- Email: [email protected]
Positive–negative impulse noise removal based on row and column filtering
Accepted Manuscript Title: Positive-negative Impulse Noise Removal Based on Row and Column Filtering Author: Shuyue Chen Yang Xu Jianwu Wan PII: DOI: Reference:
S0030-4026(15)00455-6 http://dx.doi.org/doi:10.1016/j.ijleo.2015.05.136 IJLEO 55626
To appear in: Received date: Accepted date:
22-4-2014 30-5-2015
Please cite this article as:http://dx.doi.org/10.1016/j.ijleo.2015.05.136 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
*Manuscript
Positive-negative Impulse Noise Removal Based on Row and Column Filtering
Shuyue Chen,Yang Xu, Jianwu Wan School of Information Science and Engineering Changzhou University
ip t
Changzhou, China e-mail: [email protected]
cr
Abstract- In the field of photo-electronic imaging, impulse noise seriously affects the image quality. In order to
remove it, a new positive-negative impulse denoise filter based on the row and column filtering(PN-RCF) is
us
proposed and its performance of noise removal and detail protection is analyzed. Compared with other nonlinear filters through the simulations, the given filter not only can suppress the positive-negative impulse noise, but also can protect the edges and detail well with simple row and column processing.
an
Keywords- nonlinear filter, positive-negative impulse noise, protection of edge detail, image enhancement
INTRODUCTION
Image processing, such as edge detection, image segmentation and so on[1], is to make the images
M
damaged by external causes restored more better[2]. However, the impulse noise may influence the effect of image processing. During the image acquisition and transmission, the impulse noise is caused by various factors, such as the imaging conditions, the quality of their components.
ed
In order to deal with the impulse noise, many filter based approaches have been proposed, and the key problem to the filter’s based approaches is to improve the filter ability. In general, the filter ability is inherently decided by: (1) the ability of removing the noise; (2) the filter’s edge and detail protection. [3]
, such as median filter, recursive median filter, adaptive filters, robust
ce pt
The conventional linear filters
statistical adaptive filter and so on, have a better ability in removing noise. However, they all have some drawbacks. For example, the median filter and its transform methods[4-8], may cause the edge of image evacuation and smears, which make the image distortion [9]. In addition to the traditional filters, there are many new filters appeared. For example, a two-output nonlinear filter[10] can remove impulse noise and has a good deal with the image detail, but it only deals with black and white images. Quaternion switching filter for impulse noise reduction for color image[11] is another filter that has a better
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
performance of noise suppression and detail preservation, but the method is more complicated. An effective 2-stage techniques for removing impulse noise is presented[12], it is of great noise removal effect, but the edge and detail protection is limited. To deal with the positive and negative impulse noise, we carried out a nonlinear filter (PN-RCF) based our previous work[13]. Compared with several kinds of nonlinear filters, the experimental results demonstrated that PN-RCF can remove positive-negative impulse noise, and it has a better ability to protect the edge and detail than others, which is very important in some cases.
FILTER BASED ON ROW AND COLUMN PROCESSING AND ITS PROPERTIES A.Definition of PN-RCF
Page 1 of 6
PN-RCF based on RCF distinguishes the impulse noise by following two steps: (1)comparing the difference of adjacent data with thresholds, (2)replace impulse noise with the following operation on adjacent data. Suppose X={x(i,j)∈ Z;i,j∈ Z} be the original image that contains positive-negative impulse, x(i,j) be the gray value of pixel (i,j), Z be the set of integers. An image is firstly filtered by line-by-line, and then by column by column. The definition of the filter is defined as follows:
ip t
x(i, j 1),|x(i,j)-x(i,j-1)| k x(i, ) yr (i, j ) (1) otherwise x(i, j ), y (i 1, j ),| yR (i, j ) yR (i 1, j ) | k [yR (, j )] (2) yRC (i, j ) R otherwise yR (i, j ), x (i , ) and y (, j ) R
stand for the first order difference of the i-th row of image X and j-th
cr
where
us
column of image yR ,respectively. σ is the sign of standard deviation. k is a tuning coefficient, and is generally seted as k=1.0. If k is larger (than one), only higher gray-level impulses are removed and vice versa. B.Filtering Properties of PN-RCF Impulse noise in image can be divided into positive and negative impulse noise. Gray-levels of the
an
positive impulse noise are larger than their neighborhood and negative impulse noises’ gray-levels are less than their neighborhood. According to the limitation, the PN-RCF can not remove more than 2×2
M
noise blocks. The possible situations both in one-dimension and two-dimension are listed below: Fig.1 shows the possible one-dimensional situations: single, interval and adjacent impulse (each case includes two situations: the positive and the negative). According to (1), the value of single impulse at
ed
j-th position in Fig.1(a) may be replaced by its neighborhood j+1th shown in Fig1(b). Similarly, the two impulses in Fig.1(c) may be removed as shown in Fig.1(d). Under the situation of adjacent impulse in Fig.1(e), the value of unit j-th may be substituted by the unit j+1th, and unit j+1th remains the same. (
stands for the positive and negative impulse noise respectively in Fig.1. the positive impulse noise,
the negative impulse noise)
ce pt
and
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Figure 1 row or column filtering performance (a), (c) and (e) are three typical distribution of impulses, (b), (d) and (f) are filtering results respectively by
PN-RCF. Fig.2 is a similar discussion which is opposite to the situation mentioned above.
