i jc ta 2014050532

8
Abstract: Visual perceptions can be represented in pictorial form which can be in 2D or 3D known as Image. Collective forms of pixels constitute a form of Image. Image can be distorted by various types of noise which blur and loss of image details may take place. Thus number of filtering techniques are imposed which preserve the blurness of image and even edges but overall cannot preserve the image details. In these existing techniques lots of changes are performed but still denoising in images require improvement. Nowadays soft computing techniques are used instead of filters for the removal of noise. Fuzzy system rules are imposed for the removal of noise but this system has various disadvantages such as training is not provided by this system. This is overcome by the neural network. Present paper focuses on hybrid working of two soft computing techniques fuzzy system and neural network which overcome the disadvantages of each other. New technique used is named as Neuro-Fuzzy technique. This technique overcomes the disadvantages of the existing filters namely median filter, weighted median filter, switching median filter, recursive median filter, adaptive median filter, averaging median filter and so on. Comparison results are shown on the basis of various parameters namely PSNR, MSE, RMSE Keyword: Denoising; PSNR; MSE; RMSE; Neuro-Fuzzy I. INTRODUCTION Image can be defined as an artifact else which can be said as a visual representation of the visual perceptions recorded by camera or the scanning products. These recorded perceptions are than further digitized for the conversion of particular recorded data to the form which can be stored in the memory of the computer or the storage media like hard disk, floppy disk or the compact disk. During capturing of an image or processing of an image noise can be edited in the image which may distort the image and results in loss of image details. Different types of the digital images are likely to be intruded by number of varieties of noise named as speckle noise, impulsive noise (Salt and pepper noise), Gaussian noise, gamma noise etc. Image edges and corners contain high frequency content and details. Thus image edges should be preserved. For preservation of image details various filters are introduced. These filters remove the noise from images but various filters fail in protecting the sharp edges and the image details including the information of the format along with the time of capturing the image and its size. Various techniques are proposed for the enhancements of images which contribute in the denoising of image. Filtering techniques include linear filtering and Non-linear filtering. Without explicitly identifying noise, it is removed by nonlinear filters. In low pass filters the upper area is occupied by the frequency spectrum. Thus due to this assumption it is used on the group of pixels of the spatial filters. Generally noise is eliminated by spatial filters to a reasonable extent, but edges in images remain invisible. There are various filters proposed which can remove these all above disadvantages are weighted median [39], relaxed median [18] and the rank conditioned rank selection [6]. In the removal of salt and pepper noise the Standard Median filtering [34] process is useful as it replaces the central pixel value by the median pixel values of the filtering window. Thus this filter removes the noise along with the degrading performance as it removes the details of the image and even the blurred edges are removed. Proposing the Weighted median filters [28] is due to the removal of disadvantages of the standard median filters. These filters use the approach of the selecting the filtering window and provide the more weight to the pixels selected. There is a vast disadvantage of weighted median filters that these filters have difficulty to differentiate the noisy pixels and noise free pixels. The disadvantages of above standard and the weighted median filter are overcome by the Switching Median filter. Middle pixel of the filtration window is classified by the switching pulse detector. In the case of central pixel is corrupted than for further filtration procedure the standard median filter is used. Thus the replacement of middle pixel is made by the value of the median filter. Else if the central pixel is uncorrupted it remains the same. It depends on the pulse sensor that working of this type of filtering process is improved and more appropriate than the standard median filter. Switching Median filter is improved for the further improvement in the removal of noise thus for such improvement various filters namely median filter three states [11] and the median filter of the multi-state [9] are proposed which also includes the features of the weighted median filter. As these filters can handle the complex calculations thus they are more appropriate according to the execution process performed than the standard median filter and the weighted median filter. Even there are number of filter for improving the performance such as median filter of progressive switching [43] which performs the filtering process in the iterative method by the detection of the corrupted pixels and then removal of that particular pixel. Due to its iterative nature it has a great disadvantage which is high computational complexity. To overcome above disadvantages Adaptive median filter [10] and Ranking median filter [1] are proposed but still enhancement is required in images for complete removal of noise. Image Denoising using Improved Neuro-Fuzzy based Algorithm Amaninder Kaur Brar 1 , Vikas Wasson 2 1 Research Scholar, 2 Assistant Professor Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779 IJCTA | Sept-Oct 2014 Available [email protected] 1772 ISSN:2229-6093

Upload: ali-hassan

Post on 21-Nov-2015

215 views

Category:

Documents


1 download

DESCRIPTION

jjf jgy tykghj j kgh k fjf kf f

TRANSCRIPT

  • Abstract: Visual perceptions can be represented in pictorial form which can be in 2D or 3D known as Image. Collective forms of

    pixels constitute a form of Image. Image can be distorted by

    various types of noise which blur and loss of image details may

    take place. Thus number of filtering techniques are imposed

    which preserve the blurness of image and even edges but overall

    cannot preserve the image details. In these existing techniques

    lots of changes are performed but still denoising in images

    require improvement. Nowadays soft computing techniques are

    used instead of filters for the removal of noise. Fuzzy system rules

    are imposed for the removal of noise but this system has various

    disadvantages such as training is not provided by this system.

