 Research Article
 Open Access
Impulse Noise Filtering Using Robust PixelWise SEstimate of Variance
 Vladimir Crnojević^{1} and
 NemanjaI Petrović^{1}Email author
https://doi.org/10.1155/2010/830702
© V. Crnojević and N. I. Petrović. 2010
 Received: 31 December 2009
 Accepted: 8 April 2010
 Published: 16 May 2010
Abstract
A novel method for impulse noise suppression in images, based on the pixelwise Sestimator, is introduced. The Sestimator is an alternative for the wellknown robust estimate of variance MAD, which does not require a location estimate and hence is more appropriate for asymmetric distributions, frequently encountered in transient regions of the image. The proposed computationally efficient modification of a robust Sestimator of variance is successfully utilized in iterative scheme for impulse noise filtering. Another novelty is that the proposed iterative algorithm has automatic stopping criteria, also based on the pixelwise Sestimator. Performances of the proposed filter are independent of the image content or noise concentration. The proposed filter outperforms all stateoftheart filters included in a large comparison, both objectively (in terms of PSNR and MSSIM) and subjectively.
Keywords
 Decision Boundary
 Impulse Noise
 Noisy Pixel
 Corrupted Pixel
 Noise Concentration
1. Introduction
Corruption by the impulse noise is a frequent problem which appears in digital images. It occurs as a consequence of transmission errors, timing problems in analogtodigital conversion, or damaged pixel elements in image sensors [1]. Regardless of its origin, the impulse noise has two important aspects: only certain parts of the image pixels are corrupted by the noise and the intensities of contaminated pixels are significantly different from the other noisefree pixels in their neighborhoods. These properties can easily make any kind of subsequent processing, such as segmentation, edge detection, or object recognition, difficult or even impossible. Therefore, the suppression of the impulse noise is usually a required preprocessing step.
The major issue in impulse noise suppression is to satisfy two opposing requests. The corrupted pixels should be filtered whereas the image details have to be preserved. This task is exceptionally difficult because even the smallest amount of noise impulses which are not detected and filtered causes significant deterioration of image quality due to the nature of impulse noise. There have been proposed a large number of filtering techniques for removal of impulse noise. The classical approach is based on using median or its modifications [2]. These spaceinvariant methods are applied uniformly throughout the whole image, that is, apart from the noisy pixels, they unnecessarily change the noisefree pixels and impair image details. Most of the modern impulse noise filters utilize the solution based on switching scheme [3]. The noisy pixels are detected first and filtered whereas the noisefree pixels are left intact. Thus, this approach is space variant, and it is proven to be effective in preserving image details.
The impulse noise detection is usually performed by comparison of some robust statistics calculated in a local neighborhood to the corresponding fixed or adaptively calculated thresholds. A plethora of the algorithms has been developed which uses this approach, for example, switching median (SM) filter [3], threestate median (TSM) filter [4], multistate median (MSM) [5], adaptive center weighted median (ACWM) filter [6], statedependent rankorder mean (SDROM) filter [7], progressive switching median (PSM) filter [8], conditional signaladaptive median (CSAM) filter [9], pixelwise MAD (PWMAD) filter [10], threshold boolean filter (TBF) [11], and so forth. The detectors of the previous filters are constructed heuristically, but it is also possible to use previous knowledge and machine learning techniques in order to find an optimal decision rule. Genetic programming is utilized in GP [12–14] filters, and neural networks are employed in improved adaptive impulsive noise suppression (IAINS) filter [15].
The other popular approach relies on fuzzy logic. Impulse noise detection in fuzzybased techniques models ambiguities between noisy impulses and image structures in order to preserve image details [16–18]. Further enhancement of the fuzzy techniques is achieved by combining them with neural networks into neurofuzzy systems [19].
Robust statistics play a central role in impulse detection, being capable of producing correct estimates in the presence of unreliable data. The most frequently used statistics are the median and its variants such as centerweighted median. Nevertheless, robust statistics based on absolute differences are proven to be successful. The trilateral filter [20] was the first one which employed rankordered absolute difference (ROAD) statistics. Effective modifications are given by the rankordered logarithmic difference (ROLD) detector [21] and rankorderedrelative difference (RORD) detector [22]. A slightly different technique, which is also based on absolute differences but makes use of directional information, is employed in the directional weighted median (DWM) filter [23].
