A Survey on Digital Image Enhancement Techniques

Document Sample
A Survey on Digital Image Enhancement Techniques Powered By Docstoc
					                                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                                       Vol. 8, No. 8, November 2010

A Survey on Digital Image Enhancement Techniques
                                         V.Saradhadevi1 , Dr.V.Sundaram2
     Research scholar, Karpagam University, 2 Director of MCA , karpagam Engineering College, Coimbatore, India.

Abstract---Image enhancement is one of the major research                cause degradation of an image and image restoration is one of
fields in image processing. In many applications such as                 the key fields in today's Digital Image Processing due to its
medical application, military application, media etc., the               wide area of applications. Commonly occurring degradations
image enhancement plays an important role. There are many                include blurring, motion and noise. Blurring can be caused
techniques proposed by different authors in order to remove              when object in the image is outside the camera’s depth of field
the noise from the image and produce the clear visual of the             sometime during the exposure, whereas motion blur can be
image. Also, there are many filters and image smoothing                  caused when an object moves relative to the camera during an
methods available. All these available techniques are designed           exposure. The general model for image degradation
for particular kind of noises. Recently, neural networks turn to         phenomenon is given as y = Hf + n, where y is the observed
be a very effective tool to support the image enhancement.               blurred and noisy image, f is the original image, n is additive
Neural network is applied in image enhancement because it                random noise and H is the blurring operator. The main
provides many advantages over the other techniques. Also,                objective is to estimate the original image from the observed
neural network can be suitable for removal of all kinds of               degraded image. Whatever the degraded process, image
noises based on its training data. This paper provides survey            distortions can fall into two categories, namely, spatially
about some of the techniques applied for image enhancement.              invariant or space invariant and spatially variant or space
This survey deals with the several existing methods for image            variant. In a space invariant distortion all pixels have suffered
enhancement using neural networks.                                       the same form of distortion. This is generally caused by
                                                                         problems with the imaging system such as distortions in
Keywords--- Image Enhancement, Image Denoising, Neural                   optical system, global lack of focus, or camera motion. In a
Network, Image Filter, Image Restoration.                                space variant distortion, the degradation suffered by a pixel in
                                                                         the image depends upon its location in the image. This is
                      I. INTRODUCTION                                    because of internal factors, such as distortions in the optical
                                                                         system, or by external factors, such as object motion. This
The intention of image enhancement is to improve the                     survey provides many techniques available for image
interpretability or perception of data in images for human               enhancement.
visual or to provide better input for other automated image
processing techniques.                                                                      II. LITERATURE SURVEY

Image enhancement methods can be broadly divided into two                Uma et al., [1] proposed a Morphological Neural Network for
categories:                                                              color image restoration. This paper considers the problem of
                                                                         color image restoration degraded by a blur function and
      •   Spatial domain methods, which involves direct                  corrupted by random noise. A new approach developed by
          operation on image pixels, and                                 multilayer morphological (MLM) neural network is presented,
      •   Frequency domain methods, which involves Fourier               which uses highly nonlinear morphological neuron for image
          transform of an image for its operation.                       processing to get a high quality restored color image. In this
                                                                         paper color images are considered into RGB distribution. Then
Regrettably, there is no general theory for determining what             each subspace can be considered as a gray image space and is
good image enhancement is when it comes to human                         processed by morphological way used in gray images. This
perception. If it looks good, it is good! However, when image            method is advantageous because of its low computational
enhancement methods are used as pre-processing tools for                 overhead, improved performance in terms of signal to noise
other image processing methods, then quantitative measures               ratio with less number of neurons.
can decide which techniques are most suitable.
                                                                         Gallo et al., [2] presented an adaptive image restoration using
Image Restoration is the technique of retaining the original             local neural approach. This work aims at usage of neural
image from the degraded image given the knowledge of the                 learning for defining and experimentally evaluating an
degrading factors. There are a variety of reasons that could             iterative strategy for blind image restoration in the presence of

