Performance analysis of image filtering algorithms for mri images
1 like186 views
This document presents a performance analysis of different image filtering algorithms applied to MRI images to remove various types of noise, such as speckle and Gaussian noise. It evaluates three algorithms—median filter, Wiener filter, and center weighted median filter—comparing their effectiveness based on mean square error (MSE) and peak signal-to-noise ratio (PSNR), with the Wiener filter yielding the best results. The study demonstrates the importance of noise removal in enhancing image quality for medical diagnostics.
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 438
PERFORMANCE ANALYSIS OF IMAGE FILTERING ALGORITHMS
FOR MRI IMAGES
Sivasundari .S1
, R. Siva Kumar2
, M.Karnan3
1
P.G scholar, Department of Computer and Communication, Tamilnadu College of Engineering, Coimbatore,
TamilNadu, India
2
Professor, Department of Information Technology, Tamilnadu College of Engineering, Coimbatore, TamilNadu, India
3
Professor & Head, Department of Computer Science and Engineering, Tamilnadu College of Engineering,
Coimbatore, TamilNadu, India
Abstract
In Image Processing, the digital images are much sensitive to noise which results due to the Image Acquisition errors and
transmission errors.MRI images captured usually are prone to speckle noise, Gaussian noise, salt and pepper noise etc., Image
filtering algorithms are applied over the noisy images to remove the noise and preserve the image details. In this work three
different filtering algorithms such as Median filter (MF), Weiner filter (WF) and Center Weighted Median filter (CWM) are used to
remove the noise present in the MRI images. The performance of these filters are compared using the statistical parameters such as
Mean Square Error(MSE) and Peak Signal to Noise Ratio(PSNR).The study shows that the Weiner filters reconstructs a high
quality image than Median and Center weighted Median filter.
Keywords: Median filter (MF), Weiner filter (WF), Center Weighted Median filter (CWM), Mean Square Error (MSE)
and Peak Signal to Noise Ratio (PSNR).
--------------------------------------------------------------------***-----------------------------------------------------------------
1. INTRODUCTION
In image processing, it is important to obtain quality image
for facilitating the image classification accuracy. As the poor
image quality is an obstacle for efficient image processing.
There is a need for the noise removal in medical images. In
this work MRI images are used for the experiment.
Magnetic Resonance Imaging are now becoming the most
indispensible medical imaging tool used for analysis of
different parts of the human brain.MRI produces the detailed
information of the internal parts of the brain which can be
used for the radiologists to detect the presence of tumors in
the patient’s brain. The scan does not penetrate the adjacent
healthy cells and thus preserves to be safest diagnostic tool.
However one of the primary problems with MRI images are
the presence of added unwanted noises like speckle noise,
Gaussian noise,salt,Rician noise ,salt and pepper noise etc.,
These noise have much influence on the quality of the image
which in turn degrades the performances of Feature
Extraction, Reduction and Classification of the processed
images. The noise have to be removed before these processing
stages. Many image filtering algorithms are available in the
literature.
This paper evaluates the three filtering algorithms for different
noise types. Simulation is done with the MATLAB
R2010a.The organization of this paper is as follows. Section 2
is described with image filtering methods. Section 3 discuss
about the Experiment results and evaluation and finally
Section 4 gives conclusion.
2. FILTERING METHODS
Noise is the unwanted variations or fluctuations of the image
capture. These variations are caused due to errors in the
sensors or data transmission which corrupts the image details
either by causing brightness or frequency changes. Speckle
noise are caused due to transmission errors. Rician noise are
caused due to the Gaussian noise in the original frequency
domain measurements [1]. The amplifier produces a standard
model of noise which is additive, Gaussian. These noise are
independent at each pixel and its intensity. Amplifier noise is
a cause of the "read noise" of an image sensor that is, having
constant noise level in dark areas of the image [2]. A salt-and-
pepper noise will have dark pixels and bright pixels alternate
bright and dark regions [2].Speckle noise is a granular noise
degrades the quality due to transmission errors [3].