After the detection of noisy impulses, selective, spacevariant estimation is applied. The classical approach is based on the utilization of robust estimates of location, but recently an edgepreserving regularization method emerged as an alternative. This method was applied for the first time in the detailpreserving variational method (DPVM) [24]. However, it was applied uniformly for all the pixels in the image, which resulted in relatively moderate results. This estimation method showed much better results when it was combined with the impulse detection scheme in the adaptive centerweighted median with edgepreserving regularization: (ACWMEPR) filter [25] or ROLDEPR filter [21].
In this paper we concentrate on denoising images corrupted by the mixture of saltandpepper and randomvalued impulse noises. It has already been shown with the DUMMY filter [14] that detection of pure saltandpepper impulse noises is almost a trivial problem. Therefore, improvements in saltandpepper noise filtering are directed either toward efficient implementation, like in a decisionbased algorithm (DBA) filter [26] or toward better estimation, implemented by a fuzzy impulse noise detection and reduction method (FIDRM) [27], switchingbased adaptive weighted mean (SAWM) filter [28], and the edgepreserving (EP) filter [29]. The more challenging impulse noise model is the randomvalued model, and the filters designed to remove it are, in general, capable of treating saltandpepper noise as well. However, there exist some filters for which that is not the case [13, 20]. Therefore, it is an important property of impulse noise filters that they are capable of suppressing both types of noise equally well. In order to evaluate the overall performance of impulse noise filters, the mixed impulse noise model was proposed in [14]. The same impulse noise model is used in this paper.
In Section 2 the assumed noise model is described. Section 3 introduces a pixelwise Sestimate of variance, and in Section 4 the proposed method for impulse noise suppression based on usage of this robust estimate is presented. The results are given in Section 5.
2. Noise Model
where and denote noise impulses generated according to randomvalued and saltandpepper impulse noises, respectively. A noisefree image pixel at location is symbolized by . If and are the minimum and maximum values from the dynamic range of pixel values in the image, then and . In this way, half of the noisy pixels are corrupted by the randomvalued impulse noise and the other half by saltandpepper noise. The reason for using the mixed impulse noise model is twofold. Firstly, the mixed impulse noise model is more realistic than existing models. The impulse noise is a result of disturbances caused by noise signals with random amplitudes. The amplitude of the noise signal can fall either into the dynamic range or out of that range. If it is out of the range, the corrupted pixel in the resulting noisy image will be saturated to the maximal or minimal value of the dynamic range, and that situation corresponds to the saltandpepper model. Alternatively, if the impulse noise is within the dynamic range, it will appear as randomvalued impulse noise in the noisy image. Secondly, it is expected from the highquality impulse noise filter to perform well in the presence of both saltandpepper and randomvalued models. Since the filters should handle both impulse noise types equally, it is reasonable to choose a percent ratio for testing. Accordingly, the mixed impulse noise model represents the model which is suitable for proper evaluation of the impulse noise filters.
3. PixelWise Sestimate
where , is the sample containing elements. Firstly, for each the inner median of is calculated. This yields a new sample of elements, and their median (the outer median) gives the final estimate S. A beneficial property of the Sestimate is that it does not rely on a calculation of the location estimate, and consequently it produces an accurate estimate of variance for nonsymmetric sample distributions, that is, the image regions containing edges. The major drawback of the Sestimate is its computational complexity. For a sample of elements, it is necessary to calculate the median exactly times. This could be a serious problem in image processing, where the size of the sample is usually the square of the filtering window dimension and it should be calculated for each pixel.
where is high median value, which is the order statistic of rank . Since the differences are calculated with respect to the central pixel, MAd is calculated on the set for the sake of more efficient implementation.
4. Proposed Method
4.1. Detection
where and denote the statistics calculated in the th iteration for the pixel at position . The values of the constants and depend on the iteration and they are determined experimentally.
