MEDIAN FILTER AND ITS VARIATIONS- APPLICATION TO SICKLE CELL ANEMIA BLOOD SMEAR IMAGES

MEDIAN FILTER AND ITS VARIATIONS- APPLICATION TO SICKLE CELL ANEMIA BLOOD SMEAR IMAGES Aruna N.S. Research Scholar, Electrical Engineering, College of Engineering, Trivandrum, India arunasurendran2006@gmail.com Dr. Hariharan S. Professor, Electrical Engineering, College of Engineering, Trivandrum, India harikerala2001@yahoo.com Abstract: - Median filter is one of the most important and useful filter commonly used in image processing applications. Its operating frequencies lies between that of low pass filters and high pass filters. The performance characteristics of median filters are found to be median value and hence this filter is utilized by many image processing researchers. In many image processing and image analysis work first of all median filtering is performed on images as a pre-processing step to improve the quality of input image. In this paper, the result obtained after the median filtering using various median filters for the blood smear images is also found to be fairly good. The significance of various median filters is highlighted and compared its performance on the basis of peak signal to noise ratio (PSNR) and image enhancement factor (IEF). Variations in noise density are plotted against PSNR value and IEF value to show the stable performance of various median filters. Keywords: Median filter, PSNR, IEF. I. Introduction In earlier times noise detection and noise filtering are the two processes that is carried out for image restoration. This cannot fully restore the image. Filtering is an image processing technique for modifying or enhancing an image. Image processing operations that is implemented with filtering includes smoothing, sharpening and edge enhancement. Filtering enables the operation of a given value of pixel in the output image which will be determined by operating in some algorithm to the value of the pixel in the neighborhood of the input pixel. Filters can be either linear filters or nonlinear filters. Linear filtering can be defined as one in which value of an output pixel is a linear combination of the values of the pixel in the input pixel neighborhood. The output pixels in a linear filter are obtained by summing the weights of neighboring input pixels. The filter is a correlation kernel that can be rotated 180. Examples of linear filter include Low pass filters (LPF) and High Pass Filters (HPF). An example of nonlinear filter is median filter. II. Theory Standard median filter: The standard median filter (SMF) is a nonlinear filter. It is also called as median smoother. It removes the impulse noise by changing the intensity values of the centre pixel of the window with the median of the luminous values of the pixel contained within the window. Even though median filter offers excellent noise removal performance, it removes thin lines & blurred images. Research is going on for the improvement of noise removal which has been discussed in [1-5]. No reduction in contrast of output image, no shift in boundaries, the edges will be minimally degraded and low sensitivity than mean to extreme values are the advantages of median filter. Median filter cannot distinguish fine details of noise, it is expensive and has the tendency to erase lines narrower than half the width of neighborhood pixels are the disadvantages. ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 160

Fig1 shows the different median filters that are used in this work. Median Filter Hybrid Median Filter Weighted Median Filter Trimmed Median Filter Adaptive Median Filter Fig1. Different Median filters Unsymmetric Median Cascaded Median Filter Fuzzy based Median Filter 1. Hybrid median filter It is a nonlinear windowed filter that removes impulse noise while preserving edges. It has a good edge preserving characteristics. For any element of the image which applies median technique several times by varying window shape and then takes the median is the basic idea behind hybrid filter. It preserves edges better than square kernel median filter because it is a three step ranking operation [6]. Data from different special directions are ranked separately. Three median values are calculated.mr is the median of horizontal and vertical R pixels. MD is the median of diagonal D pixels. The filtered value is the median of the two median values and the central pixel C. Edge treating is a problem in this filter. If we place window over an element at the edge, some part of the window will be empty and to fill this gap the signal should be extended. Hybrid median filter can extend image symmetrically. It preserves the corners and other features that are eliminated by the 3x3 and 5x5 median filters with repeated application. The hybrid median filter does not excessively smooth image details and typically provides superior visual quality in the filtered image [7]. The advantage of the hybrid median filter is that due to its adaptive nature, it allows the filter to perform better than the standard median filter. 2. Weighted median filter It is one of the branches of median filter. The operations involved are similar to standard median filters except that WMF has weight associated with each of its filter element [8-11]. This weight corresponds to the number of sample duplications for the calculation of median value. 3. Trimmed median filter There are two types of trimmed median filters. Symmetrical trimmed median filter and unsymmetrical trimmed median filter. Trimmed mean filter is an improved version of mean filter where test pixel is replaced by the trimmed mean value of the window pixels. In order to calculate the trimmed mean value of the pixel, trimming value must be known. All available window pixels are arranged in the ascending order of their intensity values. The first & last pixels are removed from ordered list of pixels & the restoration values for the corrupted pixels are the mean value of the removing pixels [12, 13]. The process of removing the first and last set of pixels from the ordered list is called trimming. 4. Unsymmetrical Trimmed Median Filter: An unsymmetrical trimmed median filter (UTMF) was proposed to overcome the issues with alpha trimmed median filter (ATMF). The chosen window 3x3 issues in un-symmetric trimmed median filter (UTMF) are organized in either increasing or decreasing order. Then the pixel values 0's and 255's (or salt and pepper) are the pixels in the image. Then the median value of the remaining pixel is then taken. This median value is used to filter the input image. This filter is termed trimmed median filter as a result of the ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 161