Conventionally used simplest methods of linear filtering
algorithms include mean filters. The filtering is performed by
using a mask over every pixel component in the image. Then
by averaging the pixels which falls under the mask results in a
single pixel to be studied. This technique also called as
averaging filter. The main drawback of this filter is that it is
poor in edge preserving.
2.1 Median Filter
The Median filter is advantageous over mean filter and its a
non-linear filtering technique, helps removing noise.
Convolution methods like smoothing filters reduces noise but
blurs the edge features. In order to preserve the edge features
and [4] sharpness of the image we go for Median Filters. It](https://image.slidesharecdn.com/performanceanalysisofimagefilteringalgorithmsformriimages-140819063258-phpapp02/85/Performance-analysis-of-image-filtering-algorithms-for-mri-images-1-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 439
does excellent job in rejecting impulse noise, in which some
individual pixels have extreme values. In Median filtering, for
each pixel, the pixel values in the neighborhood window are
organized according to the intensity and the middle
value(Median) becomes the new value for the central
pixel.i.e., each pixel is set to an ‘average’ of the pixel values
in the neighborhood pixels rather than the mean. This method
can be repeated if necessary since it causes only less blur of
edges. Median filter is less sensitive than the mean filter to
extreme values .The median filter (M) can be implemented by
the formula:
M(P1… P N) =Median (||P 1||2
…….|| P N||2)
- (1)
This is an optimal filter much suitable for de-noising images
of various noise types[5].Causes no changes in the shift
boundaries and no reduction in contrast Causes no changes in
the shift boundaries and no reduction in contrast and proved
to be more robust in the presence of noise.
2.2 Weiner Filter
Weiner filtering carry out an optimal tradeoff between
inverse filtering and noise smoothing. It removes additive
noise and deblurring concurrently [6].This proves to be
optimal in reducing the overall Mean Square Error
(MSE).The operation involves two parts. One is inverse
filtering and the other part is noise smoothing. Wiener filters
belong to a kind of optimum linear filters with the noisy data
as input which involves the calculation of difference between
the desired output sequences from the actual output.
Main purpose of this statistical approach is to diminish the
noise present in a signal by comparing it with an estimation
of the desired noiseless signal. Wiener filters are
characterized by an assumption that signal and (additive)
noise are stationary linear random processes and their
spectral characteristics are calculated. The Performance can
be measured using Minimum Mean-Square Error.
The Wiener filter is implemented as:
𝑊 𝑚, 𝑛 =
𝐻∗ 𝑚,𝑛 𝑃𝑠 (𝑚,𝑛)
𝐻 𝑚,𝑛 2 𝑃𝑠 𝑚,𝑛 +𝑃 𝑛 (𝑚,𝑛)
-(2)
Dividing through by 𝑃𝑠 :
𝑊 𝑚, 𝑛 =
𝐻∗ 𝑚,𝑛
𝐻 𝑚,𝑛 2+
𝑃 𝑛 (𝑚 ,𝑛)
𝑃 𝑠 𝑚 ,𝑛
-(3)
Where
H (m,n) = Degradation function ,
H*
(m,n) = Complex conjugate of degradation function ,
Pn (m,n) = Power Spectral Density of Noise ,
Ps (m,n) = Power Spectral Density of un-degraded image.
The term Pn/Ps can be interpreted as the reciprocal of the
signal-to-noise ratio [5].The statistical parameters Mean
Square Error (MSE) and Peak Signal-to-Noise ratio (PSNR)
are used to evaluate the enhancement performance.
2.3 Center Weighted Median Filter (CWM)
The CWM filter is an extension of Weighed median filter.
Median filters can remove the noise by preserving the edges
and sharpness which is suitable to reduce salt and pepper
noise. In Weighted median filter, the weights of the filter are
assigned based on intensity value of pixels in the MRI image.
