Volume 4, Issue 3, March 2014

ISSN: 2277 128X

International Journal of Advanced Research in

Computer Science and Software Engineering
Research Paper
Available online at: www.ijarcsse.com
Special Issue: Computing Terminologies and Research Development
Conference Held at SCAD College of Engineering and Technology, India

Automated Defect Detection of Steel Surface Using Neural

Network Classifier with Co-occurrence Features
M.Selvi 1#, D.Jenefa 2*
1#, Assistant Professor, Department of Computer Science and Engineering,
2* , PG Scholar, Department of Computer Science and Engineering,
Sri Lakshmi Ammaal Engineering College, Tamil Nadu, India
AbstractAutomatic defect detection of steel surface is challenging due to its localization of defect , variation in shape,
and their rare occurrences in the surface. The manual inspection might give inconsistent results especially when there
are plentiful defects need to be evaluated due to human factor. But, An automatic defect identification system used to
detect the defect correctly. It is based on defect classification. Image classification will be used for automated visual
inspection. It will be performed through textures analysis and Probabilistic Neural Network. The textures are
extracted using wavelet filters with co-occurrence features. This is exactly extract the structure by removing the back
ground noise and the false defects. The Probabilistic Neural Network is used to differentiate the defect image from
defect free one. Finally, the weld defects of the steel surface are extracted with better accuracy.
Keywords - Texture Analysis, Wavelet Filters, Co-occurrence Features, Probabilistic Neural Network
I. INTRODUCTION
Many automated manpower saving systems have been developed and employed to reduce the production cost and
improve the quality of steel products. However, in the finishing product, quality assurance with respect to surface
defects is partly automated and mostly performed by manual inspection. In order to meet the growing demand for highquality product in short duration, the use of intelligent visual inspection systems is increasingly essential in production
lines. The use of automated inspection technique is necessary at each level of production to improve the quality of
product as well as to eliminate the need of a human intervention in a hazardous environment. This will facilitate to make
the surface information available immediately after the rolling process. This has an obvious benefit in terms of quality
assurance. The surface defects often arise as a result of systematic process problems, such as partially damaged
machinery or metallurgical drift. Thus, the early detection of defects can also have a direct cost benefit in terms of saving
of time as well as to prevent rejection from being generated in large quantities, in downstream.
There are certain major difficulties in defect detection of hot rolled steel surface. These are as follows: The
classification of local area of surface defects on hot-rolled steel surface is a challenging task due to the variability in
manifestations of the defects grouped under the same defect label. Surface defects of steel plates can be broadly
categorized into two main groups: textural defects and geometric defects. Textural defects, such as heavy scale, rolled in
scale, salt and pepper, slab edge, and so on, and it can be detected by the methods based on textural analysis. Geometric
defects, such as cracks, scratches, etc., can be detected by algorithms based on wavelet filtering. In addition, these defects
on steel surface can be effectively identified from the uniform images with the small-sample statistics like the mean and
variance of gray levels. They can also be detected using simple thresholding or edge-detection techniques.
However, the large variety of defects requires different threshold values for different types of defects. However, a good
defect detection system should detect all types of defects equally by the same algorithm. There is a great deal of scales
with different colours and appearances on the surface of steel plates, which makes the background detection task
difficult. In spite of careful arrangement of uniform illumination over the steel surface, the problem of vibration-induced
uneven illumination is difficult to overcome. Therefore, a suitable defect detection algorithm of hot-rolled steel surfaces
needs to be insensitive to scales and uneven illumination to the maximum possible extent. The defect free steel surface
images are as follows:

Fig 1.1: Examples of Normal steel products

2014, ICCTRD All Rights Reserved

Page | 60

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
Two Major types of defects are encountered with respect to the space localization: the local defect and the distributed
defect. Local defects are limited in space but may appear in a discontinuous fashion on different places and different
shapes on the steel surface. For example, scratches, crack, rupture, blister, bruise, head mark, etc., belong to this
category. On the other hand, distributed defects are spread over the large area of the surface that is it appear in
continuous fashion. For example, rolled in scale, salt and pepper, scale grain, and slab edge are categorized in this group.
The local and distributed defect of the steel surface images are as follow:

