Anda di halaman 1dari 5

Second International Conference on Computer Research and Development

AN IMPROVED METHOD OF SEGMENTATION USING FUZZY-NEURO


LOGIC

S.Sathish Kumar1, M.Moorthi2, M.Madhu3, Dr.R.Amutha4


1, 2, 3 Research Scholar,
Sri Chandrasekhardendra Saraswathi Viswa Mahavidyalaya University, Kanchipuram, India.
4. Professor and Head of the Department
Sri Venkateswara college of Engineering, Chennai, India.
msskm10@gmail.com, ramutha@svce.ac.in

Abstract—Image segmentation is an important process to


extract information from complex medical images.
Segmentation has wide application in medical field. The main
objective of image segmentation is to partition an image into II. EXISITNG METHODS
mutually exclusive and exhausted regions such that each
region of interest is spatially contiguous and the pixels within A. K-Means
the region are homogeneous with respect to a predefined K-means is one of the simplest unsupervised learning
criterion. Widely used homogeneity criteria include values of
algorithms that solve the well known clustering problem.
intensity, texture, color, range, surface normal and surface
The procedure follows a simple and easy way to classify a
curvatures. During the past, many researchers in the field of
medical imaging and soft computing have made significant given data set through a certain number of clusters.
survey in the field of image segmentation. This paper aims to
develop an improved method of segmentation using Fuzzy- Step1: Choose K Initial Centers Z1(1), Z2(2), which are
Neuro logic to detect various tissues like white matter, gray arbitrary.
matter; cerebral spinal fluid and tumor for a given magnetic Step2: At the Kth Iterative Step, distribute the sample
resonance image data set. Generally magnetic resonance {X} among the K Cluster Domain ,using the relation X €
images always contain a significant amount of noise caused by Sj(k)
operator performance, equipment, and the environment, which if || X- zj(k) || < || X-z i(k) ||, Where sj(k) is the set of
can lead to serious inaccuracies. So segmentation of such samples whose cluster center is zj(k).
medical images is a challenging problem in the field of image Step3: From the result of step 2, calculate the new
analysis. Several diagnostics are based on proper clusters zj(k+1), where j=1,2------k
segmentation of the digitized image. Segmentation of medical zj(k+1)=1/nj ∑ x ,X € Sj(k)
images is needed for applications involving estimation of the Where nj is the number of samples in sj(k) and the
boundary of an object, classification of tissue abnormalities, cluster centers are sequentially updated.
shape analysis, contour detection. In particular Fuzzy-Neuro
Step4: If zj(k+1)= zj(k) ,then the algorithm is said to have
logic segmentation algorithm is used to provide satisfactory
results compared to K-means, Fuzzy C-Means, Neural
converged and the procedure is terminated ,otherwise go to
Network and Fuzzy logic. step 2.
The major disadvantage of K- means algorithm is that it
contains several misclassified data points after segmentation
Keywords- Fuzzy-Neuro logic, segmentation, K-means, Fuzzy C- of brain image [2].
Means, Neural Network, Fuzzy logic.
B. Fuzzy C- Means
I. INTRODUCTION The fuzzy c-means clustering takes more time to perform
the classification distinct of tissue types. Fuzzy C-means
Image segmentation may be defined as a technique, Clustering (FCM) is a clustering technique which is employs
which partitions a given image into a finite number of fuzzy partitioning such that a data point can belong to all
nonoverlapping regions with respect to some groups with different membership grades between 0 and
characteristics, such as gray value distribution, texture 1.FCM is an iterative algorithm. The aim of FCM is to find
distribution, etc. Segmentation of medical images is required cluster centers (centroids) that minimize a dissimilarity
for many medical diagnoses like radiation treatment, function. The algorithm is defined by the following steps[3].
planning volume visualization of regions of interest (ROI) Step1. Randomly initialize the membership matrix U
defining boundary of brain tumor and intra cerebral brain according to
hemorrhage, etc. Many approaches are based on fuzzy logic,
K means and Neural Networks (NN).

