ISSN XXXX XXXX © 2016 IJESC

Research Article Volume 6 Issue No. 8

A Comparison of Salt and Pepper Noise Removal Filters

Shivani Sharma 1 , Er. Kamal Ku mar2
M.Tech Student (ECE) 1 , Assistant professor in ECE Depart ment 2
MVEC, JA GA DHRI, India

Abstract:
The image enhancement is the basic problem in the digital image processing. The salt & pepper noise (Impulse Noise) is the on e
type of noise that occurs in images during transmission or analog to digital conversion. This paper deals the performance
comparison of the various filters wh ich are used to remove the impulse noise fro m the images. The performance criteria are the
mean square error (MSE) and peak signal to noise ratio (PSNR).

Keywords: salt & pepper No ise; Impulse Noise; MSE; PSNR.

INTRODUCTION Impulse noise removal Filters

The basic problem in image processing is the image
enhancement and the restoration in the noisy environment. If
we want to enhance the quality of images, we can use
various filtering techniques which are available in image
processing. There are various filters which can remove the
noise from images and preserve image details and enhance
the quality of image. Salt & Pepper noise is the one type of
noise which occurs due to the dead pixels or analog to digital
conversion of images. To remove this impu lse noise we have
filters like M in. filter, Max. filter , M inMax . Filter, Mean
filter, Median filter, weighted median filter, Adaptive
Median Filter. In this paper we check that which filter is best
for impulse noise removal.
Noise Model
Image noise is the degradation of the quality of the image.
Image noise is produced due to the random variation of the
brightness or the color information in images that is
produced by the sensors and the circuitry of the scanner or
digital cameras. Image noise can also originate in film grain
and in the unavoidable shot noise of an ideal photon
detector. Image noise is generally regarded as an undesirable
by-product of image capture.
Salt-and-pepper noise
The image which has salt-and-pepper noise present in image
will show dark p ixels in the bright regions and bright pixels
in the dark regions. [1]. The Impu lse noise occurs due to the
quick transitions such as faulty switching , can be caused by
the dead pixels, or due to analog-to-digital conversion
errors, or b it erro rs in the transmission, etc. This all can be
eliminated in large amount by using the technique dark
frame subtraction and by interpolating around dark/bright
pixels.
The pdf of Impu lse noise is given by:
p(z) =
There are various filters wh ich can be used to remove the
impulse noise which is explained below:
Max. Filter
The max filter is used to find the brightest point in an image.
It uses the maximu m intensity value in a sub image area.
This filter reduces the pepper noise because it has very low
values of intensities. So it can only remove pepper noise.
(x,y ) =
Where, g(s ,t) is the sub image area of m*n image.
Mi n Filter
This filter is used to find the darkest point in an image. It
uses the min imu m intensity value in a sub image area. This
filter reduces the salt noise as a result of min operation.
(x,y ) =
Where g(s, t) is the sub image area of the m*n image.
MINMAX Filter
This filter is the combination of the Min and Max filter. This
filter does not remove the impu lse noise but it is best suitable
for the Gaussian or uniform noise.
(x,y )= 1/2[
Where g(s, t) is the sub image area of the m*n image.
Mean Filter
This filter is used to remove the salt & pepper noise both
simu ltaneously fro m the image. We consider a sub image
area of size m*n centered at (x,y).We find the mean value of
that sub image area and replaces the mean value with the
central value. The image details are not preserved in this
operation, some details are lost.
(x,y ) =
Where g(s, t) is the sub image area of the m*n image.