pixel values 0's and 255's are removed from the selected window [14, 15]. This procedure removes noise in a better method than the ATMF. 5. Adaptive Median Filter: This filter is developed in order to improve the standard median filter. The basic difference between the two filters is that, in the adaptive median filter, the size of the window surrounding each pixel is variable. This variation will be depending on the median of the pixel in the current window. If the median value is an impulse, the size of the window will be expanded [16, 17]. One of the main benefits of this median filter is that repeated application of this filter will not erase any edges or other small structure in the image. 6. Cascaded Median Filter: Without blurring and retains the fine edge details, the modified cascaded filter contains a new kind of decision based median filter which are connected in cascade [18, 19]. This filter replaces the corrupted pixels only with a median value of its neighborhood pixels while the uncorrupted pixels are left unchanged. The UTMF performs un-symmetric trimming of the impulse values and averages the remaining pixels. This will remove only the impulse noise lying at the extreme ends, while the original pixel values are retained. 7. Fuzzy based median filter: For preserving local details of the image, median filter change the intensity of corrupted pixels on the damaged image. It is difficult to detect the corrupted pixel from the image [20-22]. For fixed value impulse noise, simple thresholding method cannot classify the pixels accurately. This is because some of the uncorrupted pixels are also been presented by the 0 and L-1 values. Thus researchers such as [23-28] incorporate fuzzy logic approach into median filtering process. There are several ways fuzzy logic can be used in median filtering process. Based on the fuzzy degradation measure, a proper correction will be applied. Some of the methods use fuzzy logic as a decision maker that selects a proper filter, from a filter bank, for a given input image. In order to use fuzzy logic, the damaged image must undergo a fuzzification process. The input for the fuzzification process is the intensity of the pixels with its surrounding. The system then executes the noise filtering process based on the fuzziness values obtained through a de-fuzzification process. Fuzzy logic can improve impulse noise suppression methods such as [27, 28] use too many fuzzy rules to obtain an acceptable result. The restoration results are also too dependent to the number of membership functions, and also to the parameters that control the shape of the membership functions [29]. Therefore such methods are difficult to implement as an automatic impulse noise reduction filter, and also cannot be used for real-time processing. III. Results and Discussion The images are acquired through camera attached microscope which is interfaced with computer will display the images on the screen. The acquired blood samples are used for preprocessing before performing the digital processing of images. Noises seriously affect the processing of images. The noises cannot be erased completely. One of the noises that degrade the digital images is impulse noise. The impulse noise which is a set of random pixels and have high contrast compared to its surroundings. This appears as bright and dark sprinkled spots on the image. Thus it can degrade the quality and appearance of the image. One of the best choices among the researchers for the removal of noise is the median filter. As the median filter replaces the center pixel with a value that is median of all pixels in a window will reduce the high and low intensity value pixels which will improve the clarity of the image. Median filters are non-linear filters and it works in spatial domain. There are ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 162

various types of median filters which are applied in this work to remove noises. Figure 2 shows the original blood smear image. Fig4. Standard median filtered image Fig2. Original image The standard median filter will try to remove impulse noise by changing the luminance value of centre pixel with median of luminance value of the pixels within a window. Thus it removes thin lines and blurs the image at low noise densities. As this filter cannot differentiate between uncorrupted and corrupted pixels, it will alter the level of corruption in both pixels. The local histogram used for median value calculation gives a filtered image which removes more noise pixels. Salt and pepper added image is shown in fig 3. Removal of salt and pepper noise by applying standard filter is shown in fig 4. The original blood smear image may consist of Gaussian noise and impulse noise which can be easily removed using hybrid median filters. In this paper the hybrid filter uses a topological approach. The topological approach maintains the sharpness of edges. The salt and pepper added to the image before filtering is shown in fig 5. The microscopic image that consists of Gaussian and impulse noise are removed and the result is shown in fig 6. Fig5. Salt and pepper applied with 10% ND Fig3. Salt and pepper applied with 10% ND Fig6. Hybrid median filtered image ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 163