    This is overcome by the neural network. Present paper focuses on

    hybrid working of two soft computing techniques fuzzy system

    and neural network which overcome the disadvantages of each

    other. New technique used is named as Neuro-Fuzzy technique.

    This technique overcomes the disadvantages of the existing filters

    namely median filter, weighted median filter, switching median

    filter, recursive median filter, adaptive median filter, averaging

    median filter and so on. Comparison results are shown on the

    basis of various parameters namely PSNR, MSE, RMSE

    Keyword: Denoising; PSNR; MSE; RMSE; Neuro-Fuzzy

    I. INTRODUCTION

    Image can be defined as an artifact else which can be

    said as a visual representation of the visual perceptions

    recorded by camera or the scanning products. These recorded

    perceptions are than further digitized for the conversion of

    particular recorded data to the form which can be stored in the

    memory of the computer or the storage media like hard disk,

    floppy disk or the compact disk. During capturing of an image

    or processing of an image noise can be edited in the image

    which may distort the image and results in loss of image

    details. Different types of the digital images are likely to be

    intruded by number of varieties of noise named as speckle

    noise, impulsive noise (Salt and pepper noise), Gaussian noise,

    gamma noise etc. Image edges and corners contain high

    frequency content and details. Thus image edges should be

    preserved. For preservation of image details various filters are

    introduced. These filters remove the noise from images but

    various filters fail in protecting the sharp edges and the image

    details including the information of the format along with the

    time of capturing the image and its size.

    Various techniques are proposed for the enhancements

    of images which contribute in the denoising of image.

    Filtering techniques include linear filtering and Non-linear

    filtering. Without explicitly identifying noise, it is removed by

    nonlinear filters. In low pass filters the upper area is occupied

    by the frequency spectrum. Thus due to this assumption it is

    used on the group of pixels of the spatial filters. Generally

    noise is eliminated by spatial filters to a reasonable extent, but

    edges in images remain invisible. There are various filters

    proposed which can remove these all above disadvantages are

    weighted median [39], relaxed median [18] and the rank

    conditioned rank selection [6]. In the removal of salt and

    pepper noise the Standard Median filtering [34] process is

    useful as it replaces the central pixel value by the median pixel

    values of the filtering window. Thus this filter removes the

    noise along with the degrading performance as it removes the

    details of the image and even the blurred edges are removed.

    Proposing the Weighted median filters [28] is due to the

    removal of disadvantages of the standard median filters. These

    filters use the approach of the selecting the filtering window

    and provide the more weight to the pixels selected. There is a

    vast disadvantage of weighted median filters that these filters

    have difficulty to differentiate the noisy pixels and noise free

    pixels. The disadvantages of above standard and the weighted

    median filter are overcome by the Switching Median filter.

    Middle pixel of the filtration window is classified by the

    switching pulse detector. In the case of central pixel is

    corrupted than for further filtration procedure the standard

    median filter is used. Thus the replacement of middle pixel is

    made by the value of the median filter. Else if the central pixel

    is uncorrupted it remains the same. It depends on the pulse

    sensor that working of this type of filtering process is

    improved and more appropriate than the standard median

    filter. Switching Median filter is improved for the further

    improvement in the removal of noise thus for such

    improvement various filters namely median filter three states

    [11] and the median filter of the multi-state [9] are proposed

    which also includes the features of the weighted median filter.

    As these filters can handle the complex calculations thus they

    are more appropriate according to the execution process

    performed than the standard median filter and the weighted

    median filter. Even there are number of filter for improving

    the performance such as median filter of progressive switching

    [43] which performs the filtering process in the iterative

    method by the detection of the corrupted pixels and then

    removal of that particular pixel. Due to its iterative nature it

    has a great disadvantage which is high computational

    complexity. To overcome above disadvantages Adaptive

    median filter [10] and Ranking median filter [1] are proposed

    but still enhancement is required in images for complete

    removal of noise.

    Image Denoising using Improved Neuro-Fuzzy

    based Algorithm

    Amaninder Kaur Brar1, Vikas Wasson

    2 1Research Scholar,

    2Assistant Professor

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1772

    ISSN:2229-6093

  • Linear filters are also introduced for the removal of

    noise. Linear filters were primary tools which were initially

    used for many electrical signaling and processing of image

    applications due to the presence of analytical theory of design

    and analysis. For the corruption of image by the AWGN [17]

    type noise these filters show good performance for those

    particular images. In the case of mean square error for the

    preservation of images from the Gaussian noise these filters

    also can be used. But the great disadvantage of linear filter is

    that they cannot deal the non-linearities of the image

    formation model. These filters also abolish the lines, also the

    important information of the image even blur the sharp edges.