If b>a, intensity b will appear as a light dot in the image or Median Filter
level a will appear like a dark dot or vice -versa. Median filtering is a nonlinear operation used in image
processing to reduce "salt and pepper" noise. Also Mean
filter is used to remove the impu lse noise. Mean filter

International Journal of Engineering Science and Computing, August 2016 2627 http://ijesc.org/
replaces the mean of the pixels values but it does not
preserve image details. Some details are removes with the
mean filter. In the median filter, we do not replace the pixel
value with the mean of neighboring pixel values, we replaces
with the median of those values. The median is calculated by
first sorting all the pixel values fro m the surrounding
neighborhood into numerical o rder and then replacing the
pixel being considered with the middle pixel value. (If the
neighboring pixel which is to be considered contains an even
number of pixels, than the average of the two middle pixel
values is used.) Fig.1 illustrates an examp le calculation.

original cameraman image in .tif format

Fig.1

Fig.1:Exp . of median filtering

The med ian filter g ives best result when the impulse noise
percentage is less than 0.1 %. When the quantity of impulse
noise is increased the median filter not gives best result.
Now we consider a sub image area of total image.

Where g(s, t) is the sub image area of the m*n image.

Impulse Noisy Image With % 0.01
Weighted Median Filter
In the wmf the weights are assigned to the each element in a Fig.2
window. These weights are mult iplied to each element in the MSE=1603.35,PSNR=16.11
window. This filter also removes the impulse noise but
image details are lost in this filter.
MS E & PSNR
The term MSE (mean square error) is the difference between
the original image and the recovered image and it should be
as minimu m as possible.
The term peak signal-to-noise ratio, PSNR, is the ratio
between the maximu m possible
Power o f a signal and the power of corrupting noise signal.

MSE=
output of Min. Filter
The PSNR is defined as:
Fig.3
MSE=2024.04,PSNR=15.10
PSNR = 10.

= 20.

Where,
MAXI is the maximu m possible pixel value of the image.

Simulati on Results

We perform the result on the original cameraman.tif image.

We check the performance on the different % of the impulse
noise. The med ian filter gives the best result. These filters
output of Max. Filter
give the best result when the impulse noise percentage is
0.1%. The PSNR is increasing when the noise % is Fig.4
decreasing.

International Journal of Engineering Science and Computing, August 2016 2628 http://ijesc.org/
 MSE=143.43,PSNR=26.60 Table 1 shows that when the % of impu lse noise is
decreasing, the mse is decreasing & PSNR is increasing.
Impulse
Mi n. Filter Max. Filter
Noise
(In
Percentage) MS E PSNR MS E PSNR
0.2 11268.04 7.65 13931.34 6.72
0.1 7263.7 9.55 8750.2 8.74
0.01 1603.4 16.11 2024 15.1
0.001 956.1 18.36 1074.3 17.85
0.0001 885.96 18.69 1010.4 18.12
output of MinMax. Filter
Table 1

Fig.5 Table 2 shows that when the % of impu lse noise is

MSE=28.41,PSNR=33.63 decreasing, the mse is decreasing & PSNR is increasing.

Impulse
Mi nMax. Filter Mean Filter
Noise
(In
Percentage) MS E PSNR MS E PSNR
0.2 150.98 26.13 77.74 29.26
0.1 150.8 26.38 60.25 30.37
0.01 143.43 26.8 28.41 33.63
0.001 141.43 26.92 23.43 34.47
0.0001 141.34 27.06 23.06 34.54
output of Mean. Filter
Table 2
Fig.6
MSE=28.29,PSNR=33.65
Table 3 shows that when the % of impu lse noise is
decreasing, the MSE is decreasing & PSNR is increasing.

Impulse Weighted
Median Filter
Noise Median
Filter
(In
Percentage) MS E PSNR MS E PSNR
0.2 32.48 33.05 19.55 35.25
0.1 30.28 33.35 17.37 35.77
0.01 28.29 33.65 15.85 36.16
output of WMF. Filter 0.001 28.08 33.68 15.7 36.21
Fig.7 0.0001 28.09 33.69 15.68 36.21
MSE=15.85,PSNR=36.16
Table 3

Fro m the above table we conclude that when the impulse

noise % is decreasing the PSNR.

Conclusion
Fro m the above tables & figures we conclude that the
med ian filter is the best salt & pepper noise removal filter.

