[2] Randall B. Smith."Interpreting Digital Radar Images". MicroImages, January 2012.
[3] R.Gonzalez and R.Woods, “Digital image processing,” 3rd Edition, Prentice Hall, New York, 2008,
pp. 695.
[4] E.Trucco and A.Verri.” Introductory Techniques for 3-D Computer Vision” Prentice Hall, New York
,1998. Chapter 4.2.
[5] C. Kanga,Wen-JuneWang, “Anovel edge detection method based on the maximizing objective
function,” Taiwan, April 2007.
[6] J.Canny. “A Computational Approach to Edge Detection Edge Detection”.IEEE Transactions on
Pattern Analysis and machine intelligence, Vol. PAMI-8, NO. 6, November 1986.
[7] N.T. Thomopoulos, Arvid C. Johnson.” Tables And Characteristics of the Standardized Lognormal
Distribution”, Proceedings of the Decision Sciences Institute, 2003, pp. 1031-1036.
[8] Relia Soft.” Characteristics of the Lognormal Distribution”.2006.
[9] E. Al-Owaisheque, A. Al-Owaisheq and A. El-Zaart, “A New Edge Detector Using 2D Beta
Distribution". Proceedings of the 3rd IEEE International Conference on Information &
Communication Technologies : from Theory to Application. April, 9-11, 2008, Syria.
[10] N.Nain, G. Jindal, A. Garg and A. Jain. “Dynamic Thresholding Based Edge Detection”. Proceedings
of the World Congress on Engineering 2008 Vol I. WCE 2008, July 2 - 4, 2008, London, U.K.
[11] R. J. Qian and T. S. Huang, “Optimal edge detection in two-dimensional images,” Proc. Image
Understanding Workshop, 1994, pp. 1581–1588.
[12] M. Basu, “Gaussian-Based Edge-Detection Methods A Survey,” IEEE Transactions on systems, man,
and cybernetics part C: applications and reviews, vol. 32, no. 3, August 2002.
[13] R. Gurcan, Isin Erer and Sedef Kent, “An Edge Detection Method Using 2-D Autoregressive Lattice
Prediction Filters for Remotely Sensed Images,” Istanbul Technical University Maslak, stanbul,
Turkey 2004.
[14] H.Chidiac et Djemel Ziou, “Classification of Image Edges,” Universit´e de Sherbrooke, Sherbrooke
(Qc), Canada, J1K 2R1.
[15] B. G. Schunck, “Edge detection with Gaussian filters at multiple scales,” in Proc. IEEE Comp. Soc.
Work. Comp. Vis., 1987.
[16] L.R. Liang, C.G. Looney, Competitive fuzzy edge detection, Appl. Soft Comput. J. 3 (2003) 123–137.

Computer Science & Information Technology (CS & IT) 93
Authors
Ali El-Zaart was a senior software developer at Department of Research and Development,
Semiconductor Insight, Ottawa, Canada during 2000-2001. From 2001 to 2004, he was an
assistant professor at the Department of Biomedical Technology, College of Applied
Medical Sciences, King Saud University. From 2004-2010 he was an assistant professor at
the Department of Computer Science, College of computer and information Sciences, King
Saud University. In 2010, he promoted to associate professor at the same department.
Currently, his is an associate professor at the department of Mathematics and Computer
Science, Faculty of Sciences; Beirut Arab University. He has published numerous articles and proceedings
in the areas of image processing, remote sensing, and computer vision. He received a B.Sc. in computer
science from the Lebanese University; Beirut, Lebanon in 1990, M.Sc. degree in computer science from the
University of Sherbrooke, Sherbrooke, Canada in 1996, and Ph.D. degree in computer science from the
University of Sherbrooke, Sherbrooke, Canada in 2001. His research interests include image processing,
pattern recognition, remote sensing, and computer vision.
Wafaa Kamil S. Al-jibory: Currently he is a master student in Beirut Arab University,
Department of Mathematics and Computer Science, Beirut, Lebanon (BAU).

EDGE DETECTION IN RADAR IMAGES USING WEIBULL DISTRIBUTION

Recommended

More Related Content

What's hot (19)

Similar to EDGE DETECTION IN RADAR IMAGES USING WEIBULL DISTRIBUTION (20)

More from csitconf (9)

Recently uploaded (20)

EDGE DETECTION IN RADAR IMAGES USING WEIBULL DISTRIBUTION