Fig 1.2: Examples of local and distributed defect steel products

Local defect, being constrained in space, is principally the outcome of imperfect rolling. On the other hand, the
distributed defect is mainly attributed to shortcomings in metallurgical mixing in upstream. Alternatively, the defects
may be classified into two types based on appearance: the geometric defects such as scratches, cracks, head mark,
rupture, spall, waviness, whip, etc., and the textural defects such as rolled in scale, salt and pepper, scales in grain, slab
edge, etc.
The remaining part of this paper is organized in the following way. Section II focuses on the Background and past work
of the automated defect detection system. In Section III, a brief theory of the overall System Architecture. In Section IV,
the automated defect detection scheme (feature extraction and training of classifiers) is presented. Section V, finally,
provides the conclusion and future direction of our works.
II. BACKGROUND AND PAST WORKS
There is a few amount of work related to the defect detection on the steel surface for Industrial applications. The previous
Reaches have worked out the defect detection of steel casting [1] is to find out the internal defects of the castings, to find
out the surface defect of the fabric materials [2]-[4], identify the internal defects of the glass plates [5], the image-based
defect identification system for fluted ingots using the Bayesian classifier [7] and so on. The Quality is the main goal of
the Industrial application. The automated defect detection system is used to find out the defect correctly compare to the
manpower system. It also consumes less amount of energy.
Among the recent works, Choi , Yun, Seo, Park, and Kim [9] suggested an edge-preserving filter and thresholding
technique to detect the cracks, spots, and the dark line for high speed steel bar in coil, and the noise is removed by the
morphological operation. But, It is suitable for few defects like separating defect region and defect-free region with
corner crack. Yong , Ke, Wu [10] projected an algorithm based on Undecimated Wavelet Transform and Mathematical
Morphology for detection of geometric defects such as scratches and cracks on hot rolled steel plates. Undecimated
wavelet transform(UWT) doesnt damage the original signals. Thus it produces more precise information for frequency
localization. It May lead to false alarms like water droplets and dirt in the defect classification stage.
III. SYSTEM ARCHITECTURE
Our proposed method is to classify the defect product from the defect free one. It is mainly used to avoid the false defect
of the steel surface by removing the noise of the steel surface. The input of the system is the digital radiographic image
and its noise will be removed by wavelet filters. Images Texture is extracted by using co-occurrence feature extraction
method. This method is used to extract the features exactly. The features of the training samples are extracted also by
using the co-occurrence feature extraction. Finally, the Probabilistic Neural Network (PNN) is used to detect the defect
based Radial Basis Function (RBF). It exactly classify the defect of the steel surface. Based on the generalized
regression neural network method will be used to classify the input image into either normal or defect parts.

Fig 3.1: Framework for defect Classification and Segmentation

In numerical analysis and functional analysis, a discrete wavelet transform (DWT) is an wavelet transform for which
the wavelets are discretely sampled. As with other wavelet transforms, a key advantage it has over Fourier transforms is
2014, ICCTRD All Rights Reserved