978-0-7695-4043-6/10 $26.00 © 2010 IEEE 671


DOI 10.1109/ICCRD.2010.155
c Figure 1. Architecture of RBF Neural Network
¦u i 1
ij 1, j 1,..., n . (1)

It is a multilayer feed forward network. There exists n


¦
n m
j
u xj
1 ij number of input neurons and m number of output neurons
Step2. Calculate the centroid using ci with hidden layers existing between the input and output
¦
n m
j 1
u ij layers. The interconnection between the input and hidden
layers forms hypothetical connection and between the hidden
Step3. Compute dissimilarity between centroids and data
layer and output layers forms weighted connections. The
points using
c c n
training algorithm is used for the update of weights in all the
¦ Ji ¦¦ u interconnection [7].
m 2
J (U , c1 , c2 ,..., cc ) ij d ij
i 1 i 1 j 1 The important aspect of the RBF Network is the usage of
Where, uij is between 0 and 1; ci is the centroid of cluster activation function for computing the output, where it uses
i; the Gaussian activation function. The response of such a
dij is the Euclidian distance between ith centroid(ci) and jth function is non negative for all values of x. The function is
data point; and m є [1,∞] is a weighting exponent. Stop if its defined as f(x) = exp(-x2)
improvement over previous iteration is below a threshold.
Step4. Compute a new U using The training algorithm
Step 1: Initialize the weights
1 Step 2: While stopping is false do step 3 to step 10
uij 2 /( m 1)
and go to step 2.
Step 3: For each input do step 4 to step 9
§ d ij ·
¦ ¨ ¸ Step 4: Each input unit xi ,i=1----n.Receives input signals
c
1¨ ¸ to all units in the layer above (hidden unit)
© d kj
k
¹ Step5: Calculate the radial basis function
By iteratively updating the cluster centers and the Step 6: Choose the center for the RBF, the centers are
membership grades for each data point, FCM iteratively chosen from the set of input vectors. A sufficient number of
moves the cluster centers to the "right" location within a data centers have to be selected in order to ensure adequate
set. sampling of the input vector space
FCM does not ensure that it converges to an optimal Step 7: The output of the unit vi(xi) in the hidden layer
solution because of cluster centers (centroids) being and is given as vi(xi) = exp (xji ^ -xji) 2 / σi2
initialized using U that is randomly initialized as in (1). This Where xji^ = center of the RBF unit for input
problem must be considered to improve the effectiveness of variables, σi2 = width of the ith RBF unit , Xji = j th variable
the FCM algorithm. of input pattern
C. Neural Network Step 8: Initialize the weights in the output layers
of the network to some random value
A Neural Network is used in order to significantly reduce Step 9: Calculate the output of the network using
the computation time, and misclassification rate of the Ynet= wim vi (xi)+ wo, Where, H= number of hidden layer ,
system [4-5]. The architecture of RBF Network consists of 3 Ynet= output value of the mth node in output layer for the
layers namely, the input, the hidden and the output layers as nth incoming pattern, Wim = weight between ith RBF unit and
shown in Fig.1 mth output node, Wo = biasing term at mth output node
1 Step 10: Calculate error and test the stopping condition.
D. Fuzzy Logic
Fuzzy set theory provides a host of attractive aggregation
Φ connectives for integrating membership values representing
Xi1 Zi1 uncertain information. The membership function of a fuzzy
set in a functional form typically may be a bell-shaped
function, triangle-shaped function, trapezoid-shaped
Xi1 Φ function, etc.
In fuzzy systems, describing the control rules is usually
simpler and easier, often requiring fewer rules, and thus the
Zi1
systems execute faster than conventional systems. Fuzzy
systems often achieve tractability, robustness, and overall
Φ low cost.
Xi1
The procedure for obtaining the fuzzy output of such a
knowledge base can be formulated as follows [1],[6],
Input layer Hidden layer Output layer 1) The firing level of the ith rule is determined by
Ai(x0) × Bi(y0).
2) The output of the ith rule is calculated by