Figure 2 situations opposite to Figure.1 (a), (c) and (e) are three typical distribution of impulses, (b), (d) and (f) are filtering results respectively by PN-RCF.
Page 2 of 6
Fig.3 shows the possible two-dimensional situations: single consecutive impulses shown in Fig.3(a), double consecutive impulses shown in Fig.3(d). According to (1) and (2), Fig.3(a) turns into Fig.3(b) through the row filtering line-by-line, then all of the noise are removed through the filter column-by-column, which is shown in Fig.3(c). As the double consecutive impulses，Fig.3(d) is changed into Fig.3(e) by (1), and the noise is not removed completely after column-by-column, which is shown in Fig.3(f). Therefore, we can remove all of single consecutive impulses except for double
Figure 3. two-dimensional possible situations
M
an
us
cr
ip t
consecutive impulse noise totally with PN-RCF.
2-D filtering properties, (a) and (d) are two typical distribution of impulses, (b) and (e) are filtering results of rows,
ed
(c) and (f) are final results responding to (a) and (d).
C.Detail Preservation Properties of PN-RCF
ce pt
For images, the valuable edge and detail information hides within one or several pixels. So, the detail and edge protection is very important during the filtering. As the PN-RCF filter that removes the noise is slightly worse than the 3*3 median filter, PN-RCF compares the absolute value of differences of adjacent pixels with the certain thresholds for distinguishing impulse noise: if it is the impulse noise, replace it by the pixel after it. This strategy can greatly ensure the integrity of the original image edges and details, which is superior to the other filters.
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
SIMULATIONS AND DISCUSSION For the strip-like images, the PN-RCF can’t remove the noise very well, but it can protect the image edges and details. Here we add the salt and pepper noise to the image, and make a comparison with the 3 * 3 median filter, the median filter 5 * 5, and the recursive median filter. Fig.4(a) is the original image.Fig.4(b)is the image which contains the salt and pepper noise. From Fig.4, we can observe that: (1)3*3median filter removes the most noise but can’t protect the image edge and details; (2) 5*5median filter causes the part of the image damaged; (3)the recursive median filter can remove the noise very well and deal with the edge and details better than 5*5median filter. (4) PN-RCF remove the noise worse than others, but it protects the edge and details better than others. From the Tab.1,we can find the MAE(mean absolute error) of PN-RCF is the smallest.
Page 3 of 6
ip t
(b)
(c)
(e)
(f)
an
(d)
us
cr
(a)
Fig.4 Ribbon image processing analysis
(a)original image, (b)polluted noise, (c)3*3median filter, (d)5*5midian filter, (e)Recursive median filter, (f)PN-RCF
Noise Intensity
Image
3*3
polluted by noise
0.05
0.01034
M
MAE of Several Filters
5*5
Recursive median
Median
Median
filter
filter
filter
ed
TABLE 1
0.00384
0.03279
0.00721
PN-RCF
0.00376
For the scene images, the PN-RCF can remove the noise and deal with the image edge and details
ce pt
very well. Here we also make a comparison with the 3 * 3 median filter, the median filter 5 * 5, and the recursive median filter to test PN-RCF filtering and the protection ability for the edge or detail. Fig.5(a) is an ideal image. Fig.5(b) is the image which is polluted by the salt and pepper noise.Fig5(c) and Fig.5 (d) are the images of median filtering with 3×3 and 5×5 window respectively. Fig.5 (e) is the result of recursive median filtering with 3×3 window, Fig.5 (f) is the image after PN-RCF. By the comparison, four filters are able to filter out the vast majority of impulse noise, but
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
PN-RCF have a better deal with the image’s edge and detail than the other three filters. Tab.2 shows MAE of several filters under different
noise ratio.Fig.5,6 is the second case. From the Tab.2, we can
find the MAE of PN-RCF is the smallest.
