An Unique Edge Preserving Noise Filtering Technique for Impulse Noise Removal

Signal & Image Processing : An International Journal (SIPIJ) Vol.2, No.3, September 2011
DOI : 10.5121/sipij.2011.2304 33
AN UNIQUE EDGE PRESERVING NOISE
FILTERING TECHNIQUE FOR IMPULSE
NOISE REMOVAL
Kartik Sau 1
, Amitabha Chanda 2
, Pabitra Karmakar 3
1
Department of Computer Science & Engineering, Budge Budge Institute of
Technology, Nischintapur. Kolkata 7000137, WBUT, India,
kartik_sau2001@yahoo.co.in
2
Guest faculty, UCSTA, University of Calcutta, West Bengal, India.
amitabha39@yahoo.co.in
3
Department of Computer Science & Engineerin, Institute of Engineering and
Management, WBUT, Salt Lake, Sec –V, Kolkata - 700091, India.
pab_comp@yahoo.co.in
Abstract
Image de-noising is the technique to reduce noises from corrupted images. The aim of the image de-
noising is to improve the contrast of the image or perception of information in images for human viewers
or to provide better output for other automated image processing techniques. This paper presents a new
approach for color image de-noising with Fuzzy Filtering techniques using centroid method for
defuzzification. It preserves any type of edges (including tiny edges) in any direction. The experimental
result shows the effectiveness of the proposed method.
Keywords
Impulse noise; Median filter; Color image de-noising; Defuzzification; Edge Preservation;
1. INTRODUCTION
Color provides a significant portion of visual information to human beings and enhances their
ability of object detection. In the RGB color model, the three primaries are Red, Green, and
Blue. They can be combined, two at a time to create the secondary hues. Magenta is (R+B),
cyan is (B+G), and yellow is (R+G). There are two reasons for using the red, green, and blue
colors as primaries: (1) the cones in the eye are very sensitive to these colors and (2) adding
red, green, and blue can produce many colors although not all colors. For various reasons
digital images are often contaminated with noise at the time of acquisition or transmission. The
noise introduces itself into an image by replacing some of the pixels of the original image by
new pixels having luminance values near or equal to the minimum or maximum of the allowable
dynamic luminance range. Pre-processing of an image is conducted with a view to adjusting
the image for further classification and segmentation. In the process, however, image features
should not be destroyed. This is a difficult task in any image processing system. For this
purpose various types of filters are used. Among those median filter is an important class of
filters. In the present paper we shall discuss some of the median filters.

34
1.1 Different median filters:
Some important median filters are discussed below in short.
The standard median filter [28,24,8,17,11] : In this filter luminance values in a window are
arranged in an order and the median value is selected. This filter reduces noise reasonably
well, but in the process some information is also lost at low noise densities.
The weighted median filter [30,19,11,10] and the centre-weighted median filter [16,13,25,6,11]
have been proposed to avoid the inherent drawbacks of the standard median filter by
controlling the trade off between the noise suppression and detail preservation.
The switching median filter [31,11] is a type of median filter with an impulse detector. It is so
designed that if the centre pixel is identified by the detector as a corrupted pixel, then it is
replaced with the output of the median filter, otherwise, it is left unchanged. The tri state
median filter [7] and the multi-state median filter (MSMF) [8] are two modification of
switching median filter.
Two other types of switching median filters worth mentioning are progressive switching
median filter (PSMF) [29] and Signal-dependent rank-ordered mean filter (SDROMF) [1,2].
The progressive switching median filter (PSMF) [29] is a derivative of the basic switching
median filter. In this filtering approach, detection and removal of impulse noise are iteratively
done in two separate stages. Despite its improved filtering performance it has a very high
computational complexity due to its iterative nature. Signal-dependent rank-ordered mean filter
(SDROMF) [1,2] uses rank-ordered mean filter.
Adaptive centre weighted median (ACWM) [19,11] filter avoids the drawbacks of the CWM filters
and switching median filter. Input data will be clustered by scalar quantization (SQ) method,
this results in fixed threshold for all of images.
Fuzzy-Median filters of different types have gained lot of importance in image processing.
Some such filters [3,18,27,5,22,15] are worth mentioning.
2. PROPOSED METHOD FOR DE-NOISING
2.1 Phase I
Any given digital RGB color image can be represented by a two dimensional array of size
M×N for each Red, Green and Blue primaries. When we are capturing or transmitting the
images, then there will be some noises due to improper opening and closing of the shutter,
atmospheric turbulence, misfocus of lens, relative motion between camera and objects [12].
For removing the noise from the image we can pursue the following steps:
Step 1: Let the input and output images of median filter are X(i,j) & y1(i,j) respectively. The
median filter uniformly replaces the central pixel of a window (W) by the median of the pixels
bounded by predefined window (W) size w × w. The output of the median filter [26,10,32] is
given by
y1(i,,j) = median { X(i-s; j-t) :(s,t) ε W} (1)
and W= {(s, t): -n≤ s, t ≤ n } (2)