    Wiener filtering [3] is example of linear filtering.

    In present work new improved hybrid technique is

    proposed named Neuro-Fuzzy technique which denoises the

    images along with the preservation of edges and its details.

    This technique is beneficial as it attacks on the noisy pixels

    only and keeping the rest of pixels preserved. Hence preserves

    the details of image.

    In this paper the various sections are distributed as in

    section II the proposed technique is described in which the

    pre-processing step followed by the In section III results are

    discussed in the graphical and tabular representations. Last

    section IV constitutes of conclusion and future scope of the

    proposed technique.

    II. RELATED WORK

    In literature there are various techniques and filtering

    methods discussed which are proposed to denoise the image.

    But each filtering method has the disadvantage which is

    overcome by other and at last for denoising soft computing

    techniques are implanted which preserve edges and image

    details and also remove the noise from images but still some

    improvements are required in these techniques which are

    discussed in this literature survey.

    In paper 2013Ahmed F. and Das S. [2]Removal of High Density Salt and Pepper Noise in Images with an

    Iterative Adaptive Fuzzy Filter using alpha-trimmed Mean Proposed a novel adaptive, iterative, fuzzy filter for denoising

    various images which are including the noisy pixels. It

    operates in two stages - detection of noisy pixels with an

    adaptive fuzzy detector followed by denoising using a

    weighted mean filter on the good pixels in the filter window. Fuzzy filters when used in adaptive setting are simple and

    quite efficient. The filter uses -trimmed mean which is effective for measure of central tendency than the classical

    mean than the classical mean or median measures in the

    context of impulse noise removal thus, filter is shown to be

    robust to very high levels of noise, retrieving meaningful

    detail at noise levels as high as 97%. Alpha-trimmed mean

    computes the mean of a set of elements after trimming the top

    and bottom /2 elements of the set. In paper 2012Aldinucci M. and Spampinato C.[4]A

    Parallel Edge Preserving Algorithm for Salt and Pepper Image

    Denoising proposed a two-phase filter for removing salt and pepper noise. Noisy pixels Identification and Noisy pixel

    Restoration are two phases for removal of noise. In the initial

    step, detection of the corrupted pixels is done by an adaptive

    median filter. In the second phase, restoration of the pixels is

    performed by the regulation method. The first step identifies

    noisy pixels by means of a modified Adaptive Median Filter

    (AMF) classifier; whereas second step restores them using a

    variational approach which is solved by using regularization.

    In this algorithm initial phase exhibits a linear execution time

    with respect to the total number of pixels. However, for the

    tested images, the detect phase is two-three orders of

    magnitude faster than a single iteration of the second step. The

    second phase shows the complexity of noisy pixels as well as

    number of iterations required to reach one of the convergence

    criterions. Non-noisy pixels are not changed in denoise phase

    only changes are provided to the noisy pixels. Mean absolute

    error is improved by this algorithm. It can overall complete in

    terms of quality of restoration time and execution time.

    In paper 2007 Blu T. and Luisier F. [7] The SURE-LET Approach to Image Denoising proposed a latest technique for denoising the images. SURE is the Steins Unbiased Risk Estimate. Possibility of this purposed technique

    is made by the presence of the best unbiased estimate of the

    Mean Square Error (MSE). MSE is calculated ratio of the

    initial image acquired without noise and the denoised image in

    which noise is removed. In the case of orthogonal transforms

    this technique is used. By the various experimental results it is

    observed that there are some improvements required in this

    SURE-LET algorithm.

    In paper 2001 Eng H. and Ma K. [13] Noise Adaptive Soft Switching Median Filter a filter is proposed which doesnt corporate with the fuzzification concept. This proposed filter is also known as Noise Adaptive Soft

    Switching Median (NASM) filters. This technique is purposed

    to achieve the filtration to the best by effectively removing the

    noise along with preservation of the robustness and variations

    in the noise density. There are two stages in this proposed

    technique. Each pixels characteristic is identified. In the initial level the numbers of real pixels are detected as

    uncorrupted pixels. In the next levels the rest pixels which are

    not identified are discriminated. The discrimination is done as

    such: isolated salt and pepper noise, non-isolated salt and

    pepper noise or edge pixel. In the next stage the exploitation

    of the fuzzy logic is done in which filtering scheme includes

    no filtering to identified uncorrupted pixels known as fuzzy

    weighted median (FWM). For the classification of the each

    pixel being noise free a scheme is developed namely soft-

    switching noise-detection.

    In paper 2010 Fazli S. et.al [14] Complex PDE Image Denoising Based on Particle Swarm Optimization Artificial Intelligence techniques are used along with Partial Differential

    Equations (PDE) for denoising of image. This technique is

    different from the previous techniques using PDE as in this

    method for the various complicated PDE parameters Particle

    Swarm Optimization (PSO) is used. These PDE parameters

    are tuned by reducing the Structural Similarity (SSIM)

    measure. Proposed method obtains the simulation results of

    standard images. Edge shock filter is present in PDE for

    sharpening the edges.