The weights used here are 0,0.1,0.2 and 0.3.Initially the
weight of the pixel is set to 0,if the pixel intensity is 0.The
weight assigned is 0.1 for the pixel intensity range between 1-
100 and 0.2 for pixel intensity range 101-200 else 0.3. The
weights set are multiplied with pixel intensity and Median
Filter is applied. This process is called Weighted Median
Filter.
If the center value of each window is set more weight, then its
Center Weighted Median filters (CWM). The CWM filters are
represented by two important parameters such as window size
and center weight which influence the quality of filter. The
center value is assigned large weight i.e. W(0,0)=2N+1,
where N is a non-negative integer, if N>=0 then W(i,j)=1,i
and j not equal to 0.
The performances of CWM are better analyzed by authors
[7,8].The value of weight must chosen carefully as it drives
the performance of the filter. The image characteristics and
type of noise should be considered while selecting the weight.
The larger weight provides better result in preserving image
and edge details but worse in noise removal [9].
3. EXPERIEMENTS AND RESULTS
These filtering algorithms has been tested with various types
of noisy images implemented by using MATLAB 7.1.0.. The
experimental results are tested in Intel Pentium IV 2.4 GHz
processor with 256 MB RAM. Performance of the Median
filter, Center Weighted Median filter, and Weiner filters are
analyzed and evaluated.
Table -1: shows the performance of listed filters
S.no
Filters
MSE PSNR
1 Median
Filter
148949 7.1991
2 CWM
Filter
2.7819e+006 16.312
3 Weiner
Filter
8.3636e+003 17.813](https://image.slidesharecdn.com/performanceanalysisofimagefilteringalgorithmsformriimages-140819063258-phpapp02/85/Performance-analysis-of-image-filtering-algorithms-for-mri-images-2-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 440
Fig -1 Input Image Fig -2 CWM Filtered
Fig -3 Median Filtered Fig -4 Weiner Filtered
4. CONCLUSIONS
In this paper image filtering algorithms are implemented on
MRI images to remove the different types of noise that are
either present in the image during acquisition or injected in to
the image during transmission. MRI images when captured
usually have Gaussian noise and speckle noise. In this work,
three different image filtering algorithms are compared for the
noisy images. The performances of the filters are measured
using the Peak Signal to Noise Ratio (PSNR) and Mean
Square Error (MSE). The Weiner filter gives desirable results
with large PSNR value ensuring high image enhancement.
REFERENCES
[1]. M. N. Nobi and M. A. Yousuf, 2011,” A New Method to
Remove Noise in Magnetic Resonance and Ultrasound
Images” Journal of Scientific Research,, J. Sci. Res. 3 (1),
81-89.
[2]. Shi Zhong, 2000,”Image Denoising using Wavelet
Thresholding and Model Selection”, Image Processing,
International Conference on, Volume: 3, 10-13 Sept. 2000
Pages: 262.
[3]. Suresh Kumar et al, 2010,”Performance Comparison of
Median and Wiener Filter in Image De-noising”,
International Journal of Computer Applications (0975 –
8887) Volume 12– No.4
[4]. Bhausaheb Shinde et al,2012, ” Study of Noise Detection
and Noise Removal Techniques in Medical Images”, I.J.
Image, Graphics and Signal Processing, 2, 51-60.
[5]. Dr.G.Padmavathi et al ,2009,” Performance analysis of
Non Linear Filtering Algorithms for underwater images”,
(IJCSIS) International Journal of Computer Science and
Information Security,Vol.6, No. 2,
[6]. N.Rajalakshmi and V.Lakshmi Prabha,2009,” Automated
Classification of brain MRI using color converted k-means
clustering segmentation and application of different kernel
functions with multi-class SVM”, 1st Annual International
Interdisciplinary Conference, AIIC 2013, 24-26.
[7]. E. Ben George et al, 2012,” MRI Brain Image
Enhancement Using Filtering Techniques”, International
Journal of Computer Science & Engineering Technology
(IJCSET), ISSN: 2229-3345 Vol. 3 No. 9
[8]. Tong Sun, 1995,”Analysis of two dimensional center
weighted median filters”, Multidimensional systems and
signal processing, 6,159-172.