35
Here we consider n=1
Step 2: The pixel values of the neighborhood pixels of (i,j) of original image are sorted in
some specific order.
Step 3: The Fuzzy membership [14,22,20,17] value is assigned for each pixel in a window of w
× w size using some suitable membership functions. The membership function used here is
given below.
(i) A triangular shaped membership function is used.
(ii) The highest and lowest gray values get the membership value zero.
(iii) Assign the membership value 1 to the mean value of the gray level of the window of size
w×w.
The triangular membership function [14,20,17], also called bell-shaped function with straight
lines, can be defined as follows:
µ(x;α,β,γ) = 0 if x ≤ α
= (x-α)/ (β- α+1) if α < x ≤ β (3)
= (α-x)/ (β- α+1) if β < x ≤ γ
One typical plot of the triangular membership is given in the following figure.
Gray value →
Figure 1: Triangular membership function
Step 4: Now de-fuzzyfy [14,20,21,32] the membership value using the Centroid method by the
following formula, and select the output for that window. Let it be y2(i,j) at the point (i,j) in the window
of size w×w
∑
∑
=
x x
x x
j
i
j
i
x
j
i
y
)
,
(
)
,
(
)
,
(
2
µ
µ
(4)
2.2 Phase II
If we apply the phase I on the noisy images, noise would be removed. But the problem is,
though the noises are represented as 0 and 255, so all 0 and 255 will also be removed; even of
those were not noises. For preserving the actual data we pursues the following steps

36
Step 1: For the noisy pixel at the point (i,j), we compute p(i,j) and q(i,j) by the following
formulas [21]:
p(i,j)= |f(i,j) – median{Lf(i,j)}|.
Here L(f(i,j)) is the 8-neighborhood of point (i,j).
2
)
,
(
)
,
(
)
,
(
)
,
(
)
,
(
2
1 j
i
fc
j
i
f
j
i
fc
j
i
f
j
i
q
−
+
−
= (5)
Here )
,
(
1 j
i
fc , )
,
(
2 j
i
fc are the closest values of f(i, j) in the filter window of size w×w.
Then rearrange p(i,j) and q(i,j) in ascending order for all i,j =0,1,2.
Step2: Compute w(i,j)= F(p(i,j), q(i,j)) such that
2
1
)
,
(
)
,
(
)
,
(
1
)
,
(
k
j
i
q
j
i
p
k
j
i
p
j
i
w
+
+
+
−
= (6)
Here k1 and k2 are real quantity, which depends on the quality of the input images. Select the
maximum w(i,j).
Step3: If y1(i, j) ≠ y2(i, j)
then ))
,
(
)
,
(
1
(
)
,
(
)
,
(
)
,
( 2
1
3 j
i
y
j
i
w
j
i
y
j
i
w
j
i
y −
+
= (7)
Else y3(i, j) = y1(i, j)
Step 4: Continue this process for all the pixels which are noisy.
The computational procedure is illustrated below.
Example: Consider a color image of size 3×3, as follows,
Table 1: color image with size 3×3
163 193 151 83 113 71 55 81 39
179 0 138 99 0 58 68 0 29
192 172 127 112 92 47 85 63 21
Red Green Blue

