International Journal of Advanced Computer Engineering and Communication Technology (IJACECT)
________________________________________________________________________
Local median information based adaptive fuzzy filter for impulse noise removal
1
Prajnaparamita Behera,
2Shreetam Behera
1Final Year Student, M.Tech VLSI Design, Dept. of ECE, 2Asst .Professor, Dept. ECE CIT, Centurion University of Technology & Management Jatni (Odisha), India
Email: [email protected]
Abstract— Impulse noise removal is still a great challenging job in the field of image processing. Lots of linear and nonlinear filters have been proposed earlier for the impulse noise removal but it is found that they degrade the quality of images by blurring.
In this paper a two pass median filter is used to remove impulse noise. In the first pass min-max based median filter is used for detection and correction of noisy pixel. In the second pass local median information based adaptive fuzzy filter is used to denoise the image. The proposed method is efficient, fast and results in a higher PSNR (Peak Signal to Noise Ratio) values when compared to other traditional filters.
Keywords: Impulse noise, blurring, Min-max based median filter, Adaptive, PSNR
I. INTRODUCTION
Image denoising is the most important and challenging job in the field of image processing. During the time of data acquiring, broadcasting and loading the image becomes partial. The noise is come into the images when captured by camera or scanner or while recording and when the image is transmitted by a noisy channel. Salt and pepper noise is one type of noise which is impulsive in nature and most of the techniques used for its removal has nonlinear characteristics. Median filter is the most popular nonlinear filter in image processing .The median filter is not appropriate for non-impulsive noise reduction. The Weighted Median (WM) filter is the modification of standard median filter where a specific weight is given to every pixel present in the window.
CWM is a special type of weighted median filter where weight is specified only the centre pixel of the window.
a MDB filter was introduced in [1]. This proposed technique was found to be more superior than the centre weighted median filter.
In [2] the authors introduced an algorithm in which the noisy pixel is replaced by trimmed median value for denoising the images and it is found to be better in comparison with the standard median filter.
To produce more effective and reduced noise levels , median filter is imbibed with fuzzy technique by the authors in [3] .A switching based fuzzy scheme is introduced by the authors in [4] which is able to eliminate impulse noise from grayscale images to a greater extent. It was also seen that with the increase in the processing window more accurate result was obtained in [5].In [6], the authors proposed a novel approach to detect and remove impulse noise with an additional aim of enhancing the image. The efficiency of adaptive fuzzy filter is well demonstrated in [7] with respect to other traditional median filters.
In this paper a two pass median filtering scheme is proposed for removal of impulse noise from heavily corrupted images. The proposed technique is explained in the section II. Section III analyses and explains the results of the proposed fuzzy scheme followed by the conclusion and references..
II. PROPOSED SCHEME FOR IMAGE DENOISING
In this paper a two pass median filtering scheme is proposed, where in the first pass, the noise is detected and corrected using a Min-Max Based detection based median filter and then an adaptive fuzzy filter based on local window information is used in the second pass. The flow charts give a brief outline of the proposed method.
by the median of the window pixels else left unchanged.
In the second pass to this data obtained from the first pass, local information is found out based on the median values of the absolute gradient values.
Fig.1:Flow chart for Proposed Filtering scheme
Fig.2: Flow chart for Fuzzy Filtering scheme This local information data is fuzzified using on basis of the fuzzy rules given below:
1. If L(i,j) is large ,then µ(L(i,j)) is large.
2. If L (i,j) is small ,then µ(L(i,j)) is small.
It was found that S shaped membership function given below satisfied the above rules and thus was used for the fuzzification of the local information data.
Where a & b are any fixed thresholds.
Then the corrected pixel is obtained by
Where C(i,j)=Corrected image Y(i,j)=Input image
O(i,j)=Image obtained after the first pass filtering
µ(L(i,j))=Fuzzified Local information data
III.SIMULATION RESULTS
The performance of the new scheme has been executed and compared with the existing traditional filters. In our implementation standard grayscale images of size 256 x256 of Cameraman and Lena are degraded by salt and pepper noise at various densities (10% to 80%) and restored by using various methods. Peak Signal to Noise ratio (PSNR) is used as an evaluation tool for comparing different denoising schemes. Peak signal to noise ratio for a gray scale image is defined as:
Where X (i,j) is the original pixel and Y(i,j) is the restored pixel.
Figure 3 carry out the original image of Cameraman of size 256 x256 and Lena which are restored afterwards using the proposed technique and exposed in figure 4 and 5 accordingly. The estimated value of PSNR is tabulated in Table 1 for Cameraman and Table 2 for Lena from which we can easily distinguish the prominence of the proposed method. The comparisons of different median filtering schemes with the offered technique are shown in the graphs in Figure 6 and Figure 7.
(a) (b)
Fig.3:(a)Original image of Cameraman,(b)Original Image of Lena.
Output obtained by using different filters
Noise Levels
10% 20% 30% 40% 50% 60% 70% 80%
CWM 34.0704 27.5984 23.5606 20.6747 18.3456 16.3838 14.7059 13.1256 Median 34.3729 27.7820 23.7816 20.9856 18.6689 16.7341 14.9706 13.2653 MDB 34.4074 27.8356 23.8518 21.0749 18.7612 16.8068 15.0665 13.3689 Fuzzy 34.4838 27.8702 23.8716 21.0876 18.7697 16.8127 15.0703 13.3712
(a) (b) (c) (d) (e)
Cameraman image Corrupted by 10% noise
(a) (b) (c) (d) (e)
Cameraman image Corrupted by 30% noise
(a) (b) (c) (d) (e)
Cameraman image Corrupted by 50% noise
(a) (b) (c) (d) (e)
Cameraman image Corrupted by 80% noise
Fig.4-(a) Noisy cameraman image,(b)Output of CWM filter,(c)Output of median Filter,(d)Output of MDB
filter,(e)Output of Proposed fuzzy based filter.