Page | 61

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
temporal resolution: it captures both frequency and location information that is location in time. The mother functions are
rescaled, or "dilated" by powers of two, and translated by integers. What makes wavelet bases especially interesting is the
self-similarity caused by the scales and dilations. There is no need to divide the input coding into non-overlapping TwoDimensional blocks, it has higher compression ratios avoid blocking artifacts. It allows good localization both in time
and spatial frequency domain. It is higher flexibility that is Wavelet function can be freely chosen. Since wavelet
transform has the ability to decompose complex information and patterns into elementary forms, it is commonly used in
acoustics processing and pattern recognition. Moreover, wavelet transforms can be applied to the following scientific
research areas: edge and corner detection, partial differential equation solving, transient detection, filter
design, Electrocardiogram analysis, texture analysis and business information analysis.
Texture contains important information about the structural arrangement of the surface, such as: clouds, leaves, bricks,
fabric, etc. It also describes the relationship of the surface to the surrounding environment. In short, it is a feature that
describes the distinctive physical composition of a surface. The texture properties include: Coarseness, Contrast,
Directionality, Line-likeness, Regularity and Roughness. It is one of the most important defining features of an image. It
is characterized by the spatial distribution of gray levels in a neighbourhood. In order to capture the spatial dependence of
gray-level values, which contribute to the perception of texture, a two-dimensional dependence texture analysis matrix is
taken into consideration. This Two Dimensional matrix is obtained by decoding the image file; jpeg, bmp, etc.
There are three principal approaches used to describe texture: statistical, structural and spectral. The Statistical
techniques characterize textures using the statistical properties of the grey levels of the points or pixels comprising a
surface image. Typically, these properties are computed using: the grey level co-occurrence matrix of the surface, or the
wavelet transformation of the surface. The Structural techniques characterize textures as being composed of simple
primitive structures called texels (or texture elements). These are arranged regularly on a surface according to some
surface arrangement rules. The Spectral techniques are based on properties of the Fourier spectrum and describe global
periodicity of the grey levels of a surface by identifying high-energy peaks in the Fourier spectrum,
For optimum classification purposes, what concern us are the statistical techniques of characterization. This is because
it is these techniques that result in computing texture properties. Neural network is the best method in recognition and
discrimination between different sets of signals. To get best results using the neural network, it is necessary to choose a
suitable architecture and learning algorithm. Unfortunately there is no guaranteed method to do that. The best way to do
that is to choose what is expected to be suitable according to our previous experience and then to expand or shrink the
neural network size until a reasonable output is obtained.
IV. PROPOSED METHOD
In this paper, an automatic defect identification system for detecting defects on the steel products surface from captured
digital radiographic images based on defect classification. Image classification will be used for automated visual
inspection to classify defect protects from quality one. It will be performed through textures analysis and probabilistic
neural network. The textures are extracted using wavelet filters with co-occurrence features. Here, the simplest and
fastest classifier called new probabilistic neural network with radial basis kernel function used for classification with
better accuracy.
A. Discrete Wavelet Transform
Calculating wavelet coefficients at every possible scale is a fair amount of work, and it generates an awful lot of data. It
turns out, rather remarkably, that if we choose scales and positions based on powers of two so called dyadic scales and
positions then our analysis will be much more efficient and just as accurate. We obtain such an analysis from the discrete
wavelet transform (DWT). This algorithm yields a fast wavelet transform
For many signals, the low-frequency content is the most important part. It gives the signal and its identity. The highfrequency content contains edge information. The original signal, S, passes through two complementary filters and
emerges as two signals. We produce two sequences called cA and cD. The process on the right, which includes down
sampling, produces DWT coefficients.
y[n]=(x*g)[n]= = [ ]
(4.1)
The discrete wavelet transform is obtained by applying complementary low-pass and high-pass filters and subsequent
decimation (H and L). Both H and L are applied to data vector x1, x2, ..., x8. The output of H is the four wavelet
coefficients for the first resolution; the output of L is the four coefficients of the scaling function. The wavelet
coefficients of the other resolution levels are obtained by iterating the low- and high-pass filtering steps on the
coefficients of the scaling function. Consider an defective steel surface image as:

Fig 4.1 Defective Steel Surface

2014, ICCTRD All Rights Reserved

Page | 62

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
The Detail Coefficients cD are small and consist mainly of a high-frequency as,
yhigh [n] = = [2 ]
(4.2)
The Edge information of the High Frequency defect image as follows:

Fig 4.2 High Frequency defect image

While the Approximation Coefficients cA contain much less noise than does the original signal as,
ylow [n] = = [2 ]
(4.3)
The detailed information of the Low Frequency defect image as follows:

Fig 4.3 Low Frequency defect image

The actual lengths of the detail and approximation coefficient vectors are slightly more than half the length of the
original signal. This has to do with the filtering process, which is implemented by convolving the signal with a filter. The
convolution smears the signal, introducing several extra samples into the result.
Since the analysis process is iterative, in theory it can be continued indefinitely. In reality, the decomposition can
proceed only until the individual details consist of a single sample or pixel and select a suitable number of levels based
on the nature of the signal, or on a suitable criterion such as entropy
B. Texture Analysis
Texture is that innate property of all surfaces that describes visual patterns, each having properties of homogeneity. It
contains important information. It is one of the most important defining features of an image. It is characterized by the
spatial distribution of gray levels in a neighbourhood.
Co-occurrence Matrix:
The co-occurrence matrix is obtained by the position operator P(i,j), An n x n matrix A, the number of times that points
with grey level (intensity) g[i] occur as A[i][j], in the position specified by P, relative to points with grey level g[j], the n
x n matrix that is produced by dividing A with the total number of point pairs that satisfy P as C. Finally, C[i][j] is a
measure of the joint probability that a pair of points satisfying P will have values g[i], g[j]. Here, C is called a cooccurrence matrix defined by P.
This can also be illustrated as follows: Let t be a translation, then a co-occurrence matrix Ct of a region is defined for
every grey-level (a, b)

Ct (a, b) card {(s, s t ) R2 | A[s] a, A[s t ] b}

(4.4)
Here, Ct(a, b) is the number of site-couples, denoted by (s, s + t) that are separated by a translation vector t, with a
being the grey-level of s, and b being the grey-level of s + t.
At first the co-occurrence matrix is constructed, based on the orientation and distance between image pixels. Then
meaningful statistics are extracted from the matrix as the texture representation . Haralick proposed the following texture
features :
Energy
Contrast
Correlation
Homogeneity
Entropy
2014, ICCTRD All Rights Reserved

Page | 63

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
Hence, for each Haralick texture feature, we obtain a co-occurrence matrix. These co-occurrence matrices represent the
spatial distribution and the dependence of the grey levels within a local area. Each (i,j) th entry in the matrices, represents
the probability of going from one pixel with a grey level of 'i' to another with a grey level of 'j' under a predefined
distance and angle. From these matrices, sets of statistical measures are computed, called feature vectors .
Energy: It is a gray-scale image texture measure of homogeneity changing, reflecting the distribution of image grayscale uniformity of weight and texture.
E = x y p(x,y)2
(4.5)
Contrast: Contrast is the main diagonal near the moment of inertia, which measure the value of the matrix is distributed
and images of local changes in number, reflecting the image clarity and texture of shadow depth.
I=

(x y)2 p(x,y)

(4.6)

Entropy: It measures image texture randomness, when the space co-occurrence matrix for all values are equal, it
achieved the minimum value.
S = - x y p(x,y) log p(x,y)
(4.7)
Correlation Coefficient: Measures the joint probability occurrence of the specified pixel pairs.
Correlation:sum(sum((x-x)(y-y)p(x,y)/xy))

(4.8)

Homogeneity: Measures the closeness of the distribution of elements

Homogenity = sum(sum(p(x , y)/(1 + [x-y])))

(4.9)

C. Probabilistic Neural Network

Neural network is the best method for recognition and inequity between different sets of signals. To get best results using
the neural network, it is necessary to choose a suitable architecture . But there is no guaranteed method to do that. The
best way to do that is to choose what is expected to be suitable to our project. The proposed neural network has been used
to detect and locate the defect in two cases. The first case was to detect and locate a defect in a two-dimensional sector of
the breast model. The location of defect was considered randomly at the centre and in any of the four quadratures. The
second case was to detect and locate defect anywhere in the three-dimensional model.
Probabilistic Neural Network (PNN) and General Regression Neural Networks (GRNN) have similar architectures, but
there is a fundamental difference: Probabilistic networks perform classification where the target variable is categorical,
whereas general regression neural networks perform regression where the target variable is continuous.