    In paper 2007 Feng D. et.al. [15] High Probability Impulse Noise-Removing Algorithm Based on Mathematical

    Morphology a novel filter is proposed for the large contingency removal of the impulsive noise. Initially, the

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1773

    ISSN:2229-6093

  • impulsive pixels (noisy pixels) are identified by the impulse

    noise detector by using the mathematical residues. Block

    smart erase method is used to remove the back and white

    blocks which may reduce the image quality. Open close

    sequence filter is applied in this approach. This filter is used in

    such a way that opening of the filter results in the removal of

    the salt impulsive noise and removal of the pepper impulsive

    noise is caused by the closure of the filter. Thus this filter is

    successful in the removing the impulsive noise in the images

    by removing black and white spots separately, which makes

    this process more impressive in denoising of the image.

    In paper 2013 Gargouri A. and Masmoudi D.S [16]

    Neural Network Based image denoising with Pulse Mode Operations and Hybrid on-chip learning algorithm proposed a pulse mode neural network (PMNN) based image denoising

    operation. In the purposed approach a hybrid learning

    algorithm, in which, they applied the K-means algorithm to

    adjust the centers positions of the basic activation functions, as

    well as the back-propagation algorithm to update the

    connection weights. However, early pulse mode

    implementation suffers from some constraints due to the

    complexity of the on-chip learning ability, since the back-

    propagation algorithm is probably the most used, which costs

    much of hardware resources. In this paper hardware

    implementation is presented of Radial Basis Function (RBF)

    Neural Network based removal of image. In this image

    denoising function was applied. There is an ability of

    proposed algorithm to gain the complete information from the

    corrupted or infected image, with respect to the PSNR metric.

    In paper 2009 Kumar V. et al. [24] Robust Statistics Based Algorithm to Remove Salt and Pepper Noise in Images a robust estimation filter is purposed for the removal of the

    impulse noise up to the noise density of 70% efficiently. This

    filtration process is performing better comparatively to the

    previous filters discussed earlier such as standard median

    filter, weighted median filter, recursive weighted median filter

    and the progressive switching median filter along with the

    decision based algorithm. The function of algorithm is the

    detection of the noisy pixel first as certain pixels are noisy by

    salt and pepper noise while remaining pixels remains

    uncorrupted. This will satisfactorily remove the impulsive

    noise density till the medium level along with the preservation

    of edges of an image.

    In paper 2006Luo W. [26] An Efficient Detail-Preserving Approach for Removing Impulse Noise in Images Presents an algorithm which preserves image details and

    removes the impulsive noise from the noisy images. This

    algorithm is based on the alpha-trimmed mean. This algorithm

    consists of the specific case of the order-statistics filter.

    Instead of pixel value estimation this process is used in

    detection of the impulsive noise. When identification of the

    corruptive image is made, replacement of its value is done by

    the continuous succession of the initial value and the median

    calculated for the local window. In proposed algorithm the

    filtering process is applied only to corruptive pixels either to

    all pixels in an image. Purposed algorithm consists of three

    steps: Impulse noise detection, refinement and impulse noise

    cancellation. Thus by following above steps this algorithm

    replacement is made of the values which are identified as

    noisy pixels with the predicted values. As the yielded

    subjective quality is best with the consideration to the

    impulsive noise suppression and the protection of the image

    details thus it is said that performance is better than earlier

    existing median filters.

    In paper 2012 Thirilogasundari.V, Suresh babu.V,

    Agatha Janet.S [34] Fuzzy Based Salt and pepper noise removal using adaptive switching median filter proposed a new filter which concentrate on removal of impulse noise

    caused due to fault in memory units, imperfections in

    transmission medium, errors in transmission system. This

    filter overcome the drawbacks of previous systems such as this

    system removes only noisy or corrupted pixels and preserves

    the pixels which are noise less. Proposed system is based on

    switching median filter. It consists of two stages namely:

    detection stage and filtering stage. Neighborhood mapping

    based algorithm detects the corrupted pixels in detection stage

    and in next stage of filtering fuzzy membership function is

    used. This is compared by the existing techniques by various

    parameters. This proposed algorithm has various advantages

    on the removal of the impulsive noise along with protection of

    the other image details, edges and textures and neither require

    further tuning but same filtering process cannot be used in

    case of other type of noise removal such as Gaussian noise etc.

    In paper 2010 Wang X. et.al [36] Image Denoising Based on Translation Invariant Directional Lifting for the reduction of artifacts in various denoising results translation

    invariant directional lifting is proposed by employing the

    cycle spinning based technique. For the achievement of the

    better results with the lower intricacy in this technique 2-D

    Gabor filter is adopted which also provide better denoising

    results. The results are calculated in the form of PSNR value

    for objective and SSIM evaluation which is subjective

    performance.

