An Efficient Component Based Filter for Random Valued Impulse Noise Removal Manohar Koli Research Scholar, Department of Computer Science, Tumkur University, Tumkur, Karnataka, India. S. Balaji Centre for Emerging Technologies, Jain Global Campus, Jain University, Jakkasandra Post, Kanakapura Taluk, Ramanagara Dist., Karnataka, India. Abstract In this digital world, due to faulty sensors, storage, and channels images videos are often corrupted by impulse noise, which is a frequently occurring noise type in an image. Impulse noise are classified into fixed valued (salt and pepper noise) and random valued impulse noise. This paper proposes an effective noise reduction method for images corrupted by the random valued impulse noise, handling of which is more difficult than the salt and pepper impulse noise. Our method is based on the concept that the impulse noise produces small components (patches) on an image less than 10x10 size. We convert a gray scale image into a binary image and analyze the components less than 10x10 size on various parameters and remove them using neighboring pixel connectivity. Secondly, removed pixels are filled by the estimated value of median filter calculated with the help of neighboring pixels in that area. Comparison of the proposed algorithm with other existing algorithms shows that the proposed component based filter performs better than all other existing algorithms. The visual and quantitative results show that the performance of the algorithm is very good and it handles more than 70% noise. Keywords: Filtering, Component Based Filter, Random Valued Impulse Noise, Image Restoration. Introduction The impulse noise is one which may corrupt the images during their acquisition, transmission or storage. Several algorithms have been proposed to remove the impulse noise in the images. Random Valued Impulse Noise (RVIN) assumes a noise value between the minimum value 0 and the maximum value 255 of the noise, as shown in equation (1) [1-2]. The median and the mean value based filters are the most popular non-linear filters. When an image contains a small amount of noise, they are efficient but they do not handle the large percentage of noise. Hence, in this paper, a non-linear component based filter is proposed. In literature, it is observed that only few algorithms are proposed to handle RVIN. Our (1) main aim is to provide a better solution to RVIN than the available algorithms in the literature. The proposed Component Based Filter (CBF) is compared with Adaptive Median Filter (AMF) [3], Progressive Switching Median Filter (PSMF) [4], Tri-State Median Filter (TSMF) [5], Adaptive Fuzzy Switching Filter (AFSF) [6], A New Impulse Detector Based on Order Statistics Filter (NIND) [7], An Efficient Algorithm for the Removal of Impulse Noise from Corrupted Images (AEAFRIN) [8], A New Fast and Efficient Decision-Based Algorithm (DBA) [9], An Improved Adaptive Median Filter (IAMF) [10], Robust Statistics Based Algorithm (RSBA) [11], Decision Based Adaptive Median Filter (DBAF) [12], Image Restoration in Non-linear Filtering Domain Using MDB Approach (MDBF) [13], Detail Preserving Adaptive Filter (DPAF) [14] and A Universal Denoising Framework (UDF) [15]. Proposed CBF Algorithm 1. Take input gray scale image (X). 2. Convert gray scale input image (X) to a binary image (Y). 3. Identify the connected components using 8 neighboring pixel connectivity. Remove connected components having less than 10x10 size and having pixels less than 20. In 10x10 (100 pixels) sized components, components can have maximum 100 pixels in it. Since usually noisy components contain sparsely distributed pixels, they can have very less number of pixels in 10x10 area. Hence, If component is made up of less than or equal to 20 pixels then we consider that component is noisy component else we consider component is non-noisy component and we will recover the noisy components. 4. Replace all noisy-component pixels by the median value calculated using the neighboring non-corrupted pixels to the image. 5. Calculate the restored image R (x, y) using 3x3 median filter. Using R and I, compute the difference image D (x, y) and convert D to a binary noise image (Bn). If Bn (x, y) == 1 restore pixels using their neighboring non-corrupted pixels and Set I = R. 6. Using the Centre weighted median filter restore I recursively. 1908