672
Ci(w) = Ai(x0) × Bi(y0) → Ci(w) for all w € W
Input Image
3) The overall system output, C, is obtained from the
individual rule outputs Ci by
C(w) = Agg{C1, . . . , Cn}for all w € W.
Pre-processing
III. PROPOSED METHOD
The new algorithm is proposed to overcome the above
said drawbacks
Step 1: Read the given image, and convert it into a matrix Membership functions and
FL
form where each pixel value is in the range from 0-255. Cluster centers
Step 2: Apply median filtering to remove noise.
Step 3: Create fuzzy interference system.
Step 4: Decide the number and type of Membership Feature extraction
functions for the input image by tuning the membership
functions. The following fuzzy rules are used in the proposed
algorithm for clustering. ANN
1) If the mean value is low and the standard deviation Similarity Measures
value is low then it is not an edge pixel.
2) If the mean value is medium and the standard
deviation value is low then it is an edge pixel.
Update Membership functions
3) If the mean value is high and the standard deviation and Cluster centers
value is low then it is not an edge pixel.
4) If the mean value is low and the standard deviation
value is high then it is not an edge pixel. Cost functions
5) If the mean value is medium and the standard
deviation value is high then it is not an edge pixel.
6) If the mean value is high and the standard deviation
Figure 2. Flow Diagram
value is high then it is not an edge pixel.
Step 5: Apply Fuzzification using the rules developed
above on the corresponding pixel values of the input image Finally, the parameters referring to the degree of
which gives a fuzzy set represented by a membership feature extraction are determined using RBF neural network
function. Check the rules using rule viewer and surface model. Thus, the above mentioned algorithm is used to
viewer segment the white matter, gray matter, cerebral spinal fluid
Step 6: Choose the cluster centers and extract the features (CSF) and tumor for a given MRI data set.
Step 7: Clustering is done based on similarity measures
Step 8: Update the membership function and cluster IV. RESULTS
centers
Step 9: Calculate the cost function and update the The following shows the input MRI image for
weights segmentation
Step 10: Test the stopping condition and convert the
column form to matrix form and display the segmented
image.
In Fig.2, first the input image is preprocessed using
median filter to remove noise and then the membership
functions are defined for the filtered image and clusters are
formed using fuzziness .Then neighborhood attraction is
considered to exist between neighboring pixels. This is
considered as a feature for extraction. During clustering,
each pixel attempts to attract its neighboring pixels toward
its own cluster. This neighborhood attraction depends on two
factors; the pixel intensities or feature attraction, and the Figure 3. Input Image
spatial position of the neighbors or distance attraction, which
also depends on the neighborhood structure. Consideration
of these neighboring pixels greatly restrains the influence of
noise.

673
The following are the segmented outputs of the given Thus it is seen that the misclassification rate in
MRI image using the various methods such as K-means, classifying the white , gray and CSf are highly reduced in the
Fuzzy C-means, Fuzzy Logic and Fuzzy-Neuro logic. proposed method as compared to the previous algorithms.

TABLE I. COMPARISONS OF VARIOUS ALGORITHMS

Misclassification Rate
Name of
Algorithms WHITE GRAY
CSF
MATTER MATTER
K-Means 73.68 30.10 73.68
Figure 4. K Means
Fuzzy C
29.60 30.50 29.60
Means
RBF Neural
43.51 28.28 43.51
Network
41.86 51.33 41.80
Fuzzy Logic
Fuzzy –
4.411 21.012 4.41
Neuro Logic

