Int. J. Communications, Network and System Sciences, 2012, 5, 624-630
http://dx.doi.org/10.4236/ijcns.2012.529072 Published Online September 2012 (http://www.SciRP.org/journal/ijcns)
A Hybrid Method for Edge Continuity Based on Pixel
Neighbors Pattern Analysis (PNPA) for Remote
Sensing Satellite Images
Waqas Haider, Muhammad Sharif Malik, Mudassar Raza, Abdul Wahab,
Izhar Ahmed Khan, Umar Zia, Jawad Tanveer, Hadia Bashir
Department of Computer Science, COMSATS Institute of Information Technology, Wah Cantt and Vehari Campus,
Islamabad, Pakistan
[e-mail], [e-mail], [e-mail], [e-mail]
Received June 25, 2012; revised July 26, 2012; accepted August 7, 2012
ABSTRACT
Edge enhancement is derived from lack of accurate result from edge detection techniques. The image which is captured
from long distances carries a lot of noise and blur which causes edge discontinuity. Although some novel algorithms
which are based on cellular neural network, fuzzy enhancement and binary morphology have shown accuracy in order
to obtain refined edge but still the problem of edge discontinuity arises. Eliminating discontinuity of edge a hybrid
technique is p roposed b ased on p ixel neighbors p attern analys is PNPA. In the technique Cann y operator for initial edge
detection, PNPA operation for edge enhancement are performed for remote sensing satellite image successively. The
visual and subjective evaluation shows that the proposed PNPA operation can effectively eliminate the influence of
edge discontinuity which occurred due to noise and blurr in original captured image, as comparing to existing edge
segmenting pr ocesse s .
Keywords: Ed ge Det ection; Edge Enha ncement
1. Introduction
Segmenting regions in an image applies frequently the
operations for determining edges of regions. In many
image processing applications, preserving edges of im-
ages is important. Edge is the set of pixels which de-
scribes the boundary of any region of the image. As in-
side an image there may be different regions or multiple
structures so edge describes each structure’s boundary.
The image which is captured from long distance carries
noise which causes burred image which ultimately in-
duced discontinuity in edge of that image [1,2] when
edge is detected with traditional edge detection operators.
Several contributors demonstrated there work in the area
of edge detection of remote sensing images which are
given in [3-10]. In remote sensing satellite image there
are multiple structures in it like roads, fields, rivers, for-
est buildings and residential areas. Accurate boundary
segmentation of such parts in an image totally based on
edge detection and edge enhancement. Edge enhance-
ment is a method of image segmentation in which the
detected edge should be continuous. But it is an issue
which has not been resolved completely so far because in
such images which are captured from long distances, the
illumination and noise factors are high which causes
edge discontinuity.
In this research, disco ntinuity of edge lines are tried to
minimize with proposed hybrid edge enhancement tech-
nique PNPA, which is expressed briefly in Section 4 of
the paper. The rest of this paper is organized as follows.
Section 2 reviews the existing work in the domain of edge
enhancement and edge continuity. Section 3 reviews the
working and merits/demerits of edge detection operators.
Section 4 describes the proposed method (e.g. PNPA) for
edge continuity. In Section 5, the comparative evaluation
of visual and subjective results are given and finally in
Section 6, the discussion and conclusion are given.
2. Existing Work in the Domain of Edge
Enhancement
Acquiring edge continuity has been addressed by differ-
ent new novel techniques which are based on cellular
neural network [2] and binary morphology [3] proved
better edge quality but the operators which are based on
mathematical morphology are not effective for images
which have too many ramp edges [11]. In [12] the suit-
able strategies are expressed for edge thickness and dis-
C
opyright © 2012 SciRes. IJCNS