Step by step outputs from the CBI algorithm for IMAGE-1 is shown in Figure 1. Step 1: Input Gray Image IMAGE-1 with 20% RVIN (X) Step 2: Binary Image (Y) Step 3: Noisy connected component having size less than 10X10 and less than 20 connected pixels. Step 4: Image after recovering all noisy components using neighboring non noisy pixels. Step 5: Restored image using 3X3 median filter. Step 6: Restored image using center weighted median filter recursively. Figure 1: Restoration Results of Image-1 with 20% RVIN Performance Measurements To evaluate the performance of the proposed algorithm, four different natural images (IMAGE-2, IMAGE-3, IMAGE-4, and IMAGE-5) are used. The performance is measured using Error Recovery Percentage (ERP) as shown in equations (2) and (3). Figure 2 and Figure 3 show restoration results of our algorithm for the images IMAGE-2 and IMAGE-3 for different amounts of noise. Visibility of output of 70% noisy image clearly shows that the efficiency of our algorithm is very high. Figure 4 and Figure 5 show restoration results of different filters. The visibility of the outputs of our algorithm clearly shows that efficiency of our algorithm is high compared to other algorithms. Calculated ERP for image IMAGE-4 and IMAGE-5 are shown in Table 1 and Table 2. Compared to other popular algorithms ERP value of our algorithm is very high. The results are shown graphically in Figure 6 and Figure 7. (2) 1909
Where X - Original Image R - Restored Image MXN - Size of Image MAE - Mean Absolute Error ERP - Error Recovery Percentage. Table 1: ERP Values of Filters for RVIN IMAGE-4 (300X300) NOISE 10 20 30 40 50 60 70 80 90 FILTERS AMF 57.71 50.55 42.19 35.01 28.18 21.6 17.17 13.68 11.38 PSMF 84.84 86.15 86.21 84.49 78.55 66.45 50.08 35.96 25.44 TSMF 9.61 53.49 66.61 68.97 63.03 52.89 41.04 32.18 25.79 AFSF 60.77 59.11 53.05 45.06 38.56 33.91 28.21 23.99 21.04 NIND 88.09 90.69 91.62 90.88 87.77 79.71 61.87 42.51 22.43 AEAFRIN 77.49 81.69 78.49 70.95 60.49 48.42 36.7 28.38 21.55 DBA 0.68 0.76 0.88 0.78 0.78 0.55 0.64 0.5 0.44 IAMF 85.28 85.66 84.61 82.09 65.36 40.59 20.54 8.08 2.63 RSBA 59.63 51.12 43.34 35.68 28.27 22.22 17.72 13.83 10.79 DBAF 83.8 85.04 81.23 73.13 62.07 50.18 38.78 29.18 21.84 MDBF 72.92 62.18 50.47 40.49 32.12 24.32 18.35 14.06 11.33 DPAF 65.12 56.26 47.96 39.15 29.8 22.55 17.33 13.98 11.11 UDF 73.78 81.48 83.15 82.16 74.32 58.13 41.23 30.03 21.84 PACBF 98.56 97.81 96.74 95.27 92.49 84.74 69.82 53.68 36.55 (3) Table 2: ERP Values of Filters for RVIN IMAGE-5 (300X300) NOISE 10 20 30 40 50 60 70 80 90 FILTERS AMF 79.41 69.25 59.41 50.73 43.01 36.1 29.09 23.24 18.43 PSMF 90.82 90.86 90.51 89.45 86.85 80.55 70.44 55.98 40.7 TSMF 41.42 69.6 78.83 81.71 79.7 73.03 62.14 52.13 41.77 AFSF 94.24 92.58 88.48 82.08 73.57 63.35 54.12 44 34.5 NIND 96.25 96.02 95.7 94.87 93.13 87.56 75.13 55.52 34.13 AEAFRIN 91.55 90.66 88.26 82.26 74.11 64.95 53.85 44.16 34.39 DBA 1.65 1.43 1.29 1.47 1.26 1.23 1.08 1.06 0.99 IAMF 90.26 89.81 88.97 87.81 84.15 69.7 37.7 13.03 4.93 RSBA 79.96 69.25 59.54 50.26 43.06 36.18 29.25 23.64 18.56 DBAF 96.1 94.58 91.34 85.03 76.27 66.24 54.94 44.3 34.76 MDBF 80.14 69.13 59.55 51.01 42.91 35.43 29.19 23.48 18.66 DPAF 80.37 68.96 59.7 50.75 43.08 36.05 28.8 23.15 18.39 UDF 95.63 95.29 94.8 94.03 93.04 90.98 84.37 72.55 60.18 PACBF 99.06 98.31 97.14 96.73 95.14 92.72 87.12 72.93 53.59 ORIGINAL IMAGE-2 (300X300) 05% NOISE 20% NOISE RESTORED IMAGE OF 05% NOISE ERP= 95.25 RESTORED IMAGE OF 20% NOISE ERP= 93.77 1910
35% NOISE 50% NOISE 65% NOISE RESTORED IMAGE OF 35% NOISE ERP= 91.06 RESTORED IMAGE OF 50% NOISE ERP= 88.13 RESTORED IMAGE OF 65% NOISE ERP= 79.06 Figure 2: Restoration Results of Image-2 up to 65% of Noise ORIGINAL IMAGE-3 (300X300) 10% NOISE 25% NOISE RESTORED IMAGE OF10% NOISE ERP= 97.93 RESTORED IMAGE OF 25% NOISE ERP= 96.18 1911
