
Research Article


Salt and Pepper Noise Removal Algorithm by Novel Morpho Filter 

K. Priya
and
D. Pugazhenthi



ABSTRACT

In this study, a new decision based morpho filter is proposed for denoising images that are highly corrupted images by salt and pepper noise. The main problem of denoising is how to keep the poise between degrading image noise and preserving image edge information. Hence, the main aim is to construct a denoising algorithm which not only eliminate the noises but also preserves image edge information. The algorithm replaces the noisy pixels by morphological operations. Experiments are carried out on benchmark images such as Lena, Barbara, Baboon and peppers. A competitive denoising is achieved in comparison with Standard Median Filter (SMF), Adaptive Median Filter (AMF) and Decision Based Algorithm (DBA).





Received: November 15, 2013;
Accepted: February 14, 2014;
Published: March 22, 2014


INTRODUCTION
Image denoising refers to the recovery of a digital image that has been affected by noises. The digital image can be affected by different types of noises. They are salt and pepper noise (Impulse noise), Poisson noise and Additive White Gaussian Noise (AWGN). The Standard Median Filter (SMF) is a non linear digital filter widely used in image processing. SMF is able to remove impulse noise as well as preserve the edges. However, the main drawback is it works only for low noise densities. At high noise densities, the image gets blurred and insufficient noise suppression for small window sizes (Pitas and Venetsanopoulos, 1990; PomalazaRacz and Macgillem, 1984). To overcome this, noise detection process is introduEsakkirajanced in Adaptive Median Filter (AMF) (Zhang and Karim, 2002), Decision Based Algorithm (DBA) (Florencio and Schafer, 1994) and switching median filters (Eng and Ma, 2001).
A novel OpenClose Sequence (OCS) filter to remove impulse noise in highly corrupted images based on mathematical morphology is presented (Deng et al., 2007). The morphological residue detector powerfully determinates the impulse noise with a low percentage error. The OCS filters effectively remove high probability impulse noises. A new concept in impulse noise detection and elimination through primary implicant elimination is developed (Agaian et al., 2008). The filtering algorithms are implemented based on logical transform to detect and eliminate the impulse noise.
A highly effective switchingbased adaptive weighted mean filter for removing impulse noise from the corrupted image is implemented (Zhang and Xiong, 2009). The directional difference based noise detector can realize accurate noise detection, thus facilitating the prevention of image degradation resulting from the undetected noise pixels and misidentified noisefree pixels. A modified decision based unsymmetric trimmed median filter for high density salt and pepper noise removal is implemented (Esakkirajan et al., 2011) that replaces the noisy pixel by trimmed median value when other pixel values, 0’sand 255’s are present in the selected window and when all the pixel values are 0’and 255’sec then the noise pixel is replaced by mean value of all the elements present in the selected window.
A new Fuzzy Switching Median filter (FSM) employing fuzzy techniques to denoise the corrupted image is developed (Toh and Isa, 2010). This mechanism is an extension to the classical switching median filter by employing fuzzy inference mechanism. This filter is able to remove saltand pepper noise in digital images while preserving image details and textures very well. By incorporating fuzzy reasoning in correcting the detected noisy pixel, the low complexity FSM filter is able to outperform some well known existing saltand pepper noise fuzzy and classical filters.
