International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017
Image segmentation using fuzzy Lomax-Gumbel
Distribution
Dillip Ranjan Nayak1 , Basanta Kumar Swain2
1
Asst.Prof, Dept. of Computer Science & Engg,
Govt.College of Engg. Kalahandi, Odissa, India
2
Asst.Prof, Dept. of Computer Science & Engg,
Govt.College of Engg. Kalahandi, Odissa, India
Abstract: Images contain lots of uncertainties due to
randomness. The contrast between the objects and the
background can also be viewed as fuzzy in nature. Fuzzy
approach is chosen to minimize ambiguity inherent in the
images. So, different fuzzy thresholding techniques have been
developed to remove the grayness ambiguity/vagueness using
threshold selection. In this paper, a new fuzzy thresholding
technique is introduced which processes Lomax-Gumbel
probability distribution as the membership function. To show
the effectiveness of the proposed method, the results are
compared with fuzzy divergence method using gamma
membership function. Experiments on several images show that
the proposed segmentation method performs better than the
existing method.
Keywords: Lomax- Gumbel distribution; Fuzzy
Divergence; Membership function; Thresholding; Signal to
noise ratio.
1. INTRODUCTION
Fuzzy thresholding theory is a fuzzy image processing tool
for handling the ambiguity in the images associated with
vagueness and increases the contrast of an image. It is a
popular, fast and inexpensive segmentation technique [1]
.The degree of ambiguity and uncertainty of an image can
be measured by the fuzzy entropy. Fuzzy Entropy is
analogous to the Shannon Entropy [2]. Entropy contains
randomness uncertainty but fuzzy entropy contains
vagueness uncertainty. This vagueness is defined by a
membership function. De Luca and Termini [3] uses the
concept of Shannons probabilistic entropy to define fuzzy
entropy. Pal and Rosenfeld [4], Huang and Wang [5] used
the fuzzy membership functions for image thresholding.
The discrepancy between two probability distributions is
called divergence. Kullback and Leibler [6] defined KL
divergence which based on the idea of Shannons entropy.
Measuring the difference between two fuzzy sets by means
of a function which we call fuzzy divergence [7]. Bhandari
and Pal, [8] described the measure of directed divergence
between fuzzy sets. Pal and Pal [9] used the fuzzy
exponential entropy to segment an image. Fan and Xie [10]
defined fuzzy divergence from fuzzy exponential entropy
for image segmentation.
Chaira and Ray [11], [12] used fuzzy divergence and
gamma membership function to threshold an image. Ghosh
Volume 6, Issue 5, September October 2017
ISSN 2278-6856
et al. [13] introduced an automated approach to leukocyte
recognition using modified Cauchy techniques.
The Gumbel distribution is an extreme value distribution
of type I, which applied for problems in engineering
[14].Singh and Srivastava used Gumbel probability
distribution for Retinal blood vessels segmentation [15].
The Lomax distribution is a heavy tail probability
distribution used in medical and biological sciences
[16].Zahrani and Sagor [17] defined PLD which used for a
wider range of statistical applications in image processing.
Gupta et al. [14] introduced new statistical distribution
constructed by combining the cumulative density functions
(CDF) of Lomax and Gumbel distributions named as
Lomax-Gumbel distribution.
The main problem in thresholding is to find an optimum
threshold. This process becomes complex when the image
quality is not good due to noise or imprecision, and
fuzziness. In this paper, we used fuzzy lomax-gumbel
principle to threshold the images. Here the thresholds are
selected in the gray scale on the basis of obtained
divergence values.
2. METHODLOGY
In this work, we modified fuzzy divergence of Chaira [11]
by introducing a new lomax-gumbel membership function.
The results are compared by using signal to noise ratio
calculation after thresolding in both the cases. In
multimodal histograms, a selection strategy searches
threshold values in the region between the two consecutive
peaks of the histogram, while in unimodal thresholding
linear search has been used from minimum to maximum of
the gray value. For each threshold value, the membership
values of the threshold image are compared with an ideally
threshold image
2.1 Fuzziness and Membership Function
Let A be an image of size M M having L gray levels .It
has the membership value of (i, j) pixel in A is ( ) ,
where 0 ( ) 1 . Mathematically the fuzzy image is
represented by A= { , }, . Let the count
(f) denote the number of occurrences of the gray level f in
the image. Let t is the threshold value, which separates the
Page 167
 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856
object and the background of the image. The average gray ( , )
level of the background region and object region = [1 (1 ).
respectively is finding by the relation:
. ( )
( ). ] (9)
= ( )
(1) In a similar manner, the discrimination of image B against
=
. ( )
(2) image A is:
( ) ( , )= [1 (1
The membership values of the pixels have been determined
using proposed lomax-gumbel distributions ). (( ). ](10)