Fig 4.4: Architecture of PNN

PNN networks have four layers:
Input layer There is one neuron in the input layer for each predictor variable. In the case of categorical variables, N-1
neurons are used where N is the number of categories. The input neurons (or processing before the input layer)
standardizes the range of the values by subtracting the median and dividing by the interquartile range. The input neurons
then feed the values to each of the neurons in the hidden layer.
Hidden layer This layer has one neuron for each case in the training data set. The neuron stores the values of the
predictor variables for the case along with the target value. When presented with the x vector of input values from the
input layer, a hidden neuron computes the Euclidean distance of the test case from the neurons center point and then
applies the kernel function using the sigma value(s). The resulting value is passed to the neurons in the pattern layer.
Pattern layer / Summation layer The next layer in the network is different for PNN networks and for GRNN networks.
For PNN networks there is one pattern neuron for each category of the target variable. The actual target category of each
training case is stored with each hidden neuron; the weighted value coming out of a hidden neuron is fed only to the
2014, ICCTRD All Rights Reserved

Page | 64

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
pattern neuron that corresponds to the hidden neurons category. The pattern neurons add the values for the class they
represent (hence, it is a weighted vote for that category).
For GRNN networks, there are only two neurons in the pattern layer. One neuron is the denominator summation unit
the other is the numerator summation unit. The denominator summation unit adds up the weight values coming from
each of the hidden neurons. The numerator summation unit adds up the weight values multiplied by the actual target
value for each hidden neuron.
Decision layer The decision layer is different for PNN and GRNN networks. For PNN networks, the decision layer
compares the weighted votes for each target category accumulated in the pattern layer and uses the largest vote to predict
the target category.
For GRNN networks, the decision layer divides the value accumulated in the numerator summation unit by the value in
the denominator summation unit and uses the result as the predicted target value. The following diagram is actual
diagram or propose network used in our project.
1.Input Layer:
The input vector, denoted as p, is presented as the black vertical bar. Its dimension is R 1. In this , R = 3.

Fig 4.5 : PNN Network with RBF

2.Radial Basis Layer:
In Radial Basis Layer, the vector distances between input vector p and the weight vector made of each row of weight
matrix W are calculated. Here, the vector distance is defined as the dot product between two vectors. Assume the
dimension of W is QR. The dot product between p and the i-th row of W produces the i-th element of the distance
vector ||W-p||, whose dimension is Q1. The minus symbol, -, indicates that it is the distance between vectors. Then,
the bias vector b is combined with ||W- p|| by an element-by-element multiplication, .The result is denoted as n = ||Wp||.p. The transfer function in PNN has built into a distance criterion with respect to a center. In this paper, it is defined as
radbas(n) = 2 n e. Each element of n is substituted into Eq. 1 and produces corresponding element of a, the output
vector of Radial Basis Layer. The i-th element of a can be represented as ai = radbas(||Wi - p|| ..bi) where Wi is the
vector made of the i-th row of W and bi is the i-th element of bias vector b.
Some characteristics of Radial Basis Layer: The i-th element of a equals to 1 if the input p is identical to the ith row of
input weight matrix W. A radial basis neuron with a weight vector close to the input vector p produces a value near 1
and then its output weights in the competitive layer will pass their values to the competitive function. It is also possible
that several elements of a are close to 1 since the input pattern is close to several training patterns. .
3.Competitive Layer:
There is no bias in Competitive Layer. In Competitive Layer, the vector a is firstly multiplied with layer weight
matrix M, producing an output vector d. The competitive function, denoted as C , produces a 1 corresponding to the
largest element of d, and 0s elsewhere. The output vector of competitive function is denoted as c. The index of 1 in c is
the number of defect that the system can classify. The dimension of output vector, K, is 5 in this paper.
Although the implementation is very different, probabilistic neural networks are conceptually similar to K-Nearest
Neighbour (k-NN) models. The basic idea is that a predicted target value of an item is likely to be about the same as
other items that have close values of the predictor variables. The nearest neighbour classification performed for this
example depends on how many neighbouring points are considered.
A probabilistic neural network builds on this foundation and generalizes it to consider all of the other points. The
distance is computed from the point being evaluated to each of the other points, and a radial basis function (RBF) (also
called a kernel function) is applied to the distance to compute the weight (influence) for each point. The radial basis
function is so named because the radius distance is the argument to the function.
Weight = RBF (distance)
(4.10)
Different types of radial basis functions could be used, but the most common is the Gaussian function:

Fig 4.6: Radial Based Transfer Function

2014, ICCTRD All Rights Reserved

Page | 65

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
PNNs had been used for classification problems. The PNN classifier presented good accuracy, very small training time,
robustness to weight changes, and negligible retraining time. There are six stages involved in the proposed model which
are starting from the data input to output. The first stage is should be the image processing system. Basically in image
processing system, image acquisition and image enhancement are the steps that have to do. In this paper, these two steps
are skipped and all the images are collected from available resource. The proposed model requires converting the image
into a format capable of being manipulated by the computer. The defect images are converted into matrices form by
using MATLAB. Then, the PNN is used to classify the defect images. Finally, performance based on the result will be
analyzed at the end of the development phase.
V. CONCLUSION AND FUTURE WORK
The automatic defect detection from defect free environment with supervised classifier. Here, wavelet type and lifting
scheme with co-occurrence features were used for characterize the textures regions to discriminate the normal surface
and defective surface. It was well suited for this automatic process using learning machine with radial basis kernel
function. For training and classification, probabilistic type network has involved with RBF kernel. The performance of
this classification was evaluated with metrics of sensitivity and accuracy. Finally it shown that used classifier with
texture descriptors provided better classification accuracy and compatibility. It will be enhanced to detect and locate the
defect region using image segmentation approach. In Future, the segmentation process will be done based on an effective
segmentation approach for effective partitioning a defect region from other parts. After this process, Morphological
filtering will be used to smooth the segment the region by removing the back ground noise and false defects. Finally, the
weld defects are extracted with better accuracy. This process generates accurate detection of weld defects rather than
previous methods and other clustering models.
ACKNOWLEDGEMENT
Steel Surface images were kindly provided by Mr.T.Sundar Rajan, Owner of Sree Dhanabushanam Engineering Works ,
Private Company, Vellore..
REFERENCES
[1]
X. Li, S. K. Tso, X.-P. Guan, and Q. Huang, Improving automatic detection of defects in castings by applying
wavelet technique, IEEE Trans.Ind. Electron., vol. 53, no. 6, pp. 19271934, Dec. 2006.
[2]
R. Stojanovic, P. Mitropulos, C. Koulamas, Y. Karayiannis, S. Koubias, and G. Papadopoulos, Real-time visionbased system for textile fabric inspection, Real-Time Imaging, vol. 7, no. 6, pp. 507518, Dec. 2001.
[3]
A. Kumar, Neural network based detection of local textile defects, Pattern Recognit., vol. 36, no. 7, pp. 1645
1659, Jul. 2003.
[4]
A. Kumar, Computer-vision-based fabric defect detection: A survey, IEEE Trans. Ind. Electron., vol. 55, no. 1,
pp. 348363, Jan. 2008.
[5]
S. M. Chao and D. M. Tsai, An anisotropic diffusion-based defect detection for low-contrast glass substrates,