(a)
(b)
(c)
Page 4 of 6
ip t
(d)
(e)
(f)
cr
Fig.5 Scene image processing analysis
(a)original image, (b)polluted noise,(c)3*3median filter , (d)5*5midian filter, (e)Recursive median filter, (f)PN-RCF
Noise Intensity
Image
3*3
5*5
polluted by noise
Median
Median
filter
filter
0.0250
0.0273
0.10
0.0503
0.0287
0.15
0.0747
0.0302
CONCLUSION
Recursive median
PN-RCF
filter
0.0401
0.0509
0.0180
0.0407
0.0518
0.0185
0.0414
0.0531
0.0234
M
0.05
us
MAE of Several Filters
an
TABLE 2
ed
In order to remove the positive-negative impulse noise, a nonlinear filter which is based on RCF to remove the positive-negative impulse noise is presented and its performance for impulse noise removal and edge detail protection has been described. The simulation experiments by comparing with properties.
ce pt
several typical filters show that PN-RCF is of better noise reduction and edge and detail protection
ACKNOWLEDGEMENTS
This study was supported by the National Natural Science Foundation of China under Grant 51176016.
REFERENCES
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
[1] M. Emin Yuksel ,A simple neuro-fuzzy method for improving the performances of impulse noise filters for digital images,AEU - International Journal of Electronics and Communications, Volume 59, Issue 8, 1 December 2005, Pages 463-472 [2] Peng Liu, Kie B. Eom,Restoration of multispectral images by total variation with auxiliary image Optics and Lasers in Engineering, Volume 51, Issue 7, July 2013, Pages 873-882 [3] Alberto Carini, Giovanni L. Sicuranza,Fourier nonlinear filters,Signal Processing, Volume 94, January 2014, Pages 183-194 [4] Tzu-Chao Lin, Chao-Ming Lin, Mu-Kun Liu, Chien-Ting Yeh,Partition-based fuzzy median filter based on adaptive resonance theory,Computer Standards & Interfaces, Volume 36, Issue 3, March 2014, Pages 631-640 [5] Ching-Ta Lu,Noise reduction using three-step gain factor and iterative-directional-median filter,Applied Acoustics, Volume 76, February 2014, Pages 249-261 [6] Mu-Hsien Hsieh, Fan-Chieh Cheng, Mon-Chau Shie, Shanq-Jang Ruan,Fast and efficient median filter for removing 1–99% levels of salt-and-pepper noise in images,Engineering Applications of Artificial Intelligence, Volume 26, Issue 4, April 2013, Pages 1333-1338 [7] S. Anitha, V. Radha,Enhanced Switching Median Filter for Denoising in 2D Patterned Textile Images,Procedia Engineering, Volume 38, 2012, Pages 3362-3372 [8] Ayyüce M. Kızrak, Figen Öze, A new median filter based finger print recognition algorithm,Procedia Computer Science, Volume 3, 2011, Pages 859-865 [9] Haewon Nam, Hae-Jeong Park,Distortion correction of high b-valued and high angular resolution diffusion images using iterative simulated images,NeuroImage, Volume 57, Issue 3, 1 August 2011, Pages 968-978
Page 5 of 6
[13] Chen, Shuyue, Qiu, Yun; Feng, Jun ,A Nonlinear Filter Based on Row and Column Processing for Positive Impulse
ce pt
ed
M
an
us
cr
ip t
Noise Reduction, Proceedings - 2009 Asia-Pacific Conference on Information Processing, APCIP 2009, v 2, p 44-46, 2009
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
[10] Fabrizio RussoImpulse noise cancellation in image data using a two-output nonlinear filter Measurement, Volume 36, Issues 3–4, October–December 2004, Pages 205-213 [11] Xin Geng, Xiaoguang Hu, Jin Xiao.Quaternion switching filter for impulse noise reduction in color image.Signal Processing, Volume 92, Issue 1, January 2012, Pages 150-162 [12] Sema Koç Kayhan,An effective 2-stage method for removing impulse noise in images Journal of Visual Communication and Image Representation, Volume 25, Issue 2, February 2014, Pages 478-486
Page 6 of 6
S0030-4026(15)00455-6 http://dx.doi.org/doi:10.1016/j.ijleo.2015.05.136 IJLEO 55626
To appear in: Received date: Accepted date:
22-4-2014 30-5-2015
Please cite this article as:
*Manuscript
Positive-negative Impulse Noise Removal Based on Row and Column Filtering
Shuyue Chen,Yang Xu, Jianwu Wan School of Information Science and Engineering Changzhou University
ip t
Changzhou, China e-mail: [email protected]
cr
Abstract- In the field of photo-electronic imaging, impulse noise seriously affects the image quality. In order to
remove it, a new positive-negative impulse denoise filter based on the row and column filtering(PN-RCF) is
us
proposed and its performance of noise removal and detail protection is analyzed. Compared with other nonlinear filters through the simulations, the given filter not only can suppress the positive-negative impulse noise, but also can protect the edges and detail well with simple row and column processing.