40% NOISE 55% NOISE 70% NOISE RESTORED IMAGE OF 40% NOISE ERP= 93.89 RESTORED IMAGE OF 55% NOISE ERP= 90.63 RESTORED IMAGE OF 70% NOISE ERP= 83.29 Figure 3: Restoration Results of Image-3 up to 70% of Noise ORIGINAL IMAGE-4 (300X300) PACBF AMF PSMF TSMF AFSF 1912
NIND AEAFRIN DBA IAMF RSBA DBAF MDBF DPAF UDF Figure 4: Results of Filters for IMAGE-4 (300X300) with 30% RVIN ORIGINAL IMAGE-5 (300X300) PACBF AMF 1913
PSMF TSMF AFSF NIND AEAFRIN DBA IAMF RSBA DBAF MDBF DPAF UDF Figure 5: Results of Filters for IMAGE-5 (300X300) with 70% RVIN 1914
Figure 6: ERP for the RVIN IMAGE-4 (300x300) Figure 7: ERP for the RVIN IMAGE-5 (300X300). Conclusion In this paper, an efficient component based algorithm to remove the random valued impulse noise from gray scale images is proposed. Experimental results show that the efficiency of the proposed algorithm is very high compared to the other popular algorithms reported in the literature. Further, the proposed algorithm works well in both the low and the high noise conditions up to 70%. This algorithm is a promising solution for the RVIN reduction as it very effectively handles noise and maintains consistency in performance. References [1] Manohar Annappa Koli, Review of Impulse Noise Reduction Techniques, International Journal on Computer Science and Engineering (IJCSE), Vol. 4 No. 02 February 2012, pp 184-196. [2] Sarala singh and Ravimohan, A review on the Median Filter based Impulsive Noise Filtration Techniques for FPGA and CPLD, International Journal of Emerging Technology and Advanced Engineering, Volume 3, Issue 3, March 2013,pp 821-824. [3] H. Hwang and R. A. Haddad Adaptive Median Filters: New Algorithms and Results IEEE Transactions on Image Processing, Vol. 4, No. 4, April 1995, pp 499-502. [4] Zhou Wang and David Zhang Progressive Switching Median Filter for the Removal of Impulse Noise from Highly Corrupted Images IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, Vol. 46, No. 1, January 1999, pp 78-80. [5] Tao Chen, Kai-Kuang Ma, Li-Hui Chen Tristate Median Filter for Image De-noising IEEE Transactions on Image Processing, Vol. 8, No. 12, December 1999, pp 1834-1838. [6] Haixiang Xu, Guangxi Zhu, Haoyu Peng, Desheng Wang Adaptive Fuzzy Switching Filter for Images Corrupted by Impulse noise Pattern Recognition Letters 25 (2004) pp 1657-1663. [7] Wenbin Luo A New Impulse Detector Based on Order Statistics Intl. J. Electronincs Communication (aeu) 60 (2006) pp 462-466. [8] Wenbin Luo An Efficient Algorithm for the Removal of Impulse Noise from Corrupted Images Intl. J. Electron. Commun. (aeü) 61 (2007) pp 551-555. [9] K. S. Srinivasan, D. Ebenezer A New Fast and Efficient Decision-Based Algorithm for Removal of High-Density Impulse Noises IEEE Signal Processing Letters, Vol. 14, No. 3, March 2007, pp 189-192. [10] Mamta Juneja, Rajni Mohana An Improved Adaptive Median Filtering Method for Impulse Noise Detection International Journal of Recent Trends in Engineering, Vol. 1, No. 1, May 2009, pp 274-278. [11] V.R.Vijaykumar, P.T.Vanathi, P.Kanagasabapathy, D.Ebenezer Robust Statistics Based Algorithm to Remove Salt and Pepper Noise in Images International Journal of Information and Communication Engineering 5:3 2009, pp 164-173. [12] V.R.Vijaykumar, Jothibasu Decision Based Adaptive Median Filter to Remove Blotches, Scratches, Streaks, Stripes and Impulse Noise in Image Proceedings of 2010 IEEE 17th International Conference on Image Processing, September 26-29, 2010, Hong Kong, pp 117-120. [13] S. K. Satpathy, S. Panda, K. K. Nagwanshi, C. Ardil Image Restoration in Non-linear Filtering Domain Using MDB Approach International Journal of Information and Communication Engineering 6:1 2010, pp 45-49. [14] Krishna Kant Singh, Akansha Mehrotra, Kirat Pal, M.J.Nigam A N8(p) Detail Preserving Adaptive Filter for Impulse Noise Removal 2011 International Conference on Image Information Processing (ICIIP 2011). [15] Bo Xiong, D. Zhouping Yin A Universal De-noising Framework with a New Impulse Detector and Nonlocal Means IEEE Transactions on Image Processing, Vol. 21, No. 4, April 2012, pp 1663-1675. 1915