W. HAIDER ET AL. 625
continuities inside the detected edges. Enhancing the
edge by removing the limitations in different edge detec-
tion operators a hybrid approach is presented in [13].
Smoothing edges with integrated functions in [14] an
approach is presented which utilizes the bee colony op-
timization function. Also some novel approaches for ad-
dressing edge enhancement are expressed in [15-17].
After analyzing several techniques for edge enhancement
there is still discontinuity in edges.
3. Review of Edge Detection Operators
Working with Merits and Demerits
In order to obtain edges of any image the traditional ap-
proach is shown in Figure 1.
When image is captured and during the process of
transmission, formation, reception and processing the
noise is added with it. The purpo se of pr e processing is to
remove noises from the image. When any edge detection
operator is applied, the preprocessing techniques for
de-noising are built-in. The most reliable technique for
image de-noising is Wavelet analysis [18]. In wavelet
analysis during multi scale analysis, image is decom-
posed (see Figure 2) to multi scale frequency compo-
nents, then for removing noisy data th e band passed filter
based functions starts working. Finally enhanced image
is obtained using wavelet inverse transform. In wavelet
analysis the quality of image is improved.
In the process of finding threshold value (see Figure 1)
by applying derivative operator it got grey change value,
then by using threshold, edge pixel set is selected [19].
Suppose an image is

,
f
xy after de-noising process,
its derivative operator is
f
x ,
f
y which repre-
sent grey change in two dimensions and it is defined as:
ddd
f
f
f
xy
x
y



(1)
The Equation (1) gives differential grey change value
in the direction arctan
f
f
yx





which actually gives
Figure 1. Traditional edge detection operator operation.
Figure 2. Wavelet analysis for image de-noising sketch.
the threshold value. The distance function G is applied
on two dimensional differential intensity components and
defined as:
 
22
,GFxyfx fy
 (2)
where

22
f
xfy 
 
is the maximum direc-
tional derivative. In many traditional operators to out-
stand grey level change gradient operator is used [20]
which is defined as:
22
,,,
L
k
GFKLFKL FKL  
 (3)
where



 
,, 1,
k
F
KLFKL FKL (4)



 
,,,1
LFKLFKLFKL

 (5)
In Equations (3) and (4) K = x location point value and
L = y location point value. So Equation (3) obtain thresh-
old intensity value in x direction or in k direction by cal-
culating difference between differential grey value at


,


1,
K
L and
K
L. Similarly Equation (5) gives
threshold intensity value in y or L direction by calculat-
ing difference. Finally Equation (3) process and obtain
gradient G to construct edge pixel set by setting threshold
value.
Gradient Operators
Roberts, prewitt, Sobel operators are based on differen-
tial operator processing for finding edge pixel set [21]. In
Roberts operator for extracting edge pixel set from input
image use partial difference operator. So Equation (3) is
designed for Robert operator as:

W. HAIDER ET AL.
626
   
22
1,,1FKL 

,1,1,GFKLFKLFKLFK Y  
  (6)
In Equation (6) Roberts operator consider 2 by 2 tem-
plate (e.g. 2 neighbor weight) to compute difference op-
erator so the edge extracted is quite thick, not accurate
and suitable for steep low noisy images. Also it is not
suitable for image having to many segments and blurr.
The Prewitt and Sobel operators are the same in defi-
nition and operation as Robert but the difference between
both is Prewitt consider 3 by 3 neighbors weight to
compute difference operator and Sobel uses 4 neighbor’s
weight in computing difference operator. Suppose in the
process of finding threshold value for edge pixel set , in
supposition 3 × 3 matrix given below, the point


,
f
ij
where f grey is grey level change function and


,ij

012
34
678
,
pp
pfijp
ppp

determines that point pixel which is in process.
p
Now the Prewitt and Sobel is defined as:

22
K L,GFij 
 (7)
where in case of Prewitt it utilizes less neighbors and the
detected edge by the Sobel is wider as it utilizes four
neighbors. These classical operators (sobel and prewitt)
could detect inaccurate edges incase of noise. Another
type is Log operator which is defined as:
22222
f
fk fL  (8)
In log operator the differential coefficient is calculated
for the image and the edge detected by the log operator
has double pixel boundary. The log operator is very sen-
sitive to noise and also it cannot find the orientation of
edge due to Laplacian filter. Log operator is used for
judging whether the edge pixel is in dark or bright sec-
tion. Another type of edge detectors are Canny, Shen-
Castan and Boie-Cox which are called Gaussian edge
detectors [22,23] because in these detectors Laplacian of
Gaussian combined Gaussian filtering along with the
Laplacian are included and these are well expressed in
[21]. These edge finders remove noise and smooth the
detected edge but carries complex computing.
After analyzing the various problems of edge features
in which one of them is discontinuity of edge due to
noise and blurr in captured image, it is concluded that
there is still a challenge to detect continuous edge. So in
next section the paper will specifically focus on edge
pixel continuity with proposed method.
4. Edge Continuity with Proposed Method
Unlike the traditional approach for edge detection, the
proposed edge continuity (PNPA) routine is started from
the result of Canny edge detection operator, for simula-
tion the two test images are shown in Figures 3 and 4. In
Figure 5 the edge continuity work flow is expressed.
Although Canny has goo d results but still discontinuity
occurs in edge lines so the resulting image of Canny op-
erator is further processed so that to minimize the gaps at
edge lines. In PNPA operation (see Figure 5) at any
point of binary image