A novel twostage noise adaptive fuzzy switching median filter for saltandpepper noise detection and removal is presented (Toh et al., 2008). This filter does not require any further tuning or training of parameters once optimized. It is able to yield good filtering results with efficient processing time. An effective and accurate algorithm for impulse noise detection is presented (Duan and Zhang, 2010), which consists of two iterations to make the decision as accurate as possible. Tworobust and reliable decision criteria are used for each iteration.
PROPOSED ALGORITHM
The groundwork in the proposed morpho filter is the detection of noisy pixels. The impulse corrupted pixels can take minimum (0) or maximum (255) intensity value. Hence the noisy pixels are identified by checking the pixel intensity values. If the processed pixel has 0 or 255, then the proposed morpho filter is applied to denoise the processed pixel.
The proposed algorithm replace the noisy pixels by using the morphological dilation filtering technique in which 3x3 structure element is used. Each dilation eight neighbors of the Noisy Pixel (NP) are scanned and tested. If all the neighbor pixels are uncorrupted pixels, then the median value of the neighbor pixel is used to remove the NP. If any neighbor pixels that are found noisy pixels then later dilation are applied to remove the NP. The median value of the uncorrupted neighbor pixel and the uncorrupted pixels from the later dilation is used to remove the NP. Figure 1 shows the position of the dilation pattern applied for the corrupted pixel and Fig. 2 shows the later dilation applied for neighboring pixels.
The proposed algorithm keeps track of noisy pixels in the scan order of left to right and top to bottom. The proposed algorithm is as follows :
Step 1: 
If the processed pixel N(X, Y) is noisy pixel, then the noisy pixel is replaced by following Step 2. Otherwise the next pixel is considered for the test 
Step 2: 
The dilation pattern for the corrupted pixel is identified by comparing a 3x3 window centered on the noisy pixel with dilation pattern shown in Fig. 1. The numerals in Fig. 1 denote the corresponding later dilation pattern 
Case 1: 
If all the pixels values corresponding to their later dilation pattern are uncorrupted, then the noisy center pixel N (X, Y) is replaced by the median value of the pixels considered 
Case 2: 
If any pixel found noisy, then later dilation is applied based on their corresponding numerals in Fig. 1. Except the black shaded portions in Fig. 2 are the expansion pixels of specific later dilation. The expansion pixels are found for all noisy pixels. The processed pixel N (X, Y) is replaced by the median of all expansion pixels with uncorrupted pixels in the 3x3 window 
Step 3: 
Steps 1 to 2 are repeated until all noisy pixels are denoised 
 Fig. 1:  Dilation pattern for corrupted pixel 
 Fig. 2: 
Later dilation pattern for neighbor pixels 
EXPERIMENTAL RESULTS
The denoising performance of the proposed system is quantitatively evaluated by using Peak Signal to Noise Ratio (PSNR), Image Enhancement Factor (IEF) and Mean Squared Error (MSE). First, salt and pepper noise of variance from 0.10.9 is added to the standard benchmark image. Then theses images are filtered by using the proposed decision based morpho filter and the performance measures are calculated. Also the performance of the proposed method is compared with other methods such as Standard Median Filter (SMF), AMF and DBA.
The PSNR is the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. The PSNR is most commonly used as a measure of quality of reconstruction in image enhancement. It is most easily defined via the root mean squared error (RMSE) which for two images f (x, y) and considering one of images as a noisy approximation of the other. It is defined as:
The PSNR is defined as:
Table 1: 
PSNR, IEF and MSE for various algorithms for Lena image 

