1201 2050 PDF
1201 2050 PDF
1201 2050 PDF
AND PEPPER
Tina Gebreyohannes1 and Dong-Yoon Kim2
1
ABSTRACT
In this paper, a new adaptive noise reduction scheme for images corrupted by impulse noise is presented.
The proposed scheme efficiently identifies and reduces salt and pepper noise. MAG (Mean Absolute
Gradient) is used to identify pixels which are most likely corrupted by salt and pepper noise that are
candidates for further median based noise reduction processing. Directional filtering is then applied after
noise reduction to achieve a good tradeoff between detail preservation and noise removal. The proposed
scheme can remove salt and pepper noise with noise density as high as 90% and produce better result in
terms of qualitative and quantitative measures of images.
KEYWORDS
MAG, Directional filtering, Noise detection, Noise reduction
1. INTRODUCTION
The two most common types of noise in image processing are Gaussian noise and Impulse
noise, also known as salt and pepper noise. This type of noise may appear in digital image due
contaminated impulse noise, which is caused by malfunctioning pixels in camera sensors, faulty
memory location in hardware, or transmission in noisy channel [9]. Salt and pepper noise
scattered throughout the image in such a way pixels can take only the maximum and minimum
values (0 and 255 respectively) in the dynamic range.
Many researches have been conducted and numerous algorithms were proposed to remove salt
and pepper noise. Among these noise reduction techniques, majority splits the noise removal
procedures into preliminarily detection of pixels corrupted by impulse noise followed by
filtering the noise detected on the previous phase [8]. Standard median filter (SMF) [7] was one
of famous among others due to its great denoising performance and computational efficiency.
But since the conventional median filter applies the median operation to each pixel whether it is
corrupted or not, it suffers from preserving some details of the image as the noise density
increases. More improved algorithms such as adaptive median filter (AMF) [2], decision-based
algorithm (DBA) [1] and convolution-based algorithm (CBA) [3] mainly focus on noise
detector. Pixels detected by the noise detector will be considered as noise and shall further be
processed in their respective noise reduction scheme. Having such mechanism with phases
would highly preserve the details of the image and save the restored image from having blurred
and distorted feature.
The proposed algorithm in this paper also greatly focuses on how to effectively detect the salt
and pepper noise and efficiently restore the image. The mechanism adopted by the proposed
scheme consists of three phases. The first phase determines whether a pixel is noise or not based
on some predefined threshold and calculated values. Once pixels are detected as noise in
previous phase, their new value will be estimated and set in noise reduction phase. Finally a
conditional image enhancement phase will be conducted for those images which have been
corrupted with high density noise to preserve edges and details of the restored image. This
makes the proposed algorithm to have an outstanding performance even at noise density as high
as 90%.
The rest of this paper is organized as follows. Section 2 describes the proposed scheme. Section
3 presents the experimental results. And the last section, Section 4 concludes this paper.
2. PROPOSED SCHEME
In this paper, three steps adaptive noise reduction scheme is proposed for salt and pepper noise.
For (i,j) X = {1,..,N} x {1,..,N}, let xij be the gray level of the original image X. The
observed grey level at pixel location (i,j) is given by :
(1)
(2)
Where F(i) denotes the intensity value of pixel in a region, F(0) is the intensity of the pixel in
the center as shown in Fig. 1 and N is the number of pixels in the region.
(3)
A reasonable threshold T can be determined using computer simulation. If Zij= 1, then the pixel
Xij is marked as noise candidate; otherwise the pixel Xij is noise free.
(4)
By sorting elements in M according to their distance from the center pixel, we get the sorted
sequence as follow:
i.
,
Sort the five element { ,
we get {
} where {
Median =
ii.
Sort
, and
the
other
where {
Median =
four
(i)
{
}
(5)
Where F(i) denotes the intensity values of the noisy image, N is the number of pixels in the
image and is mean of the noisy image. If standard deviation is greater than the threshold T,
then directional filter will be applied to the processed image accordingly.
A pixel Xk in the image is partitioned into (Y1, Y2, Y3, Y4) where {X1, X2, X4} belongs to Y1,
which are in the upper left portion of the mask.
So that
(7)
Y2, Y3 and Y4 with {X2, X3, X6}, {X4, X7, X8} and {X6, X8, X9} respectively will also be
calculated in the same manner.