A variation in median filter is weighted The gray scale image of the original median filter. Here a weight is associated with image is shown in fig 9 and the output of the each of the filter elements which will correspond trimmed image is shown in fig 10. to a number of duplication of the sample for calculating the median value. The filter weight will give more emphasis to the central pixel by decreasing the weight set when located little distant from the center of filtering window, hence improving the noise suppression and maintaining image details. Median filtered image is shown in fig 7 and a blurred image is obtained with weighted median filter which is shown in fig 8. Fig9. Grayscale image Fig7. Median filtered image Fig8. Weighted median filtered image As the trimmed median filter is a two stage filter, it will detect the noisy pixels first and then restores them. It will not change the noise free pixels during the restoration stage. If all the pixels in selected window contain noise, then the centre pixel will be replaced by mean of the window in restored image or if the selected window contains noisy and noise free pixels, then the centre pixel will be replaced using the median of noise free pixels in the window. Fig10. Trimmed median filtered image Adaptive filter eliminates the problem in standard median filter. Hence we choose adaptive median filter in order to show the difference that can bring to have more information from the image. The size of window located around each pixel is variable for adaptive median filter. The type of median value determines the size of window. If it is impulse value then the window is expanded otherwise the process is similar to standard median filter. The output of adaptive median filter is shown in fig 11. ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 164

Fig11. Adaptive median filtered image Un-symmetric median filter restores the gray scale images which are corrupted by high density salt and pepper noise. As already mentioned in theory of un-symmetric median filter, this filter replaces the noisy pixels by trimmed median values. It selects a window and sees whether noisy pixels with 0 s and 255 s, then those values are replaced by mean values of all pixels in the selected window. Here the elements in the selected window will be arranged decreasing or increasing order and noisy pixels will be removed using median values. This method is more efficient than all other types of median filters. The output of un-symmetric median filter is shown in fig 12. Fig13. Image applied with Un-symmetric median filter The intensity of the corrupted images should only be changed in order to preserve the local details of the images by the median filter. Thus fuzzy logic using fuzzification process has been incorporated in order to take a correct decision for choosing the effective filter that can be applied to a particular image. The original image under gone noise application and fuzzy logic filtering gives an output which is shown in fig 14 and fig 15 respectively. Fig12. Image applied with Un-symmetric median filter Filters are cascaded to get better performance in filtering. This method uses direction based difference method for the calculation of restoration value for the pixels which are noisy. The selected window will keep on expanding in size and grows until it finds that there are no noisy pixels in that window. The filtered image using cascaded filter is shown in fig 13. Fig14. Salt and pepper applied with 10% ND Fig15. Fuzzy based median filtered image ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 165

Different Sickle cell affected blood smear 2 nij rij images are used for filtering using the variations IEF of median filters. The Root Mean Square Error x 2 ij rij (3) (RMSE), Peak Signal-to-Noise Ratio (PSNR) where n- Corrupted image, r- Original image, and Image Enhancement Factor (IEF) are used to m,n- Size of image, x- Restored image. evaluate the enhancement of medical images. Various noise density levels are used to verify different types of filters. The performance 2 r ij xij evaluation of different filters that is calculated RMSE using PSNR and IEF are shown in the table 1 mn and table 2. (1) PSNR 20log 10 255 RMSE (2) Table1. Performance evaluation using PSNR values Noise Density (%) Standard Hybrid Weighted Trimmed Adaptive Cascaded Unsymmetric Fuzzy 10 38.97 37.65 36.28 45.49 42.82 19.59 43.19 37.25 20 32.02 36.66 36.44 42.31 40.48 19.59 41.31 33.02 30 24.87 33.53 34.93 40.31 38.04 19.2 41.31 25.57 40 19.58 27.11 30.49 38.43 35.92 19.1 37.43 16.88 50 15.66 20.99 24.04 35.64 34.05 18.52 36.14 13.33 60 12.67 16.14 18.24 31.26 30.56 18.31 30.67 11.67 70 10.24 12.28 13.69 26.22 23.42 18.1 25.85 10.74 80 8.34 9.39 10.14 21.33 16.36 17.4 21.54 7.34 90 6.82 7.22 7.54 16.83 10.59 10.65 14.11 5.82 Table 2. Performance evaluation using IEF values Noise Density (%) Standard Hybrid Weighted Trimmed Adaptive Cascaded Unsymmetric Fuzzy 10 217.22 160.28 116.12 971.59 525.29 2.49 771.90 117.22 20 87.57 255.35 242.53 932.37 613.56 5 730.37 85.57 30 25.30 185.97 256.74 886.40 525.14 7.51 690.40 35.30 40 9.97 56.48 123.25 765.15 429.61 10.01 665.15 19.97 50 5.04 17.26 34.65 502.71 349.08 12.51 500.71 15.04 60 3.03 6.77 11.00 220.19 187.25 15.01 245.19 13.03 70 2.02 3.25 4.49 80.51 42.25 17.51 90.83 12.02 80 1.49 1.91 2.27 29.83 9.51 20 19.58 9.49 90 1.18 1.30 1.40 11.91 2.83 22.49 10.11 1.18 ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 166