                                                                   173                               http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 8, No. 8, November 2010

blur and noise. A salient aspect of the solution is the local             image enhancement for gray scale images. But, the current
estimation of the restored image based on gradient descent                study has adapted these techniques to be used for color
strategies able to estimate both the blurring function and the            images. Even though the adapted image processing method is
regularized terms adaptively. As an alternative of explicitly             quite simple, the results signify that these methods may have
defining the values of local regularization parameters through            some potential to be used for improving the quality of Ziehl
predefined functions, an adaptive learning approach is                    Neelsen slide images. The experimental result illustrates that
proposed.                                                                 both methods proposed by the author can improve the image
                                                                          contrast and enhances the image quality when compared to its
The various restoration techniques used currently can be                  conventional techniques.
broadly viewed under two categories, namely, the transform
related techniques and the algebraic restoration techniques [3].          Pattern learning based image restoration using neural networks
The transform related techniques involve analyzing the                    is put forth by Dillon et al., [8]. The author illustrate a generic
degraded image after an appropriate transform has been                    pattern learning based image restoration scheme for degraded
applied. The two popular transform related techniques are                 digital images, where a feed-forward neural network is
inverse filtering and Kalman filtering [4]. Inverse filtering             employed for implementation of the proposed techniques. The
produces a perfect restoration in the absence of noise, but the           methodology reported here can be applied in several
presence of noise has very bad effects. The Kalman filter                 circumstances, for instance, quality enhancement as a post-
approach can be applied to non stationary image but it is                 processing of image compression schemes, blur image
computationally very intensive.                                           restoration and noise image filter, provided that the training
                                                                          data set is comprised of patterns rich enough for supervised
Algebraic techniques attempt to find a direct solution to the             learning. This paper focuses on the problem of coded image
distortion by matrix inversion techniques, or techniques                  restoration. The key points addressed in this work are
involving an iterative method to minimize a degradation                        • The use of edge data extracted from source image as
measure. The two popular algebraic techniques available are                         a priori knowledge in the regularization function to
pseudo inverse filtering and constrained image restoration.                         get better details and reduce the ringing artifact of the
The pseudo inverse spatial image restoration techniques                             coded images.
attempt to restore an image by considering the vector space                    • The theoretic basis of the pattern learning-based
model of the image degradation and attempting to restore the                        technique using implicit function theorem.
image in this vector space domain. This method does not                        • Subjective quality improvement with the use of an
consider the effects of noise in the calculations of the pseudo                     image similarity for training neural networks
inverse and so is sensitive to noise in the image. This involves               • Empirical studies with contrast to the set partitioning
determining an approximation to the inverse of the matrix                           in hierarchical tree (SPIHT) method.
blurring operator which is multiplied with the column scanned             The main advantages of this model-based neural image
image vector to produce the degraded image. Blur matrices are             restoration approach comprise strong robustness with respect
very large and it is not computationally feasible to invert them.         to transmission noise and the parallel processing for real-time
Constrained restoration techniques are often based on Wiener              applications.
estimation and regression techniques. One of the major
drawbacks in most of the image restoration algorithms is the              Reeves [9] described fast and direct image restoration with
computational complexity, so various simplifying assumptions              edge-preserving regularization. In several applications, fast
have been made to obtain computationally feasible algorithms.             restorations are required to keep up with the frame rate. FFT-
                                                                          based restoration affords a fast implementation, but it does so
Motivated by the biological neural network in which the                   at the expense of assuming that the degree of regularization is
processing power lies in a large number of neurons linked                 constant over the image. Unfortunately, this hypothesis can
with synaptic weights, artificial neural network models                   generate significant ringing artifacts in the presence of edges
attempt to achieve a good performance via dense                           as well as edges that are blurrier than necessary. Shift-variant
interconnection of simple computational elements. Neural                  regularization affords a way to vary the roughness penalty as a
network models have great potential in areas where high                   function of spatial coordinates. Virtually all edge-preserving
computation rates are required and the current best systems are           regularization techniques exploit this concept. However, this
far from equaling human performance. Restoration of a high                technique destroys the structure that makes the use of the FFT
quality image from a degraded recording is a good application             possible, since the deblurring operation is no longer shift-
area of neural nets. Joon et al., [5] proposed a Modified                 invariant. Thus, the restoration techniques available for this
Hopfield neural network model for solving the restoration                 problem no longer have the computational efficiency of the
problem which improves upon the algorithm proposed by                     FFT. The author proposes a new restoration method for the
Zhou et al. [6].                                                          shift-variant regularization approach that can be implemented
                                                                          in a fast and flexible manner. This paper decomposes the
Osman et al., [7] gives an image enhancement using bright                 restoration into a sum of two independent restorations. One
and dark stretching techniques for tissue based tuberculosis              restoration yields an image that comes directly from an FFT-
bacilli detection. This paper proposes two methods for color              based approach. This image is a shift-invariant restoration
image enhancement; bright stretching and dark stretching                  consisting of usual artifacts. The other restoration involves a
algorithms. Both techniques are well known to create good