Figure 5. Fuzzy C Means V. CONCLUSION


In this paper, the proposed method of Fuzzy – Neuro
logic, segments the different regions of the given MRI brain
image data. It is seen that the misclassification rate is less
using the FNL method as compared with the other previous
methods.
Future work will focus on developing an algorithm for
reducing the misclassification rate (MR) in clustering the
gray matter, where the MR is 21.012. More comprehensive
comparison of FNL and the generalization of the ANN
Figure 6. Fuzzy Logic model will be addressed.
REFERENCES
[1] Addallah A.Alshemawy, and Ayman A.Aly,”Edge Detection in digital
images using Fuzzy logic technique”, World Academy of science
,engineering and technology 51, pp 178 - 186 , 2009.
[2] Cheng Wen Cheng, Jiebouo, Kevin J.Parker,” Image segmentation via
Adaptive k-means clustering and knowledge based morphological
operation with biomedical application”, IEEE transaction on image
processing, Vol.7, No.12, pp 1682 – 1688, 1998.
[3] D. Q. Zhang and S. C. Chen, “A novel kernelized fuzzy c-means
algorithm with application in medical image segmentation,” Artif. Intell.
Figure 7. Fuzzy – Neuro Logic Med., vol. 32, pp. 37–52, 2004.
[4] Hall L.O,Bensaid .A, Clarke .L,” A Comparison of Neural Network and
Fuzzy clustering technique in segmenting magnetic resonance images of
It can be seen from the table I that the misclassification the brain”, IEEE transaction on Neural network, Vol.3, pp 672 – 682, 1992.
[5]Haykins .S, “Neural networks a comprehensive foundation”, Prentice
rate (MR) for white matter, gray matter and CSF is Hall Inc second edition, 1999.
73.68,30.10 and 73.68 using K-means algorithm, is 29.60, [6] S. Shen,W. A. Sandham, M. H. Grant, J. Patterson, and M. F. Dempsey,
30.50 and 29.60 using FCM algorithm, is 43.51, 28.28 and “Fuzzy clustering based applications to medical image processing,” in
43.51 using RBF Neural Network, is 41.86, 51.33 and 41.80 Proc. IEEE EMBS 25th Annu. Int. Conf., 2003, pp. 747–750.
using Fuzzy logic and 4.411, 21.012 and 4.41 using Fuzzy- [7] Shah Shen, William Sandham , “MRI fuzzy segmentation of Brain
Neuro logic. tissue using Neighborhood Attraction with Neural-Network
Optimization”IEEE Transaction on Information Technology in
Biomedicine,Vol.4,No.3, pp 459- 467 , Sept 2005.
Mr.S.Sathish Kumar is pursuing Kanchipuram. He completed his B.E degree in Electronics
his Ph.D program at Sri and Communication Engineering in the year 2001 and
Chandrasekhardendra Saraswathi Viswa M.E Applied Electronics in the year 2006 , at Arulmigu
Mahavidyalaya University,

674
Meenakshi Amman College of Engineering, Kanchipuram,
Chennai, India.
He has 9 years of teaching experience and he is
currently working as Associate Professor in the department
of Electronics and Communication Engineering at
Prathyusha Institute of Technology and management,
Chennai. He has published and presented many papers in
National and International Conference in the area of Image
processing. His research interests are Image Segmentation,
Image Compression and Image detection, speech
processing.

Mr.M.Moorthi is pursuing his Ph.D


program at Sri Chandrasekhardendra
Saraswathi Viswa Mahavidyalaya
University, Kanchipuram. He completed
his B.E degree at Arulmigu Meenakshi
Amman College of Engineering,
Kanchipuram, in Electronics and
Communication Engineering in the year 2001 and M.E
Medical Electronics in the year 2007 at Anna University,
Guindy campus, Chennai, India.
He has 9 years of teaching experience and he is
currently working as Associate Professor in the department
of Electronics and Communication Engineering at
Prathyusha Institute of Technology and management,
Chennai. He has published and presented many papers in
National and International Conference in the area of Image
processing. His research interests are Image Segmentation,
Image Compression and Image detection.

Mr.M.Madhu is pursuing his Ph.D


program at Sri Chandrasekhardendra
Saraswathi Viswa Mahavidyalaya
University, Kanchipuram. He completed
his B.E degree in Electronics and
Communication Engineering in the year
2001 and M.E Applied Electronics in the
year 2006, at Arulmigu Meenakshi Amman College of
Engineering, Kanchipuram, Chennai, India.
He has 7 years of teaching experience and he is
currently working as Assistant Professor in the department
of Electronics and Communication Engineering at Rajiv
Gandhi College of Engineering, Sriperumbudur, Chennai.
He has published and presented many papers in National
and International Conference in the area of Image
processing. His research interests are Image Segmentation,
Image Compression and Image detection, speech
processing.

675

Anda mungkin juga menyukai