However, there is a large variation in the implementation of median filters. As a result several modification of median filtering has been performed by various researchers throughout the world. Hybrid median filter, weighted median filter, trimmed median filter, adaptive median filter, un-symmetric median filter, cascaded filter, fuzzy based median filter etc. are some of them. In this work the variations of median filter is discussed and the results of various median filtered images are shown to demonstrate the differences obtained in each filtering Fig16. PSNR versus Noise Density methods. The experiment is conducted on sickle cell affected blood smear images obtained from microscopic view. The performance evaluation of Gaussian and impulse noise removal using different filters is performed using RMSE, PSNR and IEF measures. Fig17. IEF versus Noise Density The fig 16 and 17 shown the plots of PSNR and IEF with Noise density (ND) reveals the performance comparison of the filters mentioned in this work. These filters show a consistent and stable performance while varying the noise densities from 10 to 90%. The filters with high PSNR value have good quality of compressed or reconstructed image. During the noise removal process these filters preserves the edges of the object in images. IV. Conclusions Standard median filter is used in image processing for the preprocessing of input images. V. References 1. R. H. Chan, C. W. Ho, M. Nikolova, Salt-and-pepper noise removal by median-type noise detectors and detailpreserving regularization, IEEE Transaction on Image Processing, vol. 14, pp. 1479-1485, October 2005. 2. Mu-Hsien Hsieh, Fan-Chieh Cheng, Mon-Chau Shie, Shanq-JangRuan, Fast and efficient median filter for removing 1 99% levels of salt-and-pepper noise in images, Engineering Applications of Artificial Intelligence 26 pp 1333 1338, 2013. 3. Prateek Kumar Garg, PushpneelVerma, AnkurBhardwaz, A Survey Paper on Various Median Filtering Techniques for Noise Removal from Digital Images, American International Journal of Research in Formal, Applied & Natural Sciences, pp 43-47, 2014. 4. K. S. Srinivasan and D. Ebenezer, A new fast and efficient decision based algorithm for removal of high density impulse noise, IEEE Signal Processing Letters, Vol. 14, NO. 3, pp 189-162, March 2007. ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 167