                                                                    174                                http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                                       Vol. 8, No. 8, November 2010

set of unknowns whose number equals the number of pixels                 proposed in this paper. The approached operator utilizes the
with a local smoothing penalty significantly different from the          gradient operator to create the image enhancement processing
typical value in the image. This restoration represents the              focus on the interested regions, and the OTSU operator to
artifact correction image. By summing the two, the artifacts             automatically select the best threshold value, which can realize
are canceled. Since the second restoration has a significantly           a novel adaptive fuzzy image enhancement algorithm for local
reduced set of unknowns, it can be calculated very efficiently           regions. Through the experimentations of the asphalt
even though no circular convolution structure exists.                    pavement crack image detection system, the experimental
                                                                         results specify that the novel algorithm can not only attain
Noise-refined image enhancement using multi-objective                    better processing effects and higher processing speed than
optimization is illustrated by Peng et al., [10]. This paper             now-available fuzzy image enhancement algorithms, but also
presents a novel scheme for the enhancement of images using              possess the property of high practicability and generality.
stochastic resonance (SR) noise. In this scheme, a suitable
dose of noise is added to the lower quality images such that             Faouzi et al., [13] provides a directional-rational approach for
the performance of a suboptimal image enhancer is improved               color image enhancement. In this, the author presents an
without altering its parameters. In this paper, image                    unsharp masking-based approach for noise smoothing and
enhancement is modeled as a constrained multi-objective                  edge enhancing in multichannel images. The structure
optimization (MOO) problem, with similarity and some                     presented by author is similar to the conventional unsharp
desired image enhancement characteristic being the two                   masking structure, however, the enhancement is allowed only
objective functions. The principle of SR noise-refined image             in the direction of maximal change and the enhancement
enhancement is analyzed, and an image enhancement system                 parameter is computed as a nonlinear function of the rate of
is developed. A genetic algorithm-based MOO technique is                 change. This scheme improves the true details, limits the
employed to find the optimum parameters of the SR noise                  overshoot near sharp edges and attenuates noise in flat areas.
distribution. In addition, a novel image quality evaluation              In addition the use of the control function eliminates the need
metric based on human visual system (HVS) is developed as                for the subjective coefficient λ used in the conventional
one of the objective functions to guide the MOO search                   unsharp masking method.
                                                                         The noise reduction based on fuzzy image filtering is put forth
Lu et al., [11] proposed an image noise reduction technique              by Dimitri et al., [14]. A new fuzzy filter is provided for the
based on the fuzzy rules. Considering the image as non-                  noise reduction of images corrupted with additive noise. The
stationary signal, an image noise reduction method based on              filter involves two stages. The initial stage calculates a fuzzy
the fuzzy rules is proposed. This image processing system                derivative for eight different directions. The next stage uses
(IPS) is recognized as a time-variant system in which the                these fuzzy derivatives to carry out fuzzy smoothing by
system parameters change continuously based on the local                 weighting the contributions of neighboring pixel values. Both
characteristics of the images. For the purpose of noise                  these stages are dependent on fuzzy rules which make use of
reduction, Gaussian noise is considered here. The fuzzy rules            membership functions. The filter can be implemented
are implemented to consider the unstableness and uncertainty             iteratively to effectively decrease heavy noise. Especially, the
of signals. The nonlinear function indicating the fuzzy rule-            shape of the membership functions is adapted according to the
based IPS depends on the rules concerning the local                      remaining noise level after each iteration; making use of the
characteristics of the input, on the membership functions, and           distribution of the homogeneity in the image. A statistical
on the used defuzzification method. For making the system                technique for the noise distribution can be included to relate
performance as high as possible, these factors must be agreed            the homogeneity to the adaptation scheme of the membership
to be the most appropriate ones. In this paper a technique for           functions.
designing the optimum nonlinear function directly from the
local characteristics of training data is presented. Here the             Gacsadi et al., [15] makes use of cellular neural network for
rules, the membership functions, and the technique of                    the purpose of image enhancement. This technique takes both
defuzzification are not essential to be known. The design of             the denoising and the increase of the contrast into
these factors is concerned in the design of the membership               consideration. Due to whole parallel processing, computing-
function, thus attaining the optimum nonlinear function is               time reduction is achieved. In the enhancement process by
sufficient for designing the IPS. The only thing required to do          usage of nonlinear and feedback template local and also
is to choose what sort of the local characteristics of the image         regional properties will be taken into consideration due to the
should be applied to the rule-based system. Computer                     propagation of the effect between the neighbors. Considerable
simulations illustrate that the proposed technique gives better          computing power is required to solve the image processing
results in comparison with that of the weighted averaging filter         task described by variational computing. The Cellular Neural
and median filter.                                                       Networks (CNN) proved to be very useful regarding real-time
                                                                         image processing. The reduction of computing time, due to
An adaptive fuzzy image enhancement algorithm for local                  parallel processing, can be obtained only if the processing
regions is given by Yan et al., [12]. To overcome the                    algorithm can be implemented on a CNNUC or by using
drawbacks of low speed and losing image information in fuzzy             emulated digital CNN-UM implemented on FPGAs.
image enhancement algorithms, a novel fuzzy enhancement
operator with close-character and transplantable-character is