    In paper 2001Wu H.R. and Chen T. [38] Adaptive Impulse Detection Using Centre-Weighted Median Filters adaptive impulse detector was proposed in which centre

    weighted median filter is used for the efficient ejection of

    impulsive noise. In this paper a novel adaptive operator is

    devised in which estimation is based on the difference

    between the initial pixel and the output results of centre

    weighted median filters. Performance of this method is better

    only in the case of image corruption till the 50% with the

    impulsive noise.

    In paper 2012 Yang X. et.al [40] Image Denoising Based on Support Vector Machine presents the algorithm which performs denoising of an image and prevents the fine

    information of the image. The support vector machine based

    method is used for the removal of noise and is perfect method

    for denoising as it retains the image details along with the

    removal of the noise. Support vector machine is a machine

    learning which is based on the statistical learning theory and

    thus this method solves various classification problems. In this

    method wavelet transform is used which consists of various

    signaling and performing on the different forms of wavelet

    transforms for the construction of the corresponding rules.

    In paper 2009 Zaho T. et.al [41] Approach of Image Denoising Based on Discrete Multi-Wavelet Transform by using multi-wavelet transform new approach was proposed to

    perform remote sensing image denoising. Extraction of the

    signal from the corrupted data a rise is given to wavelet

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1774

    ISSN:2229-6093

  • thresholding method by wavelet theories. Multi-wavelets are

    made more appropriable for various image processing

    (denoising) by orthogonality, symmetry and short support

    properties. This technique is better than the previous

    techniques as less distortion of spatial characteristics occur in

    image denoising. Wavelet based method perform both in

    objectively and subjectively form as the result of experiments

    performed.

    In paper 2002 Zhang S. and Karim A. [44] A New Impulse Detector for switching Median Filters proposed a latest technique for detecting impulse in which switching

    median filters are present. These median filters are based on

    the least absolute value for the four convolutions. In this

    presented filtration the focus is made over the preservation of

    the lines and provides improved detection of the black and

    white pixels means distorted pixels. As the previous median

    filters were unable to differentiate between the thin lines and

    the impulses but the presented filtration method distinguish

    between these thin lines and impulses and hence is successful

    in removal of the thin lines which results in better

    performance than the existing filtering methods. Noise is

    measured by the MSE curves which show increase of noise

    (after 30% approx.) by significant increase in graph. Still lots

    of improvement is required in impulse detector as in this

    process normalization is required thus the performance of

    filter can be improved by making various significant changes.

    III. PROPOSED ALGORITHM

    A. Detection of noisy pixel in an input acquired image: An

    image is acquired by input source such as camera. A range is

    estimated in which the uncorrupted pixels may lie. The pixels

    which not lie in an estimated range are detected as noisy or

    corrupted pixels.

    Algorithm: A filtering window is slide over the highly

    contaminated image.

    ALGORITHM: SLIDING WINDOW

    BEGIN

    1: ACQUIRE (IMAGE)

    2: ADD NOISE (GAUSSIAN NOISE, SALT AND PEPPER NOISE, SPECKLE NOISE)

    3: SELECT 2*2 WINDOW (N)

    4: SLIDE N

    5: IF

    NEIGHBORING PIXELS SELECTED INCREASED

    SELECT 3*3 WINDOW

    6: ELSE

    SELECT 5*5 WINDOW

    7: ELSE

    SELECT 7*7 WINDOW

    PIXEL DETECTED (P)

    END

    During sliding the window the noisy pixels are detected by

    calculating the maximum, minimum and the median values.

    ALGORITHM: DETECTED CORRUPTED PIXELS

    BEGIN

    1: CALCULATE (MAX, MIN, MED)

    2: IF

    MAX

  • Filter and Adaptive Median Filter. This comparison process is

    performed by using various performance parameters such as

    PSNR, MSE and RMSE. Algorithm for PSNR (Peak Signal to

    Noise Ratio), MSE (Mean Square Error) and RMSE (Root

    Mean Square Error) is shown below.

    Algorithm: Calculation of PSNR, MSE and RMSE

    ALGORITHM: PSNR (PEAK SIGNAL TO NOISE RATIO),

    RMSE (ROOT MEAN SQUARE ERROR)

    WHERE MSE MEAN SQUARE RATIO

    BEGIN

    1: INPUT ARRAY (X, Y)

    2: SIZE OF ARRAY= N*M

    3: =1

    , (, ) 2 1=0

    1=0

    4: USING STEP 3 CALCULATE PSNR

    5: = 1010 2

    6: USING STEP 3 CALCULATE RMSE

    7: =

    END

    Figure 1: Proposed Methodology

    IV. RESULTS AND DISCUSSIONS

    This section includes the results which are obtained after

    implementing Neuro-Fuzzy technique which denoise the

    images from the various types of noise. In the initial part of

    this section screenshots representing different outputs by the

    different filtering techniques and in the next section various

    parameters used for performance measurement is presented

    along with the comparison results for various filtering

    techniques with the different parameters in tables. Finally for

    better understanding of the results these tabular form results

    are presented in form of Graphs.