(8)
Xk
X2
X1
y2
y1
X4
X6
Xk
y3
X7
X3
y4
X8
X9
3. EXPERIMENTAL RESULTS
In this section, the experimental result of the proposed scheme is presented. A variety of
simulations are carried out on 512x512 grey scale Lena image to verify the performances of
various restoration methods, including AMF[2], DBA[1],CBA[3] and the proposed scheme. In
the simulation, images are corrupted by salt (255) and pepper (0) noise with equal
probability. The noise density is varied from 10% to 90% with 10% incremental, and the
restoration performance is quantitatively measured by peak signal-to-noise ratio (PSNR). Peak
signal to noise ratio (PSNR) is defined as:
(9)
where 255 is the peak signal for 8 bit image and MSE is mean square error given by:
(10)
Where fij and flij are the pixel values at position (i,j) of original and processed image
respectively.
The PSNR comparison of various algorithms including DBA, AMF, CBA and the proposed
scheme are presented in Figure 3 graphically. Figure 4 and Figure 5 show the subjective visual
qualities of the filtered image using various algorithms for the image Lena corrupted by 50%
and 80% impulse noise respectively.
Figure 3.Comparison of various scheme in PSNR for Lena image with various ratio of salt and
pepper noise
When the noise density is greater than 50%, other algorithms PSNR drops sharply unlike the
proposed algorithm which doesnt show drastic change in PSNR values. On the other hand, to
further demonstrate visual quality of images for example Figure 5 presents restoration of Lena
image initially corrupted with 80% impulse noise density. For image processed with DBA, from
the subjective quality of the image one can observe that some of the impulse noises are not fully
suppressed while blurring and distortion are serious. AMF and CBA suppress much of the noise
where CBA fails to preserve edge information while maintaining details of the image unlike
AMF which suffers to maintain both. Both the performance of PSNR and the subjective visual
qualities of our method outperform other algorithms i.e. DBA, AMF and CBA in all cases.
Specially as noise density becomes high and images are greatly corrupted by the impulse noise,
our scheme shows remarkable result after images are being processed.
Figure 4.Image restoration results of Lena image. (a) 50% noise corrupted, (b) DBA, (c) AMF,
(d) CBA, (e) proposed
Figure 5. Image restoration results of the Lena image. (a) 80% noise corrupted, (b) DBA, (c)
AMF, (d) CBA, (e) proposed
4. CONCLUSION
In this paper, a new adaptive noise reduction scheme for removing salt and pepper noise is
proposed. The first phase of the scheme efficiently identifies impulse noise while the other is to
remove the noise from the corrupted image that is followed by image enhancement scheme to
preserve the details and image quality. As per the experimental results, the proposed algorithm
yields good filtering result using efficient noise detection mechanism. This is observed by
numerical measurements like PSNR and visual observations through the experiments
conducted.
ACKNOWLEDGEMENTS
The support of the Agency for Defence Development Korea is gratefully acknowledged.
REFERENCES
[1]
Sarinivasan, K.S., and Ebenezer, D., A new fast and efficient decisioin based algorithm for
removal of high-density impulse noises, IEEE signals process. Lett, pp. 189-192, 2007.
[2]
Chan, R.H., Ho, C.W., and Nikolova, M, Salt-and-pepper noise removal by median-type noise
detectors and detail-preserving regularization, IEEE Trans. Image process., pp. 1479-1485,
2005.
[3]
Wang, S.H., and Wu, C.H, A new imples detection and filtering method for removal of wide
range impulse noises, Pattern Recognitioin., pp. 2194-2202, 2009.
[4]
S. Zhang, M.A. Karim, A new impulse detector for switching median filters, IEEE Signal
Process. Lett., pp. 360363, 2002.
[5]
Chan, R.H., Chen Hu., and Nikolova, M.,An Iterative procedure for removing random-valued
impulse noise , IEEE signal process. Lett, pp. 921-924, 2004.
[6]
[7]
I. Pitas, A.N. Venetsanopoulos, Order statistics in digital image processing, Proc. IEEE., pp.
18931921, 1992.
[8]
[9]
S. Huang and J.Zhu., Removal of salt-and-pepper noise based on compressed sensing, IEEE
Electronic Letters, Vol.46(17), pp.1198-1199, 2010.
[10]
Kenny K.V. Toh and Nor A.M. Isa.,Noise adaptive fuzzy switching median filter for Salt-andPepper noise reduction, IEEE signal processing letters, Vol.17(3), pp.281-284, 2010
[11]
Chen C. , Wng J., Quin W. and Dong X., A new adaptive weight algorithm for Salt and Pepper
noise removal, Consumer Electronics, Communication and Networks, pp.26-29, 2011.
Authors
Tina Gebreyohanes received her B.Sc. degree in 2007 in Computer Science
from Addis Ababa University, Ethiopia. She worked for a Telecom company
called ZTE(H.K) Ethiopian branch as Fixed Line Next Generation
Network(FLNGN) Engineer. She is currently M.Sc. student in Information
and communication Technology Faculty, Computer Engineering Department
in Ajou University. Her research interest includes image processing and data
compression.