37
For the Green component, we calculate membership values as follows
Original value: 0 Mean value: 75 Median value y1(i,,j): 83
Table 2 : Membership values for pixels
Sorted
order 0
47 58 71 83 92 99 112 113
Member
ship
value
0.00 0.26 0.55 0.89 0.79 0.55 0.37 0.26 0.00
According to Table 2 the graph is as follows –
Gray value →
Figure 2
Now we calculate y2(i,j) using centroid method
Table 3
Sector Area of the sector Centroid of the sector
a 6.184211 23
b 4.486842 52
c 9.407894 64
d 10.105263 77
e 6.039474 87
f 3.223684 95
g 2.565789 105
h 0.013158 112

38
So, the calculated value for ∑
x x
j
i
x )
,
(
µ = 2858.328857
And the calculated value for ∑
x x
j
i )
,
(
µ = 42.02631
Therefore, )
,
(
2 j
i
y = 68 [Selected value in phase I ]
Rearranged p(i,j) and q(i,j) as follows
Table-4: Different values of p, q and w.
p(i,j) q(i,j) w(i,j)
0 7 0.956522
9 7.5 0.207792
12 8 0.509804
16 10 0.388889
25 10.5 0.388278
29 12 0.328947
30 12.5 0.202703
36 17.5 0.47222
83 52.5 0.330275
Therefore y3(i,j) = 82
Gray value →
Figure 3
The figure 3 shows the membership function for the pixel values of Green component of RGB
image. And also indicate the selected value in the proposed method of Green component.
Now, for the Red component of RGB image, the selected values are as follows
Original value: 0
Mean value: 146
Median value y1(i, j): 163

39
)
,
(
2 j
i
y = 108 [Selected value in phase I]
And y3(i,j) = 160 [in Phase II]
Gray value →
Figure 4 : Membership function for Red component
The Figure 4 shows the membership function for the pixel values of Red component of RGB
image. And also indicate the selected value in proposed method for RED component.
Similarly, for the Blue component of RGB image, the selected values are as follows
Original value: 0
Mean value: 49
Median value y1(i,,j): 55
)
,
(
2 j
i
y = 47 [Selected value in phase I]
and y3(i,j) = 54
Gray value →
Figure 5: Membership function for Blue component
The Figure 5 shows the membership function for the pixel values of Blue components of RGB
image. And also indicate the selected value in proposed method.

40
3. FLOW CHART
The flow chart of the proposed method is shown below.
Figure 6
4. EXPERIMENTAL RESULT
The effectiveness of the proposed method is shown experimentally. It eliminates the fixed
value impulse noise. And it is tested for different images in different sizes. The peak signal-to-
noise ratio (PSNR) [19,22,23,27,29,30] value gives the performance of restoration quantitatively,
which is defined as








−
=
∑
∑
k
k
k
y
k
d
PSNR 2
2
10
))
(
)
(
(
255
log
10 (8)
Where 255 is the peak pixel value of the image, d(k) represents the value of the desired output,
and y(k) represents the value of the physical output.
The experimental results were compared with median-filter (MED).

41
4.1 Original Images
4.2 Noisy Images

42
4.3 Median Filter Output
4.4 Proposed Filter Output

43
Table 5: PSNR Comparison Table
Image name PSNR for Median Filter PSNR Proposed filter
lena 27.583501 27.588477
bamboo 25.972569 25.980521
track 25.941161 25.949895
grass 24.008344 24.012961
road 25.584527 25.599688
forest 25.729469 25.743479
wires 23.657962 23.666070
s1 25.216643 25.2242
s2 24.469417 24.473686
s3 24.610376 24.618010
Table 6: Edge Count Comparison Table
Image name Original Image
Median Filter
Output
Proposed Filter
Output
lena 22384 22267 22383
bamboo 32688 32563 32663
track 34686 34468 34576
grass 27381 27224 27244
road 39945 39668 39830
forest 32062 31650 31754
wires 41625 41337 41425
s1 33467 33522 33339
s2 29437 29370 29415
s3 26717 26577 26695
From the Table 5 we observed that the proposed method is better than the median filter due to
better PSNR. From the Table 6 we conclude that more edges [9] be preserved for the proposed
method than median filter.