2.1.1 lomax-gumbel distribution Therefore, the total fuzzy divergence between images A
Lomax-Gumbel Distribution [14] a new statistical and B is
distribution constructed by combining the cumulative ( , )= ( , ) + ( , ).
density functions (cdf) of Lomax and Gumbel So, D (A, B) = [2 (1 +
distributions. The cumulative density function (cdf) of ). (1-
Lomax distribution is given by
( + ). ] (11)
( ; , )=1- 1 + , , >0, x 0 (3)
where (i,j)=0,1.(M-1)and the
The CDF of the Gumbel distribution is given by membership value of the pixel where and are
the (i,j)th pixel of the image A and B.
( )= , , 0, > 0 (4)
Gupta et al. [14] are defined a new distribution using equ 2.3 Threshold selection
(3) and equ (4) is
( ) The membership values of the thresholded image are
( ; , , , ) = = compared with an ideally threshold image. So, the above
( )
Eq. (11) becomes
1 1+ (5) D(A,B) = [2 (1 + 1).
Where = 1 (1 + ) , , , > 0, ( ). (12)
, 0 , is the shape parameter, is the location The divergence value of each pixel is calculated for whole
parameter and is the scale parameter. image and corresponding gray level is noted. The gray
For getting minimum divergence value, we proposed the value corresponding to the minimum divergence is chosen
locmax-gumbel function for our experiment is: as threshold.
( ; , , , )=1-
2.4 Signal-to-Noise ratio
1 1+ (6) The performance of proposed method with chairas
When 0, = , = 2, = 1 = 1, the method is evaluated by using the measuring signal noise
membership function of the object region is computed ratio which defined as:
Signal-to-noise ratio (SNR)
( ) =1-3 1 1 + 0.5 (
= (13)
) if aij>t (7) The comparative study is presented in Table 1.

In likewise manner, the membership function of the 3. EXPERIMENTAL RESULTS

background region is computed by choosing = is given
by
( ) =1-3 1 1 + 0.5
) if aij (8)
Where C is constant and its value is 1/ (fmax-fmin), where
fmin and fmax are the minimum and maximum gray level in
the image respectively and t is the threshold value.
The proposed fuzzy thresholding algorithm described
above is tested on cameraman image size of 256x256,
blood image size of 160x160, blood1 image size of
(
130x130 and blood cell image size of 140x140. Figs. 1(a)-
4(a) are the original images, Figs. 1(b)-4(b) are the
histograms of original images, Figs 1(c)-4(c) and 1(d)-4(d)
are the results obtained by the method in fuzzy divergence
by chaira and the proposed method, respectively.

2.2 Fuzzy divergence

Chaira extended the concept fuzzy divergence for getting
threshold for an image using from fuzzy exponential
entropy which defined by Fan and Xie [10]. In an image of
size M M with L distinct gray levels, the discrimination
of image A against image B is defined as

Volume 6, Issue 5, September October 2017 Page 168

 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856

(a)

600
(a)
1000 500

800 400
600 300
400 200
200
100
0
0
0 50 100 150 200 250
0 50 100 150 200 250
(b) (b)

(c) (c)

(d)
Figure 1 (a) Input image cameraman, (b) histogram of the
cameraman image, (c) thresholded using fuzzy divergence,
(d) thresholded using proposed method
(d)
Figure.2 (a) Input image blood (b) histogram of the blood
image, (c) threshold using fuzzy divergence, (d) threshold
using proposed method

Volume 6, Issue 5, September October 2017 Page 169

 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856

(a)
(a)

(b)

(b)

(c)
(c)

(d) (d)
Figure 3 (a) Input image blood1 (b) histogram of the Figure 4 (a) Input image blood cell (b) histogram of the
blood1 image, (c) threshold using fuzzy divergence, (d) blood cell image, (c) threshold using fuzzy divergence, (d)
threshold using proposed method threshold using proposed method

Volume 6, Issue 5, September October 2017 Page 170

 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856

0.05 0.28

0.04
0.27
0.03

0.02 0.26

0.01
0.25
0
0 50 100 150 200
0 100 200 300 (b)
(a) Figure 6 (a) Plots of gray level/fuzzy threshold using
fuzzy divergence method on the blood image, (b) plot
0.36 using proposed method

0.34 0.04

0.32
0.03

0.3
0.02
0.28

0.01
0.26 0 50 100 150 200 250
0 100 200 300
(a)
(b)
0.04
Figure 5 (a) Plots of gray level/fuzzy threshold using
fuzzy divergence method on the cameraman image, (b)
plot using proposed method 0.03

0.016
0.02
0.014

0.012 0.01
0 50 100 150 200 250
0.01 (b)
0.008 Figure 7 (a) Plots of gray level/fuzzy threshold using
fuzzy divergence method on the blood1, (b) plot using
0.006 proposed method
0.004
0 50 100 150 200
(a)

Volume 6, Issue 5, September October 2017 Page 171

 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856
-3
x 10 Blood1 112 1.0607 112 1.0607
10
Blood cell 31 0.4912 21 0.7113