                                                                   175                              http://sites.google.com/site/ijcsis/
                                                                                                    ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 8, No. 8, November 2010

The traditional analog architectures of CNN-UC are superior               appropriate for enhancement because the color components
in terms of processing speed and power dissipation. However,              are not decoupled. Alternatively, in HSV color model, hue
these implementations have a restricted applicability due to              (H), the color content, is separate from saturation (S), which
their limited features regarding accuracy, flexibility, small             can be used to dilute the color content and V, the intensity of
number of cells, their high cost and the lengthy period needed            the color content. By conserving H, and modifying only S and
for the development of such a chip. On the other hand, while              V, it is likely to enhance color image. Therefore, it is required
software solutions are extremely flexible, they are sometimes             to convert RGB into HSV for the purpose. A Gaussian type
inefficient because of limited and low processing speed.                  membership function is utilized to model S and V property of
Today, the version of CNN digital emulator implemented                    the image. This membership function utilizes only one
FPGA is a solution that achieved a compromise between speed               fuzzifier and is evaluated by maximizing fuzzy contrast.
and accuracy, but ensure repeatability, reproducibility,
flexibility, possibility for CNN implementation even for                  Muthu Selvi et al., [18] put forth a hybrid image enhancement
complex processes of processing and easy interfacing with                 technique for noisy dim images using curvelet and
digital systems. In this sense, the CNN digital emulator                  morphology techniques. The noisy dim images degrade the
implemented FPGA maximizes, for a concrete application, the               image quality. The denoising method using curvelet transform
performances of the overall CNN processing.                               outperforms than wavelet transform. The noisy dim image is
                                                                          made noise free with the help of curvelet transform to the dim
There are two basic approaches to image denoising [16] -                  image for avoiding the over illumination and under
spatial filtering methods and transform domain filtering                  illumination problems. Next the dim image is enhanced using
methods.                                                                  the morphological transformations. Closing by reconstruction
                                                                          is implemented to identify the background of the dim image.
Spatial Filtering                                                         The experimental result shows that the morphological
                                                                          restoration filter with closing by reconstruction produces
A conventional way to remove noise from image data is to                  better result than opening by reconstruction.
employ spatial filters. Spatial filters can be further classified
into non-linear and linear filters.                                       Hassan et al., [19] presented a contrast enhancement technique
                                                                          for dark blurred images. The chief goal presented by the
    i.    Non-Linear Filters                                              author is to produce a contrast enhancement technique to
With non-linear filters, the noise is removed without any                 recover an image within a given area, from a blurred and
attempts to explicitly identify it. Spatial filters utilize a low         darkness specimen, also improve visual quality of it. This
pass filtering on groups of pixels with the assumption that the           method consists of two steps unsharp masking step and
noise occupies the higher region of frequency spectrum.                   contrast enhancement step. The unsharp masking step is
Generally spatial filters remove noise to a reasonable extent             applied to the image to sharpen edges and bring out hidden
but at the cost of blurring images which in turn makes the                details. On the contrary enhancement step 3x3 slider map
edges in pictures invisible. In recent years, a variety of                window was applied to the image to determine if the
nonlinear median type filters such as weighted median, rank               corresponding pixel will be remapped or not. The new value
conditioned rank selection, and relaxed median have been                  of remapped pixel obtained is based on a sigmoid map
developed to overcome this drawback.                                      function. Good and satisfying results were obtained by
                                                                          experimentation on this technique.
   ii.   Linear Filters
A mean filter is the optimal linear filter for Gaussian noise in          Gilbao et al., [20] uses the complex diffusion processes for
the sense of mean square error. Linear filters also tend to blur          image enhancement and denoising. The linear and nonlinear
sharp edges, destroy lines and other fine image details, and              scale spaces, obtained by the inherently real-valued diffusion
perform poorly in the presence of signal-dependent noise. The             equation, are generalized to complex diffusion processes, by
wiener filtering method requires the information about the                incorporating the free Schrodinger equation. A basic solution
spectra of the noise and the original signal and it works well            for the linear case of the complex diffusion equation is
only if the underlying signal is smooth. Wiener method                    developed. Investigation of its performance shows that the
implements spatial smoothing and its model complexity                     generalized diffusion process combines properties of both
control correspond to choosing the window size.                           forward and inverse diffusion. It verifies that the imaginary
                                                                          part is a smoothed second derivative, scaled by time, when the
Sarode et al., [17] proposed the color image enhancement with             complex diffusion coefficient approaches the real axis. Based
the help of fuzzy system. This technique involves the use of              on this observation, the authors develop two examples of
knowledge-base (fuzzy expert) systems that are capable of                 nonlinear complex processes, useful in image processing: a
mimicking the behavior of a human expert. Fuzzy technique                 regularized shock filter for image enhancement and a ramp
of knowing severity of tumor is essential to determine if there           preserving denoising process.
is a need for the biopsy and it gives to user a clear idea of
spread and severity level of tumor. Fuzzy based improvement               Image denoising using non-negative sparse coding shrinkage
of color feature of tumor is an application of fuzzy in the area          algorithm is given by Shang et al., [21]. The author proposes a
of color feature extraction for enhancement of a peculiar                 new method for denoising natural images using this extended
feature. It has been determined that RGB color model is not               non-negative sparse coding (NNSC) neural network shrinkage

                                                                    176                               http://sites.google.com/site/ijcsis/
                                                                                                      ISSN 1947-5500
                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                        Vol. 8, No. 8, November 2010