[9]. J.M. Waghmare, 2013,”Removal of Noises In Medical
Images By Improved Median Filter”, The International
Journal Of Engineering And Science (IJES), ISSN(p): 2319 –
1805.
[10]. Bhausaheb Shinde et al, 2012,” Apply different
techniques to remove the speckle noise using medical
images”, International Journal of Engineering Research and
Applications (IJERA) ISSN: 2248-9622.](https://image.slidesharecdn.com/performanceanalysisofimagefilteringalgorithmsformriimages-140819063258-phpapp02/85/Performance-analysis-of-image-filtering-algorithms-for-mri-images-3-320.jpg)
Performance analysis of image filtering algorithms for mri images
- 1. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 438 PERFORMANCE ANALYSIS OF IMAGE FILTERING ALGORITHMS FOR MRI IMAGES Sivasundari .S1 , R. Siva Kumar2 , M.Karnan3 1 P.G scholar, Department of Computer and Communication, Tamilnadu College of Engineering, Coimbatore, TamilNadu, India 2 Professor, Department of Information Technology, Tamilnadu College of Engineering, Coimbatore, TamilNadu, India 3 Professor & Head, Department of Computer Science and Engineering, Tamilnadu College of Engineering, Coimbatore, TamilNadu, India Abstract In Image Processing, the digital images are much sensitive to noise which results due to the Image Acquisition errors and transmission errors.MRI images captured usually are prone to speckle noise, Gaussian noise, salt and pepper noise etc., Image filtering algorithms are applied over the noisy images to remove the noise and preserve the image details. In this work three different filtering algorithms such as Median filter (MF), Weiner filter (WF) and Center Weighted Median filter (CWM) are used to remove the noise present in the MRI images. The performance of these filters are compared using the statistical parameters such as Mean Square Error(MSE) and Peak Signal to Noise Ratio(PSNR).The study shows that the Weiner filters reconstructs a high quality image than Median and Center weighted Median filter. Keywords: Median filter (MF), Weiner filter (WF), Center Weighted Median filter (CWM), Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR). --------------------------------------------------------------------***----------------------------------------------------------------- 1. INTRODUCTION In image processing, it is important to obtain quality image for facilitating the image classification accuracy. As the poor image quality is an obstacle for efficient image processing. There is a need for the noise removal in medical images. In this work MRI images are used for the experiment. Magnetic Resonance Imaging are now becoming the most indispensible medical imaging tool used for analysis of different parts of the human brain.MRI produces the detailed information of the internal parts of the brain which can be used for the radiologists to detect the presence of tumors in the patient’s brain. The scan does not penetrate the adjacent healthy cells and thus preserves to be safest diagnostic tool. However one of the primary problems with MRI images are the presence of added unwanted noises like speckle noise, Gaussian noise,salt,Rician noise ,salt and pepper noise etc., These noise have much influence on the quality of the image which in turn degrades the performances of Feature Extraction, Reduction and Classification of the processed images. The noise have to be removed before these processing stages. Many image filtering algorithms are available in the literature. This paper evaluates the three filtering algorithms for different noise types. Simulation is done with the MATLAB R2010a.The organization of this paper is as follows. Section 2 is described with image filtering methods. Section 3 discuss about the Experiment results and evaluation and finally Section 4 gives conclusion. 