an
Keywords- nonlinear filter, positive-negative impulse noise, protection of edge detail, image enhancement
INTRODUCTION
Image processing, such as edge detection, image segmentation and so on[1], is to make the images
M
damaged by external causes restored more better[2]. However, the impulse noise may influence the effect of image processing. During the image acquisition and transmission, the impulse noise is caused by various factors, such as the imaging conditions, the quality of their components.
ed
In order to deal with the impulse noise, many filter based approaches have been proposed, and the key problem to the filter’s based approaches is to improve the filter ability. In general, the filter ability is inherently decided by: (1) the ability of removing the noise; (2) the filter’s edge and detail protection. [3]
, such as median filter, recursive median filter, adaptive filters, robust
ce pt
The conventional linear filters
statistical adaptive filter and so on, have a better ability in removing noise. However, they all have some drawbacks. For example, the median filter and its transform methods[4-8], may cause the edge of image evacuation and smears, which make the image distortion [9]. In addition to the traditional filters, there are many new filters appeared. For example, a two-output nonlinear filter[10] can remove impulse noise and has a good deal with the image detail, but it only deals with black and white images. Quaternion switching filter for impulse noise reduction for color image[11] is another filter that has a better
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
performance of noise suppression and detail preservation, but the method is more complicated. An effective 2-stage techniques for removing impulse noise is presented[12], it is of great noise removal effect, but the edge and detail protection is limited. To deal with the positive and negative impulse noise, we carried out a nonlinear filter (PN-RCF) based our previous work[13]. Compared with several kinds of nonlinear filters, the experimental results demonstrated that PN-RCF can remove positive-negative impulse noise, and it has a better ability to protect the edge and detail than others, which is very important in some cases.
FILTER BASED ON ROW AND COLUMN PROCESSING AND ITS PROPERTIES A.Definition of PN-RCF
Page 1 of 6
PN-RCF based on RCF distinguishes the impulse noise by following two steps: (1)comparing the difference of adjacent data with thresholds, (2)replace impulse noise with the following operation on adjacent data. Suppose X={x(i,j)∈ Z;i,j∈ Z} be the original image that contains positive-negative impulse, x(i,j) be the gray value of pixel (i,j), Z be the set of integers. An image is firstly filtered by line-by-line, and then by column by column. The definition of the filter is defined as follows:
ip t
x(i, j 1),|x(i,j)-x(i,j-1)| k x(i, ) yr (i, j ) (1) otherwise x(i, j ), y (i 1, j ),| yR (i, j ) yR (i 1, j ) | k [yR (, j )] (2) yRC (i, j ) R otherwise yR (i, j ), x (i , ) and y (, j ) R
stand for the first order difference of the i-th row of image X and j-th
cr
where
us
column of image yR ,respectively. σ is the sign of standard deviation. k is a tuning coefficient, and is generally seted as k=1.0. If k is larger (than one), only higher gray-level impulses are removed and vice versa. B.Filtering Properties of PN-RCF Impulse noise in image can be divided into positive and negative impulse noise. Gray-levels of the
an
positive impulse noise are larger than their neighborhood and negative impulse noises’ gray-levels are less than their neighborhood. According to the limitation, the PN-RCF can not remove more than 2×2
M
noise blocks. The possible situations both in one-dimension and two-dimension are listed below: Fig.1 shows the possible one-dimensional situations: single, interval and adjacent impulse (each case includes two situations: the positive and the negative). According to (1), the value of single impulse at
ed
j-th position in Fig.1(a) may be replaced by its neighborhood j+1th shown in Fig1(b). Similarly, the two impulses in Fig.1(c) may be removed as shown in Fig.1(d). Under the situation of adjacent impulse in Fig.1(e), the value of unit j-th may be substituted by the unit j+1th, and unit j+1th remains the same. (
stands for the positive and negative impulse noise respectively in Fig.1. the positive impulse noise,
the negative impulse noise)
ce pt
and
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Figure 1 row or column filtering performance (a), (c) and (e) are three typical distribution of impulses, (b), (d) and (f) are filtering results respectively by
PN-RCF. Fig.2 is a similar discussion which is opposite to the situation mentioned above.