    Initially image is acquired or selected from the number

    of dataset. After the selection of image, addition of noise is

    made to an image. Gaussian, Salt & pepper or Speckle noise

    either can be edited to the image. Finally the filter is selected

    from the image with which the image is to be filtered. These

    all filters are introduced by using Matlab implementation

    A. Denoised Image Outputs

    (a) Original Image

    (b) Filtered Image by Median Filter

    (c) Filtered Image by Averaging Filter

    (d) Filtered Image by Neuro-Fuzzy

    Technique

    B. Comparison Tables

    In this section comparison of various techniques is shown by

    the various parameters namely MSE, RMSE and PSNR. The

    tabular results are represented such as:

    Acquire an Image

    Noise is added in Image

    Filtering Window is slide over the

    corrupted Image

    Corrupted pixels are detected

    Uncorrupted pixels

    are kept as such

    Noise is removed

    from corrupted

    pixels using Neuro-

    Fuzzy Technique

    Denoised Image

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1776

    ISSN:2229-6093

  • Table 1: Performance of MSE for different filtering technique on pepper

    image

    NOISE

    DENSITY %

    50%(NOISE

    DENSITY)

    80%(NOISE

    DENSITY)

    90%(NOISE

    DENSITY)

    MEDIAN

    FILTER

    492.4 973 1253.02

    AVERAGING

    FILTER

    369.2 873.34 1053.67

    ADAPTIVE

    FILTER

    257.1 480.06 986.0

    PROPOSED

    NEURO-FUZZY

    149.79 297.99 640.42

    AVERAGE 317.12 656.09 983.28

    Table 2: Performance of PSNR for different filtering techniques on pepper

    image

    NOISE

    DENSITY %

    50%(NOISE

    DENSITY)

    80%(NOISE

    DENSITY)

    90%(NOISE

    DENSITY)

    MEDIAN

    FILTER

    23.20 19.48 18.19

    AVERAGING

    FILTER

    25.68 20.43 17.65

    ADAPTIVE

    FILTER

    28.02 26.31 25.38

    PROPOSED

    NEURO-FUZZY

    33.32 28.27 26.69

    AVERAGE 27.51 23.62 21.98

    Table 3: Performance of RMSE for different filtering techniques on pepper

    image

    NOISE

    DENSITY %

    50%(NOISE

    DENSITY)

    80%(NOISE

    DENSITY)

    90%(NOISE

    DENSITY)

    MEDIAN

    FILTER

    22.20 31.19 35.39

    AVERAGING

    FILTER

    19.21 29.55 32.46

    ADAPTIVE

    FILTER

    16.03 21.91 31.40

    PROPOSED

    NEURO-FUZZY

    12.24 17.26 25.30

    AVERAGE 17.42 24.98 31.13

    C. Comparison Graphs

    In this section comparison of various tabular results is

    represented in the graphical form which shows the results

    clearly. These graphical representations are such as:

    Graph 1: Graphical representation of MSE for different filtering techniques on pepper image.

    Graph 2: Graphical representation of PSNR for different filtering technique

    pepper image

    Graph 3: Graphical representation of RMSE for different filtering techniques

    on pepper image

    V. CONCLUSION AND FUTURE SCOPE

    A. Conclusion

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1777

    ISSN:2229-6093

  • In this research work a new hybrid technique is described

    which is known as Neuro-Fuzzy technique. In this technique

    combined features of neural network and fuzzy technique are

    applied for the removal of noise to preserve image details and

    edges. Thus this proposed filter is seen a quite effective in

    eliminating different types of noise such as salt and pepper

    noise, Gaussian noise and speckle noise along with preserving

    the edges. The efficacy of this technique is illustrated by

    applying on number of test images as shown in above results

    and discussions. In terms of qualitative and quantitative

    measures proposed technique outperforms the existing based

    filters with which it is compared such as Median filter,

    averaging filter, adaptive filter etc. the images filtered are

    quite pleasant for visual perception. It is also quite suitable for

    the real time implementations due to its adaptive nature.

    B. Future Scope

    The proposed technique removes the noise along with

    preservation of edges quite efficiently and effectively. But as

    neural network requires training of the system which takes

    some time to complete the training of the system. Along with

    the increment of layers of neural network the training time

    increases thus some improvement is required to decrease the

    time of the training or learning process.

    Present work is done on 2-D images but in future this can

    be processed on the 3-D images

    In future the time for training of neural network can be

    reduced to more extent.

    As the present technique is implemented on various sample

    images but in future this removal technique can be

    implemented on various videos.