44
5. CONCLUTION
In this paper, we propose a new filtering approach for noise removal. It based on soft
computing techniques which can remove the noise and at the same time tiny edges can be
preserved. This enhances the reliability for getting the output image, which is more or less
same with original image. From the Table 5 we conclude that the best PSNR can be achieved
by using the proposed method. The proposed method is tested more than 1000 pictures,
which gives the better result for each test image.
REFERENCE
[1] Abreu E., Lightstone M., Mitra S. K., Arakawa K., “A new efficient approach for the
removal of impulse noise from highly corrupted images,” IEEE Trans Image Processing,
vol. 5, no. 6, pp. 1012-1025, 1996
[2] Abreu E., Mitra S.K., “A signal-dependent rank ordered mean (SD-ROM) filter. A new approach
for removal of impulse From highly corrupted images”, Proceedings of IEEE ICASSP-95,
Detroit, MI, 1995, pp. 2371-2374.
[3] Arakawa K., “Median filters based on fuzzy rules and its application to image restoration”,
Fuzzy Sets and Systems 77 (1996) 3-13.
[4] Baudes, A., B. Coll, Morel J, M, 2005. “A non-local algorithm for noise denoising”, Computer Vision
and Pattern Recognition, CVPR. IEEE Conf., 2: 60-65, 20-25
[5] Bezdek J. C., Pattern Recognition with Fuzzy Objective Function Algorithms. New York:
Plenum, 1981
[6] Chen T.,. Wu H. R., “Adaptive impulse detection using center- weighted median filters,” IEEE
Signal Processing Letters, vol. 8, no. 1, pp. 1–3, 2001.
[7] Chen T,. Ma K. K ,. Chen L.-H, “Tri-state median filter for image denoising”, IEEE Trans. Image
Processing, vol. 8, no. 12, pp. 1834 – 1838, 1999.
[8] Chen. T., Wu H. R., “Space variant median filters for the restoration of impulse noise corrupted
images,” IEEE Trans. on Circuits and Systems II: Analog and Digital Signal processing, vol.
48, no. 8, pp. 784–789, 2001.
[9] Gonzalez. R. C, Digital Image processing. 2nd
Edition TMH
[10] Hamza A., Krim H., “Image denoising A non linear robust statistical approach.” IEEE. Trans. Signal
Processing 49(2), pp. 3045-3054; 2001
[11] Hwang H., Haddad R. A., “Adaptive Median filters: New algorithms and results”. IEEE transaction
on Image processing, 4, pp. 499-502 1995
[12] Jayaraman S., Esakkirajan, S., Veerakumar T, “Digital Image processing”, Tata McGraw Hill
Education Private Limited.
[13] Jeong B. , Lee Y. H., “Design of weighted order statistic filters using the perceptron algorithm,”
IEEE Trans. Signal Processing, vol. 42, no. 11, pp. 3264–3269, 1994.
[14] Konar A, Computational Intelligence, Springer, New York, 2004

45
[15] Lakshimiprabha S., A new method of image denoising based on fuzzy logic. International Journal of
Soft computing 3(1): 74-77, 2008
[16] Lee Y. H., Ko, . S.-J “Center weighted median filters and their applications to image enhancement,”
IEEE Trans. Circuits and Systems, vol. 38, no. 9, pp. 984 – 993, 1991.
[17] Lee, C. S. V H Kuo, P. T Yu. “Weighted Fuzzy mean filters for Image processing”. Fuzzy Sets Sys.
89, 157-180; 1997
[18] Lin T.-C., Yu P.-T., Partition fuzzy median filter based on fuzzy rules for image restoration, Fuzzy
Sets and Systems 147 (2004) 75 - 97.
[19] Lin T.C., Yu P.-T, “ A new adaptive center weighted median filter for suppressing impulsive noise in
images”, Information Sciences 177 (2007) 1073-1087.
[20] Ross. T. J, Fuzzy logic with Engineering Applications, 2nd edition, University of New Mexico, USA,
Wiley India.
[21] Sadoghi H., Yazdi, Homayouni F., “Impulsive Noise Suppression of Images Using Adaptive Median
Filter”, International Journal of Signal Processing, Image Processing and Pattern Recognition Vol. 3,
No. 3, September, 2010
[22] Sau K., Chanda A.,“Image De-Noising with A Fuzzy Filtering Technique”, India, 2011
[23] Sau K., Chanda A., Karmakar P, “Image De-noising with Edge Preservation Using Soft Computing
Techniques” Annual International Conference on Emerging Research Areas(AICERA). 2011 in
Collaboration with IUPUI United States
[24] Sonka, M., Hlavac V., Boyle R.,Image Processing, Analysis, and Machine Vision, PWS Publishing,
Pacific Grove, Calif, USA, 1999.
[25] Sun T. , Neuvo Y, “Detail-preserving median based filters in image processing,” Pattern
Recognition Letters, vol. 15, no. 4, pp. 341– 347, 1994.
[26] Somasundram K. Shanmugavadivy P. Adaptive iterative order Statistics Filter, ICGST-GVIP journal
Vol. 9 issue 4, pp 23-32, 2009
[27] Smolka B., Chydzinski A., Fast detection and impulsive noise removal in color images, Real-Time
Imaging 11 (2005) 389 - 402.
[28] Umbaugh, S. E. Computer Vision and Image Processing, Prentice-Hall, Englewood Cliffs, NJ, USA,
1998.
[29] Wangand Z., Zhang D., “Progressive switching median filter for the removal of impulse noise
from highly corrupted images,”IEEE Trans. on Circuits and Systems II: Analog and Digital Signal
Processing, vol. 46, no. 1, pp. 78–80, 1999.
[30] Yli-Harja O., Astola J., and Neuvo Y., “Analysis of the properties of median and weighted median
filters using threshold logic and stack filter representation,” IEEE Trans Signal Processing, vol. 39,
no. 2, pp. 395– 410, 1991.
[31] Zhang S.,. Karim M. A, “A new impulse detector for switching median filters,” IEEE Signal
Processing Letters, vol. 9, no. 11, pp. 360–363, 2002
[32] Zimmermann H. J “Fuzzy set theory- And its applications”, second edition, Allied publishers limited.