The same recursive approach is used when and values are calculated. Instead of the input values , the initial estimates are always used if they are already calculated. The initial estimates are later replaced by final estimates obtained by edgepreserving regularization. The details about the final estimation are given in Section 4.2.
We denote the proposed detection scheme PWS detector since the noise classifier is based on PWS and MAd statistics, but MAd is actually calculated within the calculation of PWS. The detector parameters and are determined experimentally. Parameter defines the slope of the classification line in the PWS/MAd plane, defined by (7), whereas defines the offset from the origin. It has been found through experimentation with images having 8 bits per pixel that the optimal value of should be in the range . In our simulations we set it to be constant in every iteration and to be at the middle of this range, that is, . On the other hand, the slope parameter should change throughout the iterations. In the beginning it should be in the range and it should become smaller in subsequent iterations, in order to become less conservative and allow detection of more noise. We reduce this parameter for a constant factor in each iteration . These parameters are very robust and produce satisfactory results for different noise levels and different images.
It has been experimentally verified that the proposed detection procedure is robust to the variation of parameters and . However, the quality of the output is mainly influenced by the optimal number of iterations , because iterative filtering has to be stopped before it starts to severely destroy image details. This issue is discussed and handled by the algorithm described in Section 4.3.
The last parameter, which is the only one that actually has to be set manually, is the window size , defined in (3). We follow the rules given in [20]: if the noise ratio is higher than , we use window; otherwise the window of the size is applied. This yields satisfactory performance in most cases.
4.2. Estimation
Many spacevariant filtering methods designed for suppression of the impulse noise use estimators based on the median and its derivatives. We utilize this kind of estimate just as a first approximation during the iterative impulse detection. The final estimate is found by replacing the noisy pixels with values found through the procedure of edgepreserving regularization (EPR) similar to [24]. We combine the proposed PWS detector with edgepreserving regularization and denote it as PWSEPR filter.
similarly as in [25], where Note that in contrast to [25] or [32] the functional in (12) contains only a regularization term and not the data term. This is because the data are fitted exactly for the uncorrupted pixels, while for the corrupted pixels it is expected that the difference between observed end estimated values will be large due to the nature of the impulse noise.
A global minimization of the functional is very difficult. Therefore, we apply the optimization procedure for each pixel separately and repeat the procedure iteratively across the whole image until the process converges to the stable solution. Since the edgeregularization potential function is strictly convex, we perform the local optimization by Brent optimization method [33]. The same convergence criteria as in [34] are applied.
4.3. Stopping Criteria
The important issue which is common for iterative filtering approaches is to determine the optimal number of iterations. Most of the stateoftheart algorithms set that parameter to some fixed value which gives satisfactory results in most cases [14, 21, 25] or set it according to noise concentration [20, 23]. Still, this is a challenging problem because the optimal number of iterations usually depends on both image content and noise concentration. The proposed PWSEPR filter in each iteration calculates the PWS estimate of the variance, and this is utilized to determine the stopping criteria for the iterative filtering.
where the summation goes over all pixel coordinates, is the number of pixels in the image, and is the iteration number.
5. Results
The performance of the proposed PWSEPR filter was compared to a number of stateoftheart impulse noise filters. The experiments were conducted on standard test images and for different concentrations of mixed impulse noise. The parameters of the PWSEPR filter were set to the constant experimentally found values as explained in Section 4. The only parameter which needs to be tuned with respect to the noise level is the window size used by the detector. It is set to be when noise concentration % and otherwise. The other parameters were set as follows: , , and . In all experiments, the parameters were kept constant, that is, they were not tuned for a particular image or noise level. The optimal number of iterations is calculated dynamically, as explained in Section 4.3.
Filtering results in PSNR (dB) for images corrupted with mixed impulse noise.