5. K. Aiswarya, V. Jayaraj, and D. Ebenezer, A new and efficient algorithm for the removal of high density salt and pepper noise in images and videos, in Second Int. Conf. Computer Modeling and Simulation, pp. 409 413, 2010. 6. GnanambalIlango and R. Marudhachalam, New Hybrid Filtering Techniques for Removal of Gaussian Noise from Medical Images, ARPN Journal of Engineering and Applied Sciences, vol. 6, no. 2, February 2011. 7. S. Peng and L. Lucke, A hybrid filter for image enhancement, Proc. of International Conference on Image Processing (ICIP). 1: 163-166, 1995. 8. Ko S.J. and Lee Y.H., "Center weighted median filters and their applications to image enhancement," IEEE Trans. Circuits Systems, Vol. 38, No. 9, pp. 984 993, 1991. 9. Brownrigg D.R.K., 'The weighted median filter,' Communication, ACM, Vol. 27, No.8, pp. 807-818, 1984. 10. Ching-Ta Lu, Tzu-Chun Chou, Denoising of salt-and-pepper noise corrupted image using modified directional-weighted-median filter, Springer Pattern Recognition Letters, (33), pp 1287 1295, 2012. 11. Wang Chang-you, LI Lin-lin, YANG Fu-ping, GONG Hui, A new kind of adaptive weighted median filter algorithm, IEEE 2010 International Conference on Computer Application and System Modeling, pp 667-671, 2010. 12. N.Krishna Chaitanya, P. Sreenivasulu, Removal of Salt and Pepper Noise Using Advanced Modified Decision Based Unsymmetric Trimmed Median Filter, IEEE 2014 International Conference on Electronics and Communication System, 2014. 13. Athi Narayanan, G. Arumugam, Prof. Kamal Bijlani, Trimmed Median Filters for Salt and Pepper Noise Removal, International Journal of Emerging Trends & Technology in Computer Science, Volume 2, Issue 1,pp 35-40, January February 2013. 14. S. Esakkirajan, T. Veerakumar, Adabala N. Subramanyam, and C. H. Prem Chand, Removal of High Density Salt and Pepper Noise Through Modified Decision Based Unsymmetric Trimmed Median Filter, IEEE Signal Processing Letters, vol. 18, no. 5, pp 287-290, May 2011. 15. Shruti Singh, Anil Kumar Tiwari, An Enhanced Decision Based Unsymmetric Trimmed Median Filter for Removal of High Density Salt and Pepper Noise, IEEE 2015 Fifth National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), 2016. 16. KesariVermaa, Bikesh Kumar Singhb, A.S. Thokec, An Enhancement in Adaptive Median filter for Edge Preservation, International Conference on Intelligent Computing, Communication & Convergence, Procedia Computer Science 48, pp 29 36, 2015. 17. Jiafu Jiang, Jing Shen, An Effective Adaptive Median Filter Algorithm for Removing Salt & Pepper Noise in Images IEEE Symposium on Photonics and Opto-electronic, 2010. 18. Xin Wang, Generalized Multistage Median Filters, IEEE Transactions on Image Processing, Vol. I. No. 4, Pp 543-545, October 1992. 19. Hakan Güray Senel, Richard Alan Peters and Benoit Dawant. Topological Median Filter, IEEE Trans on Image Processing. 11(2): pp 89-104, 2002. 20. Yan Zhou and Quan-hua Tang, Wei-dong Jin, Adaptive Fuzzy Median Filter for Images Corrupted by Impulse Noise, IEEE Congress on Image and Signal Processing, pp 265-269, 2008. 21. K. K. V. Toh and N. A. M. Isa, Noise adaptive fuzzy switching median filter for ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 168

salt-and-pepper noise reduction, IEEE Signal Processing Letters, vol. 17, pp. 281-284, March 2010. 22. Joan-Gerard Camarena, Valent ın Gregori, Samuel Morillas, and Almanzor Sapena, A Simple Fuzzy Method to Remove Mixed Gaussian-Impulsive Noise from Color Images, IEEE Transactions On Fuzzy Systems, vol. 21, no. 5, pp 971-978,, October 2013. 23. A. Asano, K. Itoh, and Y. Ichioka, Optimization of the weighted median filter by learning, Optics Express, vol. 16, no. 3, pp. 168 170, 1991. 24. C. Spence and C. Fancourt, An iterative method for vector median filtering, In IEEE International Conference on Image Processing (ICIP 2007), vol. V, pp. 265 268, 2007. 25. A. Toprak and I. Guler, Suppression of impulse noise in medical images with the use of fuzzy adaptive median filter, Journal of Medical Systems, vol. 30, no. 6, pp. 465 471, 2006. 26. A. Toprak and I. Guler, Impulse noise reduction in medical images with the use of switch mode fuzzy adaptive median filter, Digital Signal Processing, vol. 17, no. 4, pp. 711 723, 2007. 27. A. Toprak, M. S. Ozerdem, and I. Guler, Suppression of impulse noise in MR images using artificial intelligent based neuro-fuzzy adaptive median filter, Digital Signal Processing, vol.18, no. 3, pp. 391 405, 2008. 28. W. Luo, Efficient removal of impulse noise from digital images, IEEE Transactions on Consumer Electronics, vol. 52, no. 2, pp.523-527, 2006. 29. Siddhartha A., R. A. Patil, Denoising Techniques for biomedical images, International Journal of Engineering Research and General Science, vol 4, issue 2, pp 706-710, 2016. 30. R. Gonzalez and R. Woods. 1992. Digital Image Processing. Adison -Wesley, New York. Aruna N.S. received M.Tech and B.Tech Degrees in Electrical and Electronics Engineering from University of Kerala during the year 2009 and 2006 respectively. Her area of interest is in Biomedical image processing. Currently doing research work in image processing at Department of Electrical Engineering, College of Engineering, Trivandrum, University of Kerala, India. Dr. Hariharan S. received Ph.D from Indian Institute of Technology, Kharagpur, India during 2002 and M.Tech degree from Indian Institute of Technology, Bombay, India during 1992. His area of interest is in Biomedical Image Processing. Currently he is working as Professor and UG Co-ordiantor at Department of Electrical Engineering, College of Engineering, Trivandrum, Kerala, India. ISSN 2394-0573 All Rights Reserved 2017 IJEETE Page 169