algorithm, which is self-adaptive to the statistic property of            proposed method offering the best efficiency in terms of
natural images. The fundamental principle of denoising using              image denoising and edge preservation.
NNSC shrinkage is similar to that using standard sparse
shrinkage and wavelet soft threshold. Using test images                   Zhang et al., [25] presented image denoising using a neural
corrupted by additive Gaussian noise, this paper evaluated the            network based non-linear filter in wavelet domain. Images are
method across a range of noise levels. This method utilized the           often distorted as a result of various factors that can occur
normalized mean squared error as a measure of the quality of              during acquisition and transmission processes. Image
denoising images and the signal to noise rate (SNR) value as              denoising is intended at removing or reducing noise, so that a
an evaluative feature of different denoising approaches. The              good-quality image can be obtained for various applications.
experimental result shows that the NNSC shrinkage certainly               The author presents a neural network based denoising method
is effective in image denoising. Otherwise, the author also               implemented in the wavelet transform domain. A noisy image
compares the effectiveness of the NNSC shrinkage with sparse              is first wavelet transformed into four subbands, and then a
coding shrinkage and wavelet soft threshold method. The                   trained layered neural network is applied to each subband to
simulative tests show that this denoising method outperforms              generate noise-removed wavelet coefficients from their noisy
any other of the two kinds of denoising approaches.                       ones. The denoised image is then obtained through the inverse
                                                                          transform on the noise-removed wavelet coefficients.
Gupta et al., [22] designed a FIR filter for image restoration            Compared with other techniques performed in the wavelet
using principal component neural network. The neural                      domain, it requires no a priori knowledge about the noise and
network can be applied in many image denoising applications               needs only one level of signal decomposition to obtain very
because of its inherent characteristics such as nonlinear                 good denoising results.
mapping and self-adaptiveness. The design of filters widely
depends on the a-priori knowledge about the type of noise.                                           III. CONCLUSION
Because of this, standard filters are application and image
specific. Extensively used filtering algorithms reduce noisy              This survey discusses about several existing image
artifacts by smoothing. Though, this operation normally                   enhancement methods. All those methods discussed have their
results in smoothing of the edges as well. Alternatively,                 own advantages and disadvantages. This survey helps in
sharpening filters enhance the high frequency details making              choosing the better suitable image enhancement scheme for
the image non-smooth. An integrated general technique to                  particular kind of noise in the image. Also, the filtering
design a finite impulse response filter based on principal                algorithms used for removing the noise from the image are
component neural network (PCNN) is provided in this study                 presented in this thesis. This survey explains about the
for image filtering, optimized in the sense of visual inspection          importance of neural network for image enhancement as the