Lena  Goldhill  Boats  Bridge  

Methods 












Noisy  13.95  10.91  9.14  13.82  10.81  9.04  13.76  10.75  8.99  13.61  10.63  8.88 
MED3 3  30.97  27.05  22.66  29.26  26.83  22.92  28.64  25.55  21.98  24.61  22.68  20.06 
TSM  33.85  26.32  19.79  31.86  25.99  19.68  31.15  25.18  19.34  26.93  23.11  18.36 
ACWM  34.72  28.86  23.09  32.52  28.37  22.94  31.78  27.04  22.08  27.17  23.96  20.24 
SDROM  34.74  28.87  23.47  32.97  28.39  23.26  31.85  27.39  22.48  27.56  24.25  20.57 
PSM  28.96  27.51  25.66  28.83  27.12  25.11  28.89  26.17  23.90  26.84  23.74  21.59 
PWMAD  34.77  28.88  19.59  32.38  27.85  19.22  31.69  26.93  18.94  27.22  23.68  17.80 
Trilateral  33.98  25.68  16.37  32.18  25.40  16.10  31.39  24.70  16.02  27.12  22.38  15.10 
DWM  34.49  30.95  26.49  32.22  29.28  25.84  31.41  28.22  24.57  26.40  24.33  21.78 
FRINR  35.24  30.88  24.96  32.70  29.05  24.51  32.53  27.91  22.67  27.55  24.16  19.39 
GP  35.49  30.73  24.73  32.97  29.55  24.47  32.39  28.40  23.56  27.60  24.83  21.28 
ACWMEPR  35.88  30.50  22.74  33.36  29.12  22.29  32.88  28.49  21.64  26.91  24.56  19.97 
ROLDEPR  35.08  30.66  25.31  33.26  29.90  24.85  32.58  27.72  22.84  28.04  24.28  20.42 
PWSEPR  36.38  31.46  27.23  33.95  30.10  26.78  33.49  28.43  24.58  28.10  24.84  21.97 
Filtering results in MSSIM for images corrupted with mixed impulse noise.
Lena  Goldhill  Boats  Bridge  

Methods 












Noisy  0.115  0.051  0.026  0.132  0.056  0.028  0.145  0.067  0.035  0.226  0.103  0.052 
MED3 3  0.880  0.803  0.641  0.796  0.719  0.561  0.856  0.775  0.610  0.689  0.591  0.443 
TSM  0.944  0.801  0.491  0.919  0.773  0.477  0.928  0.777  0.478  0.861  0.715  0.456 
ACWM  0.958  0.864  0.638  0.926  0.820  0.594  0.946  0.841  0.611  0.870  0.748  0.535 
SDROM  0.951  0.830  0.593  0.928  0.801  0.574  0.941  0.811  0.575  0.884  0.755  0.539 
PSM  0.737  0.673  0.631  0.772  0.692  0.614  0.768  0.668  0.595  0.832  0.711  0.581 
PWMAD  0.958  0.860  0.447  0.923  0.803  0.420  0.942  0.835  0.438  0.868  0.730  0.416 
Trilateral  0.942  0.790  0.384  0.919  0.761  0.372  0.931  0.777  0.388  0.870  0.712  0.370 
DWM  0.949  0.892  0.776  0.920  0.837  0.701  0.934  0.863  0.734  0.843  0.734  0.583 
FRINR  0.948  0.892  0.811  0.916  0.814  0.704  0.943  0.864  0.754  0.874  0.726  0.565 
GP  0.960  0.891  0.685  0.930  0.844  0.639  0.949  0.870  0.661  0.872  0.767  0.572 
ACWMEPR  0.959  0.864  0.573  0.933  0.826  0.554  0.949  0.847  0.555  0.859  0.760  0.531 
ROLDEPR  0.947  0.908  0.791  0.932  0.868  0.720  0.941  0.874  0.727  0.896  0.756  0.571 
PWSEPR  0.963  0.903  0.792  0.944  0.860  0.723  0.957  0.876  0.736  0.902  0.777  0.573 
6. Conclusion
This paper has two major contributions. Firstly, we introduced a pixelwise Sestimator as an effective and computationally efficient robust estimator of variance that can be successfully utilized in an iterative scheme for mixed impulse noise filtering. In addition, we developed a novel method for determining the optimal number of filtering iterations, also based on the pixelwise Sestimator. The proposed PWSEPR filter outperforms other filters included in the comparison, both objectively (in terms of PSNR and MSSIM) and subjectively. Further improvements of the proposed filter are possible since the linear classifier used for noise detection is a rather simple solution. The utilization of a quadratic classifier instead which could provide better separation of noisy and noisefree pixels should further reduce the number of iterations, increase the detector accuracy, and improve the overall filtering performance.