2. FILTERING METHODS Noise is the unwanted variations or fluctuations of the image capture. These variations are caused due to errors in the sensors or data transmission which corrupts the image details either by causing brightness or frequency changes. Speckle noise are caused due to transmission errors. Rician noise are caused due to the Gaussian noise in the original frequency domain measurements [1]. The amplifier produces a standard model of noise which is additive, Gaussian. These noise are independent at each pixel and its intensity. Amplifier noise is a cause of the "read noise" of an image sensor that is, having constant noise level in dark areas of the image [2]. A salt-and- pepper noise will have dark pixels and bright pixels alternate bright and dark regions [2].Speckle noise is a granular noise degrades the quality due to transmission errors [3]. Conventionally used simplest methods of linear filtering algorithms include mean filters. The filtering is performed by using a mask over every pixel component in the image. Then by averaging the pixels which falls under the mask results in a single pixel to be studied. This technique also called as averaging filter. The main drawback of this filter is that it is poor in edge preserving. 2.1 Median Filter The Median filter is advantageous over mean filter and its a non-linear filtering technique, helps removing noise. Convolution methods like smoothing filters reduces noise but blurs the edge features. In order to preserve the edge features and [4] sharpness of the image we go for Median Filters. It
- 2. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 439 does excellent job in rejecting impulse noise, in which some individual pixels have extreme values. In Median filtering, for each pixel, the pixel values in the neighborhood window are organized according to the intensity and the middle value(Median) becomes the new value for the central pixel.i.e., each pixel is set to an ‘average’ of the pixel values in the neighborhood pixels rather than the mean. This method can be repeated if necessary since it causes only less blur of edges. Median filter is less sensitive than the mean filter to extreme values .The median filter (M) can be implemented by the formula: M(P1… P N) =Median (||P 1||2 …….|| P N||2) - (1) This is an optimal filter much suitable for de-noising images of various noise types[5].Causes no changes in the shift boundaries and no reduction in contrast Causes no changes in the shift boundaries and no reduction in contrast and proved to be more robust in the presence of noise. 2.2 Weiner Filter Weiner filtering carry out an optimal tradeoff between inverse filtering and noise smoothing. It removes additive noise and deblurring concurrently [6].This proves to be optimal in reducing the overall Mean Square Error (MSE).The operation involves two parts. One is inverse filtering and the other part is noise smoothing. Wiener filters belong to a kind of optimum linear filters with the noisy data as input which involves the calculation of difference between the desired output sequences from the actual output. Main purpose of this statistical approach is to diminish the noise present in a signal by comparing it with an estimation of the desired noiseless signal. Wiener filters are characterized by an assumption that signal and (additive) noise are stationary linear random processes and their spectral characteristics are calculated. The Performance can be measured using Minimum Mean-Square Error. The Wiener filter is implemented as: 𝑊 𝑚, 𝑛 = 𝐻∗ 𝑚,𝑛 𝑃𝑠 (𝑚,𝑛) 𝐻 𝑚,𝑛 2 𝑃𝑠 𝑚,𝑛 +𝑃 𝑛 (𝑚,𝑛) -(2) Dividing through by 𝑃𝑠 : 𝑊 𝑚, 𝑛 = 𝐻∗ 𝑚,𝑛 𝐻 𝑚,𝑛 2+ 𝑃 𝑛 (𝑚 ,𝑛) 𝑃 𝑠 𝑚 ,𝑛 -(3) Where H (m,n) = Degradation function , H* (m,n) = Complex conjugate of degradation function , Pn (m,n) = Power Spectral Density of Noise , Ps (m,n) = Power Spectral Density of un-degraded image. The term Pn/Ps can be interpreted as the reciprocal of the signal-to-noise ratio [5].The statistical parameters Mean Square Error (MSE) and Peak Signal-to-Noise ratio (PSNR) are used to evaluate the enhancement performance. 