(a) (b) (c) (d) (e) Lena image Corrupted by 10% noise
(a) (b) (c) (d) (e)
Lena image Corrupted by 30% noise
(a) (b) (c) (d) (e)
Lena image Corrupted by 50% noise
(a) (b) (c) (d) (e)
Lena image Corrupted by 80% noise
Fig.5-(a) Noisy Lena image,(b)Output of CWM filter,(c)Output of median Filter,(d)Output of MDB filter,(e)Output of Proposed fuzzy based filter.
Output obtained by using different filters
Noise Levels
10% 20% 30% 40% 50% 60% 70% 80%
CWM 35.5064 28.6780 2.4252 21.4766 18.9458 16.9171 15.1398 13.3418 Median 35.7050 28.9014 24.7793 21.9496 19.5421 17.4858 15.6039 13.6232 MDB 35.7499 28.9483 24.8486 22.0331 19.6121 17.5784 15.6985 13.7142 Fuzzy 35.8172 28.9759 24.6821 22.0393 19.6146 17.5793 15.6987 13.7142
Table 2:Comparison Table for PSNR values of Lena at various technique with different noise densities .
10 20 30 40 50 60 70 80 10
15 20 25 30 35
noise
psnr
comparision of %salt and pepper noise vs. PSNR
psnr=cwm psnr=median psnr=mdb psnr=fuzzy
Fig.6: Comparison of PSNR vs. salt and pepper noise of Cameraman at various noise densities.
10 20 30 40 50 60 70 80
10 15 20 25 30 35 40
noise
psnr
comparision of %salt and pepper noise vs. PSNR
psnr=cwm psnr=median psnr=mdb psnr=fuzzy
Fig.7: Comparison of PSNR vs. salt and pepper noise of Lena at various noise densities.
IV .CONCLUSION
Filtering effect becomes appreciable with higher PSNR values. For the images the subjective analysis of the image depicts the quality of the image. In this research work, the PSNR values for the proposed filtering scheme was found to be higher than the other traditional methods and it was also found the images have better quality when analyzed subjectively with respect to other denoising methods.
V. REFERENCES
[1] S. K. Satpathy, S. Panda, K. K. Nagwanshi and C. Ardil “Image Restoration in Non-Linear Filtering Domain using MDB approach”
,International Journal of Information and Communication Engineering Volume 6,Issue 1 2010,pp45-49.
[2] Aswini Kumar Samantaray and Priyadarshi Kanungo “First order neighborhood decision based median filter” 2012 World Congress on Information and Communication Technologies Vol.6, pp 785-789.
[3] Bhavana Deshpande, H.K. Verma & Prachi Deshpande “Fuzzy based median filtering for removal of salt & pepper noise” International journal of soft computing &
Engineering,IJSCE,ISSN:2231-2307,Volume- 2,Issue-3,pp 76-80,July 2012.
[4] R.Pushpavalli & G.Sivarajde, “A Fuzzy Switching Median Filter for Highly Corrupted Images” International journal of Science and Research Publications ,ISSN 2250-3156,Volume- 3,Issue-6.pp.1-6,June 2013.
[5] Isavani Perumal.P&Murugappriya.S,
“Implementation of Cluster based Adaptive Fuzzy Switching Median Filter for Impulse Noise Removal”, IJMER, ISSN: 2249-6645, Volume- 2.Issue-3, pp 1306-1309,May-June 2012.
[6] M.Suneel, K.Samba Siva Rao, M.Lavanya &
M.Sai Sasanka “Fuzzy Enhancement Technique using S-membership Function in Medical applications”, IJECE, ISSN: 2278-9901, Volume- 2, Issue-2, pp.121-126, May 2013.
[7] CharuKhare and Kapil Kumar Nagwanshi
“Image Restoration Technique with Non Linear Filter” International Journal of Advanced Science and Technology Vol. 39, February.
[8] Shanmugavadivu P1 and Eliahim Jeevaraj P S2 “ Adaptive PDE based median filter for the restoration of high density impulse noise corrupted images” International Journal of Advanced Information Technology (IJAIT) Vol.
1, No.6, December 2011,pp 43-51.
[9] J.Sorubamarcel, A.Jayachandran, G.Kharmega &
Sundararaj, “An efficient algorithm for removal of impulse noise using adaptive fuzzy switching weighted median filter”, IJCTEE,ISSN 2249- 6343,Volume-2,Issue-2,pp 1-8,2012.
[10] Roli Bansal, Priti Sehgal & Punam Bedi, “A Simplified Fuzzy Filter for Impulse Noise Removal using Thresholding”,WCECS,2007 October pp24-26, San Francisco, USA.
[11] Rafael C. Gonzalez& Richard E. Woods, “Digital Image Processing”, Pearson,3rd Edition,2009.
[12] S. N. Sivanandam& S. N. Deepa, “Principles of Soft Computing”, Wiley India, Second Edition, 2011.