and error metric. This technique utilizes the inter-pixel                 neural networks have the advantages such as nonlinear
correlation by iteratively updating the filter coefficients using         mapping and self-adaptiveness. To overcome the demerit of
PCNN. This technique performs optimal smoothing of the                    the techniques discussed in this survey, Adaptive Neuro-Fuzzy
noisy image by preserving high and low frequency features.                Interference Systems (ANFIS) can be used for image
Experimental results state that this filter is robust under               enhancement as it combines the advantages of Artificial
various noise distributions. Additionally, the number of                  Neural Networks (ANN) and Fuzzy Interference System
unknown parameters is very few and most of these parameters               (FIS).
are adaptively obtained from the processed image.

Image denoising based on combined neural networks filter is                                             REFERENCES
proposed Junhong et al.,[23]. A new image restoration
technique based on combined neural networks Alter is                      [1]   S. Uma and S. Annadurai, "Color Image Restoration Using
proposed by the author. This integrated neural networks Alter                   Morphological Neural Network", ICGST.
is posed by a BPNN Alter and an image data fusion system                  [2]   I. Gallo, E. Binaghi and A. Macchi, "Adaptive Image Restoration using
based on self-organizing mapping neural networks. And this                      a Local Neural Approach".
                                                                          [3]   H. C. Andrews & B. R. Hunt, "Digital Image restoration", Englewood
technique can use the corrupted image itself as training data to                cliffs, NJ, Prentice Hall, 1977.
avoid the problem of how to choose the training data, which is            [4]   Rafael C. Gonzalez and Richard E. Woods, "Digital image processing",
most of the other neural networks denoising methods have to                     2nd edition, Addison- Wesely, 2004.
face, by using the distributed character of WGN. Experiment               [5]   Joon K. Paik and Aggelos K. Katsaggelos, "Image restoration using a
                                                                                modified Hopfield Network", IEEE Transactions on image processing,
results show that this method can denoise the noises                            Vol 1, No.1, pp. 49-63, January 1992.
effectively.                                                              [6]   Y.T.Zhou, R. Chellappa and B.K. Jenkins, "Image restoration using a
                                                                                neural network", IEEE Trans. Acoust., Speech, Signal Processing, Vol,
Gacsadi et al., [24] describes PDE-based medical images                         ASSP-36, pp 1141-1151, July 1988.
                                                                          [7]   M.K. Osman, M.Y. Mashor and H. Jaafar2Colour, "Image Enhancement
denoising using Cellular Neural Networks. The author                            using Bright and Dark Stretching Techniques for Tissue based
presents the medical image denoising by using cellular neural                   Tuberculosis Bacilli Detection", Proceedings of the International
networks (CNN), based on the variational model of Chan and                      Conference on Man-Machine Systems (ICoMMS), 2009.
Esedoglu. By comparatively examining the proposed method                  [8]   Dianhui Wang Dillon and T. Chang, "Pattern learning based image
                                                                                restoration using neural networks", Proceedings of the International
and other CNN methods that uses variational computation, the                    Joint Conference on Neural Network, 2002.