    ACKNOWLEDGMENT

    I would like to express the deepest appreciation to my

    supervisor Er.Vikas Wasson, for the useful comments,

    remarks and engagement through the learning process of my

    research. Without his guidance and persistent help this thesis

    would not have been possible. His patience, motivation,

    enthusiasm, support and immense knowledge helped me to go

    through my research and finish this research paper. His

    guidance helped me in all the time of research and writing of

    this paper. I hope that one day I would become as good an

    advisor to my students as he has been to me. I would also like

    to express gratitude to Chandigarh University for providing

    me great opportunity to share my innovative ideas on topic

    Image Denoising using Improved Neuro-Fuzzy based Algorithm.

    REFERENCES

    [1] Abreu E, Lightstone M, Mitra S.K, and Arakawa K, A new efficient approach for the removal of impulse noise from highly corrupted

    images, IEEE Transactions Image Processing, Vol. 5, no. 6, pp. 10121025, January 1996.

    [2] Ahmed F. and Das S. Removal of High Density Salt and Pepper Noise in Images with an Iterative Adaptive Fuzzy Filter using alpha-trimmed

    Mean IEEE Transactions on Fuzzy Systems, Issue 99,October 2013. [3] A.K.Jain, Fundamentals of digital image processing Prentice-Hall, 1989 [4] Aldinucci M., Spampinato C. A Parallel Edge Preserving Algorithm for

    Salt and Pepper Image Denoising IEEE Image Processing Theory, Tools

    and Applications (IPTA), 3rdInternational conference, pp. 97-104, October

    2012. [5] Arce G. and Paredes J., Recursive Weighted Median Filters Admitting

    Negative Weights and Their Optimization, IEEE Transactions on Signal Processing, Vol. 48, No. 3, pp. 768-779, 2000.

    [6] Ben Hamza A, Luque P, Mattinez J and Roman R, Removing Noise and Preserving Details with Relaxed Median Filters, Springer, Journal of Mathematical Imaging and Vision, Vol. 11, No.2, pp. 161-177, October 1999

    [7] Blu T. and Luisier F. The SURE-LET Approach to Image Denoising IEEE Transactions on Image Processing, Vol. 16, No. 11, November 2007.

    [8] Bovik A.C., Hand book of Image and Video Processing, Academic Press, 2000.

    [9] Chen T. and Ren Wu H, Space Variant Median Filters for the Restoration of Impulse Noise Corrupted Images, IEEE trans. on circuits and systems II: Analog and Digital Signal Processing, Vol. 48, No. 8, pp.784 -789, August 2001.

    [10] Chen T and Wu H.R, Application of partition-based median type lters for suppressing noise in images, IEEE Transactions Image Process, Vol. 10, No. 6, pp. 829836, January 2001.

    [11] Chen T, Ma K.K, and Chen L.H, Tri-state median lter for image denoising, IEEE Trans. Image Process., Vol. 8, No. 12, pp. 18341838, December1999.

    [12] Elad M. et al. Image Sequence Denoising via Sparse and Redundant Representations, IEEE Transactions on Image Processing, Vol. 18, No. 1, January 2009

    [13] Eng H. and Ma K.Noise Adaptive Soft Switching Median FilterIEEE Transactions on Image Processing, Vol. 10, No.2 February 2001.

    [14] Fazli S., Bouzari H. and Moradi H. Complex PDE Image Denoising Based on Particle Swarm Optimization IEEE International congress on Ultra Modern Telecommunications and Control Systems and Workshops

    (ICUMT, pp. 364-370,October 2010

    [15] Feng D. et.al. [26] High Probability Impulse Noise-Removing Algorithm Based on Mathematical Morphology IEEE signal processing letters, Vol. 14 No. 1 January 2007.

    [16] Gargouri A., Masmoudi D.S. Neural Network Based image denoising with Pulse Mode Operations and Hybrid on-chip learning

    algorithmIEEE Transactions Image Process, Vol. 6, pp. 978-984, June 2013

    [17] Gonzalez R.C. and Woods R.E., Digital Image Processing, Addison-

    Wesley Publishing Company, 2002.

    [18] Hardie R.C and Barner K.E, Rank conditioned rank selection filters for signal restoration, IEEE Trans. Image Processing, Vol. 3, pp.192206, March 1994.

    [19] Hong-qio Zhang, Xin-jun Ma, Wu-Ning, A New Filter Algorithm of Image Based on Fuzzy Logical, IEEE Computer Science and Society (ISCCS), pp.315-318, July 2011

    [20] Jain M. et al. Effect of Blur and Noise on Image Denoising based on PDE, International Journal of Advanced Computer Research (IJACR),Vol. 3, No. 1, Issue-8, March 2013.