,pxy, the 4 connected and 8
connected neighbors are analyzed for achieving the pat-
tern of neighbor pixels and then a value is imposed to
that chosen pixel (point) according to achieved pattern.
As in binary image zeros shows black (back ground) and
ones shows white (edge pixels). The PNPA operation is
graphically shown in Figure 6. PNPA addresses the
point’s left, right, top, bottom and diagonal neighbor pix-
els to check whether these neighbor pixels are zeros or
one to get pattern of edge line and then giving value to
chosen pixel.
PNPA (Function P)
The hybrid function Pixel neighbor pattern analysis got
the working and decision making logic of two edge en-
hancing filters comparison and selection (CS) and
weighted majority of samples with minimum range
Figure 3. Test image A.
Figure 4. Test image B.

W. HAIDER ET AL. 627
Figure 5. Edge continuity work flow.
Figure 6. PNPA operation on point.
(WMM) [24-26]. The function is defined as:

P
1ii
n

,
x
yxy
i
pp



Pp
where n is the size of the upcoming matrix, ,
x
y is the
new point value which is to be determined according to
neighbor pattern as i
p
x
pp
0, if1,2,1,20
1,if1,2,1,21
X
xx xx
Pxx xx
by Equation (10) in case of
x-direction. Similarly in case of y-direction the i
y is
calculated by Equation (11). In Equation (9) x direction
edge continuity is further given as:







0, if1,2,1,20
1,if1,2,1,2 1
Y
yy yy
Pyy yy
(10)
Similarly for y direction, the continuity is given as:








10 1 1
11,1
00 1 0
11 0 1
Pij
(11)
Expressing the informal definition of Equation (10)
and Equation (11) suppose the upcoming matrix which
has to enhance in the problem domain is as follows:

At point

,Pij function P addresses its left, right,
top, bottom and diagonal neighbor pixels (see Figure 6)
whether they are zeros or one to get pattern of edge line
and then decide what to give value to chosen pixel, both
Equations (10) and (11) works. Also the more accurate
analysis for pattern can be achieved by considering
neighbors at multiple directions. The neighbor pixels
tells PNPA algorithm that what value to be given to cho-
sen pixel whether it remain 0 or converted to 1 to fill up

W. HAIDER ET AL.
628
unnecessary edge gaps.
5. Comparative Evaluation of Visual Results
To analyze the comparative continuity of detected edge
different traditional edge detecting operators and PNPA
are applied on test image A as shown in Figure 7.
In Figure 8 the original image has blur and the arrows
shows the specific part of image which are countered
with PNPA edge enhancement function. Due to the blur
in original image, in Figure 8 the canny operator missed
the desired edges which should be detected. This lack of
canny is enhanced with neighbor pixel pattern analysis
logic PNPA which is shown in Figure 8. Similarly the
test image B is analyzed under different edge detection
operators as shown in Figures 9 and 10 expresses the
enhancement of edge which is achieved by PNPA opera-
tion.
Subjective Evaluation of Edge Detectors and
Proposed Method PNPA
In Table 1 the method of subjective comparative evalua-
tion is taken from [21]. Table 1 shows that comparing to
existing edge detecting operators, the proposed method is
capable of serving images in the domain of edge continu-
ity in highly noisy and blurry conditions. Also the me-
thod keeps the edge features as well as accuracy and ul-
timately no un necessary gaps which occurred due to
blurr and noise in the original image.
6. Conclusion
Each edge detection technique has its own significance
according to the capturing conditions of an image. Some
are better for low noisy images and some are better for
highly noisy images. So noise in an image effects the
Figure 7. Visual comparative analysis of edge operators result with proposed PNPA operation on image A. (a) Original image;
(b) Sobel edge; (c) Robert edge; (d) Log edge; (e) Prewitt edge; (f) Cannv edge; (g) Edge after PNPA.
Figure 8. Result elaboration with image A.