Figure 2 situations opposite to Figure.1 (a), (c) and (e) are three typical distribution of impulses, (b), (d) and (f) are filtering results respectively by PN-RCF.
Page 2 of 6
Fig.3 shows the possible two-dimensional situations: single consecutive impulses shown in Fig.3(a), double consecutive impulses shown in Fig.3(d). According to (1) and (2), Fig.3(a) turns into Fig.3(b) through the row filtering line-by-line, then all of the noise are removed through the filter column-by-column, which is shown in Fig.3(c). As the double consecutive impulses，Fig.3(d) is changed into Fig.3(e) by (1), and the noise is not removed completely after column-by-column, which is shown in Fig.3(f). Therefore, we can remove all of single consecutive impulses except for double
Figure 3. two-dimensional possible situations
M
an
us
cr
ip t
consecutive impulse noise totally with PN-RCF.
2-D filtering properties, (a) and (d) are two typical distribution of impulses, (b) and (e) are filtering results of rows,
ed
(c) and (f) are final results responding to (a) and (d).
C.Detail Preservation Properties of PN-RCF
ce pt
For images, the valuable edge and detail information hides within one or several pixels. So, the detail and edge protection is very important during the filtering. As the PN-RCF filter that removes the noise is slightly worse than the 3*3 median filter, PN-RCF compares the absolute value of differences of adjacent pixels with the certain thresholds for distinguishing impulse noise: if it is the impulse noise, replace it by the pixel after it. This strategy can greatly ensure the integrity of the original image edges and details, which is superior to the other filters.
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
SIMULATIONS AND DISCUSSION For the strip-like images, the PN-RCF can’t remove the noise very well, but it can protect the image edges and details. Here we add the salt and pepper noise to the image, and make a comparison with the 3 * 3 median filter, the median filter 5 * 5, and the recursive median filter. Fig.4(a) is the original image.Fig.4(b)is the image which contains the salt and pepper noise. From Fig.4, we can observe that: (1)3*3median filter removes the most noise but can’t protect the image edge and details; (2) 5*5median filter causes the part of the image damaged; (3)the recursive median filter can remove the noise very well and deal with the edge and details better than 5*5median filter. (4) PN-RCF remove the noise worse than others, but it protects the edge and details better than others. From the Tab.1,we can find the MAE(mean absolute error) of PN-RCF is the smallest.
Page 3 of 6
ip t
(b)
(c)
(e)
(f)
an
(d)
us
cr
(a)
Fig.4 Ribbon image processing analysis
(a)original image, (b)polluted noise, (c)3*3median filter, (d)5*5midian filter, (e)Recursive median filter, (f)PN-RCF
Noise Intensity
Image
3*3
polluted by noise
0.05
0.01034
M
MAE of Several Filters
5*5
Recursive median
Median
Median
filter
filter
filter
ed
TABLE 1
0.00384
0.03279
0.00721
PN-RCF
0.00376
For the scene images, the PN-RCF can remove the noise and deal with the image edge and details
ce pt
very well. Here we also make a comparison with the 3 * 3 median filter, the median filter 5 * 5, and the recursive median filter to test PN-RCF filtering and the protection ability for the edge or detail. Fig.5(a) is an ideal image. Fig.5(b) is the image which is polluted by the salt and pepper noise.Fig5(c) and Fig.5 (d) are the images of median filtering with 3×3 and 5×5 window respectively. Fig.5 (e) is the result of recursive median filtering with 3×3 window, Fig.5 (f) is the image after PN-RCF. By the comparison, four filters are able to filter out the vast majority of impulse noise, but
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
PN-RCF have a better deal with the image’s edge and detail than the other three filters. Tab.2 shows MAE of several filters under different
noise ratio.Fig.5,6 is the second case. From the Tab.2, we can
find the MAE of PN-RCF is the smallest.