                                                                    177                                    http://sites.google.com/site/ijcsis/
                                                                                                           ISSN 1947-5500
                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                   Vol. 8, No. 8, November 2010

[9]    S.J. Reeves, "Fast and direct image restoration with edge-preserving
       regularization", IEEE Digital Signal Processing Workshop, 2002.
[10]   Renbin Peng, Hao Chen and Varshney, "Noise-refined image
       enhancement using multi-objective optimization", 44th Annual
       Conference on Information Sciences and Systems (CISS), 2010.
[11]   Ruihua Lu and Li Deng, "An Image Noise Reduction Technique Based
       on the Fuzzy Rules".
[12]   Yan Maode, Bo Shaobo, Li Xue and He Yuyao, "An Adaptive Fuzzy
       Image Enhancement Algorithm for Local Regions", Chinese Control
       Conference, 2007.
[13]   Faouzi Alaya Cheikh and Moncef Gabbouj, "Directional-Rational
       Approach for Color Image Enhancement".
[14]   Dimitri Van De Ville, Mike Nachtegael, Dietrich Van der Weken,
       Etienne E. Kerre, Wilfried Philips, and Ignace Lemahieu, "Noise
       Reduction by Fuzzy Image Filtering", IEEE transactions on fuzzy
       systems, vol. 11, no. 4, august 2003.
[15]   A. Gacsadi, V. Tiponut, E. Gergely, I. Gavrilut, "Variational Based
       Image Enhancement Method by using Cellular Neural Networks",
       Proceedings of the 13th WSEAS International Conference on
[16]   Mukesh C. Motwani, Mukesh C. Gadiya, Rakhi C. Motwani and
       Frederick C. Harris, Jr., "Survey of Image Denoising Techniques".
[17]   Milindkumar V. Sarode, S.A.Ladhake and Prashant R. Deshmukh,
       "Fuzzy system for color image enhancement", World Academy of
       Science, Engineering and Technology, 2008.
[18]   Muthu Selvi, Roselin and Kavitha, "A Hybrid Image Enhancement
       Technique for Noisy Dim Images Using Curvelet and Morphology",
       International Journal of Engineering Science and Technology Vol. 2(7),
[19]   Naglaa Yehya Hassan and Norio Aakamatsu, "Contrast Enhancement
       Technique of Dark Blurred Image", IJCSNS International Journal of
       Computer Science and Network Security, Vol.6, No.2A, February 2006.
[20]   Gilboa.G, Sochen.N and Zeevi.Y.Y, "Image enhancement and denoising
       by complex diffusion processes", IEEE Transactions on Pattern Analysis
       and Machine Intelligence, 2004.
[21]   Li Shang and Deshuang Huang, "Image denoising using non-negative
       sparse coding shrinkage algorithm",          IEEE Computer Society
       Conference on Computer Vision and Pattern Recognition, 2005.
[22]   Gupta.P.K and Kanhirodan.R, "Design of a FIR Filter for Image
       Restoration using Principal Component Neural Networks", IEEE
       International Conference on Industrial Technology, 2006.
[23]   Junhong Chen and Qinyu Zhang, "Image Denoising Based on Combined
       Neural Networks Filter", International Conference on Information
       Engineering and Computer Science, 2009.
[24]   Gacsadi.A, Grava.C, Straciuc.O and Gavrilut.I, "PDE-based medical
       images denoising using Cellular Neural Networks", International
       Symposium on Signals, Circuits and Systems, 2009.
[25]   Zhang.S and Salari.E, "Image denoising using a neural network based
       non-linear filter in wavelet domain", IEEE International Conference on
       Acoustics, Speech and Signal Processing, 2005.

                                                                                178                           http://sites.google.com/site/ijcsis/
                                                                                                              ISSN 1947-5500