46
AUTHORS
[1] Kartik Sau
Kartik Sau completed his B.Sc. in mathematics from R. K Mission Vidyamandira,
University of Calcutta. And M. Sc. in the same subject from Indian Institute of
Technology, M. Tech in Computer Science from Indian School of Mines,
Dhanbad. Currently he is the Head of the department of Computer Science &
Engineering in Budge Budge Institute of Technology and a member of the
governing body of the same institution. He has presented many papers in
International and National journals and Conferences. His area of interest includes
Digital Image processing, Artificial Intelligence, Pattern Recognition, Soft
computing, etc. He has more than eight years of teaching and research experience
in his area of interest.
[2] Dr. Amitabha Chanda
Dr. Amitabha Chanda: visiting Professor; Department of Computer Science, UCSTA; Calcutta
University. He completed his B.E in Chemical Engineering (Jadavpur), M.A. in Pure Mathematics and
Ph.D in Mathematics from university of Calcutta. He was a faculty member of Indian Statistical Institute
(ISI), Kolkata. Now he is also guest faculty member of ISI, Kolkata; Department of Computer Science,
Rajabazar Science College, Kolkata. His area of interest includes Digital Image processing, Pattern
Recognition Fuzzy logic, Genetic Algorithms, Computer Graphics, Control system Turbulence, Fractal,
multifractals, Clifford Algebra, Nonlinear dynamics. He has more than fifty years teaching and research
experience in his area of interest. He has presented more than 50 papers in International and National
journals and Conference. Dr. Chanda is an Associate member of American Mathematical Society.
[3] Pabitra Karmakar
Pabitra Karmakar received his M. Tech degree in Computer Science &
Engineering from Institute of Engineering & Management, Salt Lake Kolkata,
India. He completed his B. Tech degree in Computer Science & Engineering
from Dumkal Institute of Engineering & Technology, WBUT, India. Also he was
a faculty member of Institute of Engineering and Management in Department of
Computer Science.& Engineering. He have certified in SCJP 1.4 from Sun
Microsystem, USA and MTA certified form Microsoft Corporation. Also he is an
IBM certified C and C++ programmer. Beside this, He has a industry experience
in software development. Currently he is an IT professional. His area of interest
includes Digital Image processing, Pattern Recognition, Fuzzy logic, Genetic
Algorithms, Neural Networks, RDBMS and programming languages.

An Unique Edge Preserving Noise Filtering Technique for Impulse Noise Removal

More Related Content

Similar to An Unique Edge Preserving Noise Filtering Technique for Impulse Noise Removal (20)

Recently uploaded (20)

An Unique Edge Preserving Noise Filtering Technique for Impulse Noise Removal