2.3 Center Weighted Median Filter (CWM) The CWM filter is an extension of Weighed median filter. Median filters can remove the noise by preserving the edges and sharpness which is suitable to reduce salt and pepper noise. In Weighted median filter, the weights of the filter are assigned based on intensity value of pixels in the MRI image. The weights used here are 0,0.1,0.2 and 0.3.Initially the weight of the pixel is set to 0,if the pixel intensity is 0.The weight assigned is 0.1 for the pixel intensity range between 1- 100 and 0.2 for pixel intensity range 101-200 else 0.3. The weights set are multiplied with pixel intensity and Median Filter is applied. This process is called Weighted Median Filter. If the center value of each window is set more weight, then its Center Weighted Median filters (CWM). The CWM filters are represented by two important parameters such as window size and center weight which influence the quality of filter. The center value is assigned large weight i.e. W(0,0)=2N+1, where N is a non-negative integer, if N>=0 then W(i,j)=1,i and j not equal to 0. The performances of CWM are better analyzed by authors [7,8].The value of weight must chosen carefully as it drives the performance of the filter. The image characteristics and type of noise should be considered while selecting the weight. The larger weight provides better result in preserving image and edge details but worse in noise removal [9]. 3. EXPERIEMENTS AND RESULTS These filtering algorithms has been tested with various types of noisy images implemented by using MATLAB 7.1.0.. The experimental results are tested in Intel Pentium IV 2.4 GHz processor with 256 MB RAM. Performance of the Median filter, Center Weighted Median filter, and Weiner filters are analyzed and evaluated. Table -1: shows the performance of listed filters S.no Filters MSE PSNR 1 Median Filter 148949 7.1991 2 CWM Filter 2.7819e+006 16.312 3 Weiner Filter 8.3636e+003 17.813
- 3. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 05 | May-2014, Available @ http://www.ijret.org 440 Fig -1 Input Image Fig -2 CWM Filtered Fig -3 Median Filtered Fig -4 Weiner Filtered 4. CONCLUSIONS In this paper image filtering algorithms are implemented on MRI images to remove the different types of noise that are either present in the image during acquisition or injected in to the image during transmission. MRI images when captured usually have Gaussian noise and speckle noise. In this work, three different image filtering algorithms are compared for the noisy images. The performances of the filters are measured using the Peak Signal to Noise Ratio (PSNR) and Mean Square Error (MSE). The Weiner filter gives desirable results with large PSNR value ensuring high image enhancement. REFERENCES [1]. M. N. Nobi and M. A. Yousuf, 2011,” A New Method to Remove Noise in Magnetic Resonance and Ultrasound Images” Journal of Scientific Research,, J. Sci. Res. 3 (1), 81-89. [2]. Shi Zhong, 2000,”Image Denoising using Wavelet Thresholding and Model Selection”, Image Processing, International Conference on, Volume: 3, 10-13 Sept. 2000 Pages: 262. [3]. Suresh Kumar et al, 2010,”Performance Comparison of Median and Wiener Filter in Image De-noising”, International Journal of Computer Applications (0975 – 8887) Volume 12– No.4 [4]. Bhausaheb Shinde et al,2012, ” Study of Noise Detection and Noise Removal Techniques in Medical Images”, I.J. Image, Graphics and Signal Processing, 2, 51-60. [5]. Dr.G.Padmavathi et al ,2009,” Performance analysis of Non Linear Filtering Algorithms for underwater images”, (IJCSIS) International Journal of Computer Science and Information Security,Vol.6, No. 2, [6]. N.Rajalakshmi and V.Lakshmi Prabha,2009,” Automated Classification of brain MRI using color converted k-means clustering segmentation and application of different kernel functions with multi-class SVM”, 1st Annual International Interdisciplinary Conference, AIIC 2013, 24-26. [7]. E. Ben George et al, 2012,” MRI Brain Image Enhancement Using Filtering Techniques”, International Journal of Computer Science & Engineering Technology (IJCSET), ISSN: 2229-3345 Vol. 3 No. 9 [8]. Tong Sun, 1995,”Analysis of two dimensional center weighted median filters”, Multidimensional systems and signal processing, 6,159-172. [9]. J.M. Waghmare, 2013,”Removal of Noises In Medical Images By Improved Median Filter”, The International Journal Of Engineering And Science (IJES), ISSN(p): 2319 – 1805. [10]. Bhausaheb Shinde et al, 2012,” Apply different techniques to remove the speckle noise using medical images”, International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622.