Authors’ Affiliations
References
 Gonzalez RC, Woods RE: Digital Image Processing. Prentice Hall, Upper Saddle River, NJ, USA; 2002.Google Scholar
 Pitas I, Venetsanopoulos AN: Order statistics in digital image processing. Proceedings of the IEEE 1992, 80(12):18931921. 10.1109/5.192071View ArticleGoogle Scholar
 Sun T, Neuvo Y: Detailpreserving median based filters in image processing. Pattern Recognition Letters 1994, 15(4):341347. 10.1016/01678655(94)900825View ArticleGoogle Scholar
 Chen T, Ma KK, Chen LH: Tristate median filter for image denoising. IEEE Transactions on Image Processing 1999, 8(12):18341838. 10.1109/83.806630View ArticleGoogle Scholar
 Chen T, Wu HR: Space variant median filters for the restoration of impulse noise corrupted images. IEEE Transactions on Circuits and Systems II 2001, 48(8):784789. 10.1109/82.959870View ArticleMATHGoogle Scholar
 Chen T, Wu HR: Adaptive impulse detection using centerweighted median filters. IEEE Signal Processing Letters 2001, 8(1):13. 10.1109/97.889633View ArticleGoogle Scholar
 Abreu E, Lightstone M, Mitra SK, Arakawa K: A new efficient approach for the removal of impulse noise from highly corrupted images. IEEE Transactions on Image Processing 1996, 5(6):10121025. 10.1109/83.503916View ArticleGoogle Scholar
 Wang Z, Zhang D: Progressive switching median filter for the removal of impulse noise from highly corrupted images. IEEE Transactions on Circuits and Systems II 1999, 46(1):7880. 10.1109/82.749102View ArticleGoogle Scholar
 Pok G, Liu JC, Nair AS: Selective removal of impulse noise based on homogeneity level information. IEEE Transactions on Image Processing 2003, 12(1):8592. 10.1109/TIP.2002.804278View ArticleGoogle Scholar
 Crnojević V, Šenk V, Trpovski Ž: Advanced impulse detection based on pixelwise MAD. IEEE Signal Processing Letters 2004, 11(7):589592. 10.1109/LSP.2004.830117View ArticleGoogle Scholar
 Aizenberg I, Butakoff C, Paliy D: Impulsive noise removal using threshold boolean filtering based on the impulse detecting functions. IEEE Signal Processing Letters 2005, 12(1):6366.View ArticleGoogle Scholar
 Petrović N, Crnojević V: Impulse noise detection based on robust statistics and genetic programming. In Advanced Concepts for Intelligent Vision Systems, Lecture Notes in Computer Science. Volume 3708. Springer, Berlin, Germany; 2005:643649. 10.1007/11558484_81View ArticleGoogle Scholar
 Petrović NI, Crnojević VS: Evolutionary treestructured filter for impulse noise removal. In Advanced Concepts for Intelligent Vision Systems, Lecture Notes in Computer Science. Volume 4179. Springer, Berlin, Germany; 2006:103113. 10.1007/11864349_10View ArticleGoogle Scholar
 Petrović NI, Crnojević V: Universal impulse noise filter based on genetic programming. IEEE Transactions on Image Processing 2008, 17(7):11091120.MathSciNetView ArticleGoogle Scholar
 Sa PK, Majhi B: An improved adaptive impulsive noise suppression scheme for digital images. International Journal of Electronics and Communications 2010, 64(4):322328. 10.1016/j.aeue.2009.01.005View ArticleGoogle Scholar
 Schulte S, Witte VD, Nachtegael M, der Weken DV, Kerre EE: Fuzzy random impulse noise reduction method. Fuzzy Sets and Systems 2007, 158(3):270283. 10.1016/j.fss.2006.10.010MathSciNetView ArticleGoogle Scholar
 Yildirim MT, Baştürk A, Yųksel ME: Impulse noise removal from digital images by a detailpreserving filter based on type2 fuzzy logic. IEEE Transactions on Fuzzy Systems 2008, 16(4):920928.View ArticleGoogle Scholar