(a)
(b)
(c)
Page 4 of 6
ip t
(d)
(e)
(f)
cr
Fig.5 Scene image processing analysis
(a)original image, (b)polluted noise,(c)3*3median filter , (d)5*5midian filter, (e)Recursive median filter, (f)PN-RCF
Noise Intensity
Image
3*3
5*5
polluted by noise
Median
Median
filter
filter
0.0250
0.0273
0.10
0.0503
0.0287
0.15
0.0747
0.0302
CONCLUSION
Recursive median
PN-RCF
filter
0.0401
0.0509
0.0180
0.0407
0.0518
0.0185
0.0414
0.0531
0.0234
M
0.05
us
MAE of Several Filters
an
TABLE 2
ed
In order to remove the positive-negative impulse noise, a nonlinear filter which is based on RCF to remove the positive-negative impulse noise is presented and its performance for impulse noise removal and edge detail protection has been described. The simulation experiments by comparing with properties.
ce pt
several typical filters show that PN-RCF is of better noise reduction and edge and detail protection
ACKNOWLEDGEMENTS
This study was supported by the National Natural Science Foundation of China under Grant 51176016.
REFERENCES
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
[1] M. Emin Yuksel ,A simple neuro-fuzzy method for improving the performances of impulse noise filters for digital images,AEU - International Journal of Electronics and Communications, Volume 59, Issue 8, 1 December 2005, Pages 463-472 [2] Peng Liu, Kie B. Eom,Restoration of multispectral images by total variation with auxiliary image Optics and Lasers in Engineering, Volume 51, Issue 7, July 2013, Pages 873-882 [3] Alberto Carini, Giovanni L. Sicuranza,Fourier nonlinear filters,Signal Processing, Volume 94, January 2014, Pages 183-194 [4] Tzu-Chao Lin, Chao-Ming Lin, Mu-Kun Liu, Chien-Ting Yeh,Partition-based fuzzy median filter based on adaptive resonance theory,Computer Standards & Interfaces, Volume 36, Issue 3, March 2014, Pages 631-640 [5] Ching-Ta Lu,Noise reduction using three-step gain factor and iterative-directional-median filter,Applied Acoustics, Volume 76, February 2014, Pages 249-261 [6] Mu-Hsien Hsieh, Fan-Chieh Cheng, Mon-Chau Shie, Shanq-Jang Ruan,Fast and efficient median filter for removing 1–99% levels of salt-and-pepper noise in images,Engineering Applications of Artificial Intelligence, Volume 26, Issue 4, April 2013, Pages 1333-1338 [7] S. Anitha, V. Radha,Enhanced Switching Median Filter for Denoising in 2D Patterned Textile Images,Procedia Engineering, Volume 38, 2012, Pages 3362-3372 [8] Ayyüce M. Kızrak, Figen Öze, A new median filter based finger print recognition algorithm,Procedia Computer Science, Volume 3, 2011, Pages 859-865 [9] Haewon Nam, Hae-Jeong Park,Distortion correction of high b-valued and high angular resolution diffusion images using iterative simulated images,NeuroImage, Volume 57, Issue 3, 1 August 2011, Pages 968-978
Page 5 of 6
[13] Chen, Shuyue, Qiu, Yun; Feng, Jun ,A Nonlinear Filter Based on Row and Column Processing for Positive Impulse
ce pt
ed
M
an
us
cr
ip t
Noise Reduction, Proceedings - 2009 Asia-Pacific Conference on Information Processing, APCIP 2009, v 2, p 44-46, 2009
Ac
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
[10] Fabrizio RussoImpulse noise cancellation in image data using a two-output nonlinear filter Measurement, Volume 36, Issues 3–4, October–December 2004, Pages 205-213 [11] Xin Geng, Xiaoguang Hu, Jin Xiao.Quaternion switching filter for impulse noise reduction in color image.Signal Processing, Volume 92, Issue 1, January 2012, Pages 150-162 [12] Sema Koç Kayhan,An effective 2-stage method for removing impulse noise in images Journal of Visual Communication and Image Representation, Volume 25, Issue 2, February 2014, Pages 478-486
Page 6 of 6