W. HAIDER ET AL. 629
Figure 9. Visual comparative analysis of edge operators result with proposed PNPA operation on image B. (a) Original image
B; (b) LOG edge; (c) Prewitt edge; (d) Sobel edge; (e) Canny edge; (f) Edge after PNPA.
Figure 10. Result elaboration with image B.
Table 1. Subjective comparative ev aluation of edge detectors and proposed method PNPA.
Techniques Best for input image Suitable noise conditionsEdge feature Edge accuracy Edge continuity
Robert operator Less noisy Less noisy Gaps No No
Sobel operator Less noisy Less noisy Gaps No No
Prewitt operator Less noisy Less noisy Gaps No No
Log operator Noisy Noisy Gaps No No
Canny operator High noisy High noisy Little gaps at highly
noisy area Better Better
Canny+ PNPA Highly noisy Highly noisy No gaps Yes Yes
results. In order to minimize noise, before detection ap-
ply wavelet analysis and then processed for edge extrac-
tion. Also the result of any edge detection operator can
be further processed for edge enhancement because noise
is not completely removed which causes unnecessary
edge gaps (discontinuity) and other effects. Hence the
satisfactory results can be achieved if suitable edge de-
tection operator and enhancement technique are applied
according to the problem as edge discontinuity. There-
fore the proposed method could be applied to any image