References:
[1] Image Denoising using Wavelet Thresholding and Model
Selection by Shi Zhong. Image Processing, 2000,
median filter output
Proceedings, 2000 International Conference held on,
Fig.8 Vo lu me: 3, 10-13 Sept. 2000 Pages: 262.

International Journal of Engineering Science and Computing, August 2016 2629 http://ijesc.org/
[2] Performance Co mparison of Median and Wiener Filter in
Image De-noising. International Journal of Co mputer
Applications Vo lu me 12– No.4, November 2010 Page No.
(0975 – 8887).
[3]Leslie Stroebel and Richard D. Zakia, “The Focal
encyclopedia of photography”. Focal Press, 1995.
[17]Changhong Wang Taoyi Chen ZhenshenQu,“A Novel
Improved Medium Filter for Salt-and-Pepper Noise fro m
Highly Corrupted Images”, International Sy mposiu m on
Systems and Control in Aeronautics and Astronautics
(ISSCAA), Harbin, China ,10 june 2010,718-722

[4]Gonzalez C. Rafael, Woods E. Richard, “Dig ital Image

Processing” , Dorling Kinderslet, 3rd edition, 2009.

[5]Wikipedia,“http://en.wikipedia.org/wiki/ Gaussian_noise”,
July 2009.

[6]Wikipedia,“http://www.b ioen.utah.edu/wiki/ index.php?tit

le=Noise_modeling_ and_Probabiliy_Theory”.

[7]Shi Zhong, “Image Denoising using Wavelet

Thresholding and Model Selection”, International
Conference on Image Processing, 2000, Vancouver, BC,
Canada, Vo lu me: 3, 10- 13 Sept. 2000 Pages: 262-265.

[8]JohnsonNyquist,Wikipedia,“http://en.wikipedia.o rg/wiki/I
mage_noise”,
http://books.google.com/books?id=oqOK8zpM7jIC&pg=PA
207&dq=photon- shot noise+image+bright+areas,2006.

[9]http://en.wikipedia.org/wiki/ Peak_signal-to-noise_ratio
[10]http://www.kxcad.net/cae_MATLAB/toolbo x/images/f1
6-14527.ht m

[11] Matlab software, ”Image Processing”, 2007b. R.Boyle,

R.Tho mas,“http://homepages.inf.ed.ac.uk/rb f/HIPR2/ med ian
.htm,Co mputer Vision”A First Course,Blackwell Scientific
Publications, 1988, 32 - 34.

[12]Shaomin Peng, Lo ri Lucke , “Multi-level Adaptive Fu zzy

Filter for M ixed Noise Removal”, IEEE International
Symposiu m on, Seattle, WA , USA,30 Apr-3 May 1995,
1524 - 1527 vol.2 .

[13]Shaomin Peng, Lo ri Lucke,“A Hybrid Filter For Image

enhancement”, International Conference on Image
Processing, Washington, DC, 23-26 Oct 1995, 163 - 166
vol.1.

[14]Sureshkumar, Papendra ku mar, Manoj Gupta, Ashok

Ku mar Nagawat, “Performance Co mparison of Median and
Wiener Filter in Image De-noising”, International Journal of
Co mputer Applications, Dec. 2010, Vo l. 12, Issue No.-
10.5120,Page no.1664-2241.

[15]J.Najeer Ahamed, V. Rajamani, “Design of Hybrid Filter

for Denoising Images Using Network and Edge Detecting”,
American Journal of Scientific Research, ISSN 1450-223X
,Issue 3,2009.

[16]Chan, R.H., Chung-Wa Ho, Niko lova, M., “Salt-and-

Pepper Noise Removal by Median-Type Noise Detectors and
Detail-Preserving Regularizat ion”, IEEE Transctions On
Image Processing, Hongkong China,Oct. 2005.

International Journal of Engineering Science and Computing, August 2016 2630 http://ijesc.org/