 Chen HC, Wang WJ: Efficient impulse noise reduction via local directional gradients and fuzzy logic. Fuzzy Sets and Systems 2009, 160(13):18411857. 10.1016/j.fss.2008.11.020MathSciNetView ArticleMATHGoogle Scholar
 Yüksel ME: A hybrid neurofuzzy filter for edge preserving restoration of images corrupted by impulse noise. IEEE Transactions on Image Processing 2006, 15(4):928936.View ArticleGoogle Scholar
 Garnett R, Huegerich T, Chui C, He W: A universal noise removal algorithm with an impulse detector. IEEE Transactions on Image Processing 2005, 14(11):17471754.View ArticleGoogle Scholar
 Dong Y, Chan RH, Xu S: A detection statistic for randomvalued impulse noise. IEEE Transactions on Image Processing 2007, 16(4):11121120.MathSciNetView ArticleGoogle Scholar
 Yu H, Zhao L, Wang H: An efficient procedure for removing randomvalued impulse noise in images. IEEE Signal Processing Letters 2008, 15: 922925.View ArticleGoogle Scholar
 Dong Y, Xu S: A new directional weighted median filter for removal of randomvalued impulse noise. IEEE Signal Processing Letters 2007, 14(3):193196.View ArticleGoogle Scholar
 Nikolova M: A variational approach to remove outliers and impulse noise. Journal of Mathematical Imaging and Vision 2004, 20(12):99120.MathSciNetView ArticleGoogle Scholar
 Chan RH, Hu C, Nikolova M: An iterative procedure for removing randomvalued impulse noise. IEEE Signal Processing Letters 2004, 11(12):921924. 10.1109/LSP.2004.838190View ArticleGoogle Scholar
 Srinivasan KS, Ebenezer D: A new fast and efficient decisionbased algorithm for removal of highdensity impulse noises. IEEE Signal Processing Letters 2007, 14(3):189192.View ArticleGoogle Scholar
 Schulte S, Nachtegael M, Witte VD, der Weken DV, Kerre EE: A fuzzy impulse noise detection and reduction method. IEEE Transactions on Image Processing 2006, 15(5):11531162.View ArticleGoogle Scholar
 Zhang X, Xiong Y: Impulse noise removal using directional difference based noise detector and adaptive weighted mean filter. IEEE Signal Processing Letters 2009, 16(4):295298.MathSciNetView ArticleGoogle Scholar
 Chen PY, Lien CY: An efficient edgepreserving algorithm for removal of saltandpepper noise. IEEE Signal Processing Letters 2008, 15: 833836.View ArticleGoogle Scholar
 Rousseeuw PJ, Croux C: Alternatives to the median absolute deviation. Journal of the American Statistical Association 1993, 88(424):12731283. 10.2307/2291267MathSciNetView ArticleMATHGoogle Scholar
 Charbonnier P, BlancFéraud L, Aubert G, Barlaud M: Deterministic edgepreserving regularization in computed imaging. IEEE Transactions on Image Processing 1997, 6(2):298311. 10.1109/83.551699View ArticleGoogle Scholar
 Chan RH, Ho CW, Nikolova M: Saltandpepper noise removal by mediantype noise detectors and detailpreserving regularization. IEEE Transactions on Image Processing 2005, 14(10):14791485.View ArticleGoogle Scholar
 Brent RP: Algorithms for Minimization without Derivatives. Prentice Hall, Englewood Cliffs, NJ, USA; 1973.MATHGoogle Scholar
 Chan RH, Ho CW, Nikolova M: Convergence of Newton's method for a minimization problem in impulse noise removal. Journal of Computational Mathematics 2004, 22(2):168177.MathSciNetMATHGoogle Scholar
 Wang Z, Bovik AC, Sheikh HR, Simoncelli EP: Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 2004, 13(4):600612. 10.1109/TIP.2003.819861View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.