W. HAIDER ET AL.
630
where edge continuity is required.
REFERENCES
[1] J. B. Wu, Z. P. Yin and Y. L. Xiong, “The Fast Multilevel
Fuzzy Edge Detection of Blurry Images,” IEEE Signal
Processing Letters, Vol. 14, No. 5, 2007, pp. 344-347.
doi:10.1109/LSP.2006.888087
[2] G. Li, F. Yang and L. L. Wang, “An Algorithm for Re-
mote Sensing Image Edge Detection Based on Fuzzy
Sets,” Second International Symposium on Intelligent In-
formation Technology Application, Chongqing, 20-22 De-
cember 2008, pp. 1087-1090.
[3] G. B. Xu, G. Y. Zhao, Y. Lu, Y. X. Yin and Y. L. Shen,
“A CNN Based Edge Detection Algorithym for Remote
Sensing Image,” Chinese Control and Decision Confer-
ence, Zhanjiang, 2-4 July 2008, pp. 2558-2561.
[4] F.-Y. Cui, L.-J. Zou and B. Song, “Edge Feature Extrac-
tion Based on Digital Image Processing Techniques,” In-
ternational Conference on Automation and Logistics,
Qingdao, September 2008, pp. 2320-2324.
[5] Q. Song and L. X. Liu, “Compare between Several Linear
Image Edge Detection Algorithm,” Machine Vision, 2009,
pp. 259-263.
[6] J. L. Wang, C. L. Yang and C. Sun, “A Novel Algorithm
for Edge Detection of Remote Sensing Image Based on
CNN and PSO,” 2nd International Congress on Image
and Signal Processing, Harbin, 17-19 October 2009, pp.
1-5.
[7] N. H. Kaplan, I. Erer and S. Kent, “Edge Detection in
Remote Sensing Images via Lattice Filters Based Sub-
band Decomposition,” 4th International Conference on
Recent Advances in Space Technologies, Istanbul, 11-13
June 2009, pp. 437-440.
[8] M. S. Islam and L. J. Kitchen, “Straight-Edge Extraction
in Distorted Images Using Gradient Correction,” Digital
Image Computing: Techniques and Applications, Park-
ville, 1-3 December 2009, pp. 505-512.
[9] L. X. Xue and Z. C. Wang, “Study on Edge Detection of
Multispectral Remote Sensing Image in Multidimensional
Cloud-Space,” Information Science and Engineering
(ICISE), 2009, pp. 1444-1447.
[10] S. Liu, X. Y. Wang, X. F. Qiu and Y. J. He, “Mathemati -
cal Morphology Edge Detection Algorithm of Remote
Sensing Image with High Resolution,” 1st International
Conference on Information Science and Engineering
(ICISE), Chongqing, 26-28 December 2009, pp. 1323-
1326.
[11] A. B. Mansoor, Z. Khan and A. Khan, “An Application of
Fuzzy Morphology for Enhancement of Aerial Images,”
2nd International Conference on Advances in Space
Technologies, Islamabad, 29-30 November 2008, pp.
143-148.
[12] J. Fombellida, I. Melgar and J. Seijas, “Combining Edge
Detection Algorithms in Intensity SAR Images,” 7th
IEEE International Conference on Industrial Informatics,
23-26 June 2009, pp. 504-509.
[13] K. S. Sim and L. W. Thong, “Enhancement of Optical
Images Using Hybrid Edge Detection Technique,” IEEE
Conference on Innovative Technologies in Intelligent
Systems and Industrial Applications, Ayer Keroh, 25-26
July 2009, pp. 186-191.
[14] T. R. Benala, S. D. Jampala, S. H. Villa and B. Konathala,
“A Novel Approach to Image Edge Enhancement Using
Artificial Bee Colony Optimization Algorithm for Hy-
bridized Smoothening Filters,” IEEE World Congress on
Nature & Biologically Inspired Computing, Visakhapat-
nam, 9-11 December 2009, pp. 1071-1076.
[15] M. T. Alonso, C. López-Martínez, J. J. Mallorquíand and
P. Salembier, “Edge Enhancement Algorithm Based on
the Wavelet Transform for Automatic Edge Detection in
SAR Images,” IEEE Transactions on Geoscience and
Remote Sensing, Vol. 49, No. 1, 2011, pp. 222-235.
[16] T. Chaira, “A Novel Rank Ordered Filter Based Edge
Enhancement of Medical Images Using Intuitionistic
Fuzzy Set Theory,” 2010 International Conference on
Methods and Models in Computer Science, New Delhi,
13-14 December 2010, pp. 18-21.
[17] J. Wu, G. Xu and D. Q. Chen, “One Valid Framework of
Integrating Intensity Discontinuity and Edge Continuity
for Image Straight-Line Extraction,” 2009 1st Interna-
tional Conference on Information Science and Engineer-
ing (ICISE), Guilin, 26-28 December 2009, pp. 1356-
1359.
[18] Z. G. Lai, “Image Processing and Analysis Based on
MATLAB,” Defence Industry Publications, Beijing, 2008.
[19] M. Juneja and P. S. Sandhu, “Performance Evaluation of
Edge Detection Techniques for Images in Spatial Do-
main,” International Journal of Computer Theory and
Engineering, Vol. 1, No. 5, 2009, pp. 614-621.
[20] R. Gonzalez and R. Woods, “Digital Image Processing,”
2nd Edition, Prentice Hall Publisher, Upper Saddle River,
2002.
[21] M. Roushdy, “Comparative Study of Edge Detection
Algorithms Applying on the Grayscale Noisy Image Us-
ing Morphological Filter,” GVIP Journal, Vol. 6, No. 4,
2006.
[22] I. J. Cox, R. A. Boie and D. A. Wallach, “Line Recogni-
tion,” 10th International Conference on Pattern Recogni-
tion, Vol. 1, 1990, pp. 639-645.
[23] S. Castan, J. Zhao and J. Shen, “New Edge Detection
Methods Based on Exponential Filter,” 10th International
Conference on Pattern Recognition, Vol. 1, No. 16, 1990,
pp. 709-711.
[24] R. C. Hardie and C. G. Boncelet, “Gradient-Based Edge
Detection Using Nonlinear Edge Enhancing Prefilters,”
IEEE Transactions on Image Processing, Vol. 4, No. 11,
1995, pp. 1572-1577. doi:10.1109/83.469938
[25] Y. H. Lee and A. T. Fam, “An Edge Gradient Enhancing
Adaptive Order Statistic Filter,” IEEE Transactions on
Acoustics, Speech and Signal Processing, Vol. 35, No. 5,
1987, pp. 680-695.
[26] H. G. Longbotham and D. Eberly, “The WMMR Filters:
A Class of Robust Edge Enhancers,” IEEE Transactions
on Signal Processing, Vol. 41, No. 4, 1993, pp. 1680-
1685. doi:10.1109/78.212743