    [21] Kaur A. and Wasson V., Image Denoising using Improved Neuro-Fuzzy Based Algorithm International Journal of Advanced Research in Computer Science and Software Engineering,Vol.4, Issue 4, pp. 1072-

    1075, April 2014. [22] Ko S.J and Lee Y.H, Center weighted median lters and their

    applications to image enhancement, IEEE Transactions Circuit System, Vol. 38, no. 9, pp. 984993, September 1991.

    [23] Kuang K. Noise Adaptive Soft-Switching Median Filter, IEEE Transactions on Image Processing, Vol.10, No.2, pp. 242 251, October 2001.

    [24]Kumar V. et al. Robust Statistics Based Algorithm to Remove Salt and Pepper Noise in Images IEEE signals processing lettersVol.3, No.5, pp 278-285, March 2009.

    [25] Lang C. et al. Combined Transform Image Denoising based on Morphological Component Analysis, IEEE Multimedia Technology (ICMT), International Conference, pp. 4871-4874, July 2011.

    [26] Luo W. An Efficient Detail-Preserving Approach for Removing Impulse Noise in Images, IEEE signals processing letters, Vol. 13, No.7, pp. 413 416, December 2006.

    [27] Muresan D.D, and Parks T.W, Adaptive principal components and image denoising, in: Proceedings of IEEE International conference on Image processing, Vol. 1, Barcelona, Spain, 14-17, pp 101-104, September, 2003.

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1778

    ISSN:2229-6093

  • [28] Neuvo Y, Yli-Harja O and Astola J, Analysis of the properties of median and weighted median lters using threshold logic and stack lter representation, IEEE Transactions Signal Process,Vol. 39, No. 2, pp. 395410, February 1991.

    [29] Pushpavalli R., Sivarajde G. Image Denoising Using A New Hybrid Neuro-Fuzzy Filtering Technique International Journal of science & Research,Vol. 2, Issue 5, May 2013.

    [30] Rosenfield A. and Kak A.C. Digital Picture Processing, 2nd Edition, Academic Press, New York.

    [31] Scott M An Introduction to Genetic Algorithms JCSC 20, 1 (October 2004)

    [32] Srinivasan E. and Ebenezer D., A New Fast and Efficient Decision- Based Algorithm for Removal of High-Density Impulse Noises IEEE signal processing letters, Vol. 14, No. 3, pp.189 -192, February 2007.

    [33] Sun T. and Neuvo Y., Detail-preserving median based lters in image processing, Science Direct Pattern RecognitionLetters,Vol. 15, no. 4, pp. 341347, April 1994.

    [34] Thirilogasundari.V, Suresh babu.V, Agatha Janet.S Fuzzy Based Salt and pepper noise removal using adaptive switching median

    filterElsevier,Procedia Engineering ,Vol.38, pp. 2858-2865, 2012 [35] Umbaugh S.E, Computer Vision and Image Processing. Englewood

    Cliffs, NJ: Prentice-Hall, 1998.

    [36] Wang G., Shi G. and Liang L. Image Denoising Based on Translation Invariant Directional Lifting, IEEE Acoustics Speech and Signal Processing (ICASSP,pp. 1446-1449, March 2010.

    [37] Windyga P.S, Fast impulse noise removal, IEEE Transactions Image Process., Vol. 10, no. 1, pp. 173179, January 2001

    [38] Wu H.R. and Chen T. Adaptive Impulse Detection Using Center-Weighted Median Filters IEEE signal processing letters, Vol. 8, No. 1, January 2001.

    [39] Yang R, YinL, Gabbouj M, Astola J, and Neuvo Y, Optimal weighted median filters under structural constraints, IEEE Transactions Signal Processing, Vol. 43, pp. 591604, March 1995.

    [40] Yang X. et.al Image Denoising Based on Support Vector Machine, IEEE Engineering and Technology (S-CET), pp. 1-4, May 2012.

    [41] Zaho T., Wang Y., Ren Y. and Liao Y Approach of Image Denoising Based on Discrete Multi-Wavelet Transform, IEEEIntelligent Systems and Applications, Vol.2, No.5, pp. 1-4, May 2009.

    [42] Ze-Feng D. et al. High Probability Impulse Noise-Removing Algorithm Based on Mathematical Morphology, IEEE signal processing Letters, Vol. 14, No.1, pp.31- 34, 2007.

    [43] Zhang D. and Wang Z., Progressive switching median lter for the removal of impulse noise from highly corrupted images, IEEE Trans.Circuit Syst. II, Exp. Briefs,Vol. 46, No. 1, pp. 7880, January 1999.

    [44] Zhang S. and Karim M.A, A new impulse detector for switching median filters, IEEE Signal Processing Letters, Vol. 9, pp. 360363, November 2002.

    Amaninder Kaur Brar et al, Int.J.Computer Technology & Applications,Vol 5 (5),1772-1779

    IJCTA | Sept-Oct 2014 Available [email protected]

    1779

    ISSN:2229-6093

    PointTmp