Table 2: 
PSNR, IEF and MSE for various algorithms for pepper image 


Fig. 3(ab): 
Simulation results of different algorithms for 30% noise added to (a) Lena image: Noisy image, median filter, AMF, DBA, proposed approach and (b) Pepper image, noisy image median filter, AMF, DBA, proposed approach 
Here, peak to peak value of the referenced image is the maximum pixel value of the image. When the pixels are represented using 8 bits per sample, this is 255. The peak signal to noise ratio is calculated from the error using the above equation. The higher the value of the PSNR, the better is the performance of that particular local operator for the noise added. The image enhancement factor is defined by:

where, noisy is the corrupted image, f is the input image and is the denoised image. The PSNR, IEF and MSE values for different methods are shown in Table 1 and 2 for Lena and pepper image, respectively. From the table, it is observed that the PSNR of the proposed decision based morpho filter is higher than other methods. Figure 3 and 4 shows the simulation results of different algorithms for 30 and 70% noise added to Lena and pepper image, respectively. Figure 5 shows the simulation results of noise density vs. PSNR for different algorithms. Also the simulation results of noise density vs. IEF and noise density vs. MSE for different algorithms is shown in Fig. 6 and 7, respectively.

Fig. 4(ab): 
Simulation results of different algorithms for 70% noise added to (a) Lena image: Noisy image, median filter, AMF, DBA, proposed approach and (b) Pepper image, noisy image, median filter, AMF, DBA, proposed approach 
 Fig. 5: 
Simulation results of noise density versus PSNR of Lena for different algorithms 
 Fig. 6: 
Simulation results of noise density versus MSE of Lena for different algorithms 
From the Fig. 6, it is noted that IEF of the proposed morpho filter is in between 4 and 5 irrespective of noise density and the IEF of others are decreases as noise density increases.
It is observed from the Table 1 and 2, the performance of the proposed approach is better than SMF, AMF and DBA methods. The PSNR of the proposed approach is approximately minimum 4dB higher than SMF, AMF and DBA methods irrespective of the density of noise present in the given image. Also, the MSE of the proposed approach is lower compared to others.
 Fig. 7: 
Simulation results of noise density versus IEF of Lena for different algorithms 
The IEF of the proposed approach is higher than AMF and DBA based approach for those images corrupted by higher density salt and pepper noise.
CONCLUSION
In this study, a decision based morpho filter is developed for denoising images that are corrupted by salt and pepper noise. This approach uses morphological dilation to denoise the noisy pixels. The salt and pepper noise with densities 1090% is added to the input image and then the noisy pixels are removed by the proposed approach. The proposed filter is tested with benchmark images such as Lena, pepper images. To demonstrate the performance of the proposed approach, three performance metrics PSNR, MSE and IEF of the proposed approach is compared with other stateofart techniques such as SMF, AMF and DBA. The simulations result shows that the proposed filtering scheme has a very satisfactory denoising property as well as edge and detail preserving at very high noise densities.

REFERENCES 
1: Pitas, I. and A.N. Venetsanopoulos, 1990. Nonlinear Digital Filters: Principles and Applications. Vol. 84, Springer, USA., ISBN: 9780792390497, Pages: 392.
2: PomalazaRacz, C.A. and C.D. Macgillem, 1984. An adaptive non linear edge preserving filter. IEEE Trans. Acoust. Speech Signal Process., 32: 571576.
3: Zhang, S. and M.A. Karim, 2002. A new impulse detector for switching median filters. J. IEEE Signal Process. Lett., 9: 360363. CrossRef  Direct Link 
4: Eng, H.L. and K.K. Ma, 2001. Noise adaptive softswitching median filter. IEEE Trans. Image Process., 10: 242251. CrossRef  Direct Link 
5: Florencio, D.A. and R.W. Schafer, 1994. Decisionbased median filter using local signal statistics. Proceedings of the SPIE Visual Communications and Image Processing, Volume 2308, September 2529, 1994, Chicago, Illinois, pp: 268275.
6: Deng, Z.F., Z.P. Yin and Y.L. Xiong, 2007. High probability impulse noiseremoving algorithm based on mathematical morphology. IEEE Signal Process. Lett., 14: 3134. CrossRef 
7: Agaian, S.S., E.E. Danahy and K.A. Panetta, 2008. Logical system representation of images and removal of impulse noise. IEEE Trans. Syst. Man Cybern. Part A: Syst. Humans, 38: 13491362. CrossRef 
8: Zhang, X. and Y. Xiong, 2009. Impulse noise removal using directional difference based noise detector and adaptive weighted mean filter. IEEE Signal Process. Lett., 16: 295298. CrossRef 
9: Esakkirajan, S., T. Veerakumar, A.N. Subramanyam and C.H. PremChand, 2011. Removal of high density salt and pepper noise through modified decision based unsymmetric trimmed median filter. IEEE Signal Proc. Lett., 18: 287290. CrossRef  Direct Link 
10: Toh, K.K.V., H. Ibrahim and M.N. Mahyuddin, 2008. Saltandpepper noise detection and reduction using fuzzy switching median filter. IEEE Trans. Consum. Electron., 54: 19561961. CrossRef 
11: Toh, K.K.V. and N.A.M. Isa, 2010. Noise adaptive fuzzy switching median filter for saltandpepper noise reduction. IEEE Signal Process. Lett., 17: 281284. CrossRef 
12: Duan, F. and Y.J. Zhang, 2010. A highly effective impulse noise detection algorithm for switching median filters. IEEE Signal Process. Lett., 17: 647650. CrossRef 