Image Vis. Comput., vol. 26, no. 2,pp. 187200, Feb. 2008.
[6]
W.-C. Li and D.-M. Tsai, Defect inspection in low-contrast LCD images using hough transform-based nonstationary line detection, IEEE Trans. Ind. Informat., vol. 7, no. 1, pp. 136147, Feb. 2011.
[7]
A. Mukherjee, T. Ray, S. Chaudhuri, P. K. Dutta, S. Sen, and A. Patra, Image-based classification of defects in
frontal surface of fluted ingot, Mesurement, vol. 40, no. 6, pp. 687698, Jul. 2007.
[8]
T. Sugimoto and T. Kawaguchi, Development of a surface defect inspection system using radiant light from steel
products in a hot rolling line,IEEE Trans. Instrum. Meas., vol. 47, no. 2, pp. 409416, Apr. 1998.
[9]
S. H. Choi, J. P. Yun, B. Seo, Y. Park, and S. W. Kim, Real-time defects detection algorithm for high-speed steel
bar in coil, Proc. World Acad.Sci., Eng. Technol., vol. 21, pp. 6670, Jan. 2007.
[10] W. X.-Yong, X. Ke, and X. J.-Wu, Application of undecimated wavelet transform to surface defect detection of
hot rolled steel plates, in Proc. Congr. Image Signal Process., Sanya, China, May 2008, pp. 528532.
[11] Y. Zhang, H. Zhou, S. J. Qin, and T. Chai, Decentralized fault diagnosis of large-scale processes using
multiblock kernel partial least squares,IEEE Trans. Ind. Informat., vol. 6, no. 1, pp. 310, Feb. 2010.
[12] C. H. Lo, E. H. K. Fung, and Y. K. Wong, Intelligent automatic fault detection for actuator failures in aircraft,
IEEE Trans. Ind. Informat.,vol. 5, no. 1, pp. 5055, Feb. 2009.
[13] M. D. Ma, D. S. H. Wong, S. S. Jang, and S.-T. Tseng, Fault detection based on statistical multivariate analysis
and microarray visualization, IEEE Trans. Ind. Informat., vol. 6, no. 1, pp. 1824, Feb. 2010.
[14] N. Cristianini and J. Shawe-Taylor, An Introduction to Support Vector Machines. Cambridge, MA: Cambridge
Univ. Press, 2000.
[15] S. Ghorai, A. Mukherjee, and P. K. Dutta, Discriminant analysis for fast multiclass data classification through
regularized kernel function approximation,IEEE Trans. Neural Netw., vol. 21, no. 6, pp. 10201029,Jun. 2010.
[16] N. Cristianini and J. S. Taylor, Kernel Methods for Pattern Analysis. Cambridge, U.K.: Cambridge Univ. Press,
2004.
[17] G. Strang, Linear Algebra and Its Applications, 4th ed. Jackson, WY:Nelson Engineering, 2007.
[18] R. C. Gonzalez and R. C. Woods, Digital Image Processing, 2nd ed.Upper Saddle River, NJ: Pearson Education,
2007.
[19] S. G. Mallat, A Wavelet Tour of Signal Processing, 2nd ed. New York: Academic, 1999.
2014, ICCTRD All Rights Reserved

Page | 66

[20]
[21]
[22]
[23]

Selvi et al., International Journal of Advanced Research in Computer Science and Software Engineering 4 (3),
March- 2014, pp. 60-67
X.-G. Xia, J. S. Geronimo, D. P. Hardin, and B. W. Suter, Design of prefilters for discrete multiwavelet
transforms, IEEE Trans. Signal Process.,vol. 44, no. 1, pp. 2535, Jan. 1996.
C.-C. Chang and C.-J. Lin, LIBSVM: A Library for Support Vector Machines,vol. 2, no. 3, p. 27, Apr. 2001.
[Online]. Available: http://www.csie.ntu.edu.tw/~cjlin/libsvm
T. M. Mitchell, Machine Learning. Singapore: McGraw-Hill, 1997.
Peng. Adaptive Median Filtering. Available:
http://www.massey.ac.nz/~mjjohnso/notes/59731/presentations/Adaptive%20Median%20Filtering.doc

2014, ICCTRD All Rights Reserved

Page | 67