8
4. CONCLUSION
Based on the concept of fuzzy divergence with lomax-
6 gumbel, a new thresholding method is proposed in this
paper. We have conducted experiments on many images.
Proposed method compare with Chairas fuzzy divergence
method. The thresholds obtained in proposed case yield
4 better segmentation of the objects from the background. It
0 50 100 150 200 avoids over segmentation and give smooth enhance result.
(a)
0.27 References
[1] T. Chaira., 2010. Intuitionistic Fuzzy Segmentation of
0.265 Medical Images. IEEE Transactions on
Biomedical Engineering. Vol. 57, No. 6, 1430-1436.
0.26 [2] W., Sander, 1989. On measure of fuzziness, fuzzy sets
and system. 29, 49-55.
[3] A. De Luca,., Termini, S., 1979. Entropy and energy
0.255
measure of fuzzy sets, in advances in Fuzzy set
Theory and applications, Elsevier, pp.321-328.
0.25 [4] A. K. Pal,., A. Rosenfeld,., 1988. Image enhancement
and thresholding by optimization of uzzy
0.245 Compactness. Pattern recognition letters, 7, 77-86.
0 50 100 150 200 [5] L.K. Huang,., M.J. Wang,., 1995. Image thresholding
(b) by minimizing the measure of fuzziness. Pattern
Figure 8 (a) Plots of gray level/fuzzy threshold using Recognition. 28 (1), 4151.
fuzzy divergence method on the blood cell, (b) plot using [6] S. Kullback,. R..A. Leibler., 1951. On information and
proposed method sufficiency. Ann. Math. Statist. 22, 79 86.
[7] S. Montes, I. Couso, P. Gil, C. Bertoluzza, 2002.
In Table 1, signal to noise ratio and gray values are Divergence measure between fuzzy sets. Int. J.
statistically represented for each type of the images. We Approx. Reason. vol. 30, pp. 91105.
found the value of SNR of the proposed method is better in [8] D. Bhandari, N.R. Pal 1993. Some new information
all cases except blood1 image where it is equal. The SNR measure for fuzzy sets. Information Sciences. vol.67,
is high means the images are less noise and more clear. No.3, pp.209228.
Plots for fuzzy divergence against the gray level are shown [9] N. R. Pal, , S. K. Pal., 1992. Some properties of the
in Fig 5, Fig 6, Fig 7 and Fig 8. It is observed from the exponential entropy. Inform. Sci. 66, 119 137.
plots that the fuzzy divergence is minimum at a particular [10] J. Fan, W. Xian, 1999.Distance measure and induced
gray level. Proposed Figure (Fig 5(b)-8(b)) is more flat fuzzy entropy. Fuzzy Sets System 104, 305 314.
than divergence method. So it gives better result. [11] T. Chaira, , A.K. Ray,., 2003. Segmentation using
Fig. 4(a) is a low contrast, vague image. It is very clear that fuzzy divergence. Pattern Recognition Lett. 12 (24),
the image obtained by the proposed method Fig. 4(d) more 18371844.
distinct than Fig. 4(c).In the case of Fig.1 and Fig.2, we [12] T. Chaira, , A.K. Ray, 2004. Threshold selection using
found that the images obtained by lomax-gumbel method fuzzy set theory. Pattern Recognition Letters 25 (12),
are more segmented compared than other method. From 865874.
all the point of view, it is observed that lomax-gumbel [13] M. Ghosha,., D. Dasa, C. Chakrabortya, A .K. Ray,.
membership leads to better segmentation in comparison 2010. Automated leukocyte Recognition using Fuzzy
divergence. Micron .41, 840846.
with Gamma method [11].
[14] S.P Goyal., J., Gupta, M. Garg, M., Gupta, 2016. The
Lomax-Gumbel Distribution. Palestine Journal of
Table 1: Signal to noise ratio and gray values Mathematics.Vol. 5(1), 3542.
[15] N.P. Singh, R., Srivastava, 2016.Retinal blood vessels
Type Threshold SNR Threshold by SNR by segmentation by using Gumbel Probability
Image by Gamma by Gamma Lomax-Gumbell Lomax-Gumbell
Cameraman 87, 88 1.6564 86 1.6653
Distribution function based matched filter. Computer
methods and programs in Biomedicine, 129.40-50.
Blood 38 0.5766 34 0.6316

Volume 6, Issue 5, September October 2017 Page 172

 International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org
Volume 6, Issue 5, September- October 2017 ISSN 2278-6856
[16] H.A. Rady, W.A. Hassanein, T.A. Elhaddad, 2016.
The power Lomax distribution with an Application to
bladder cancer data. Springer Plus 5:18-38.
[17] B.A. Zahrani, H. Sagor, 2014.The Poisson-Lomax
Distribution. Revista Colombiana de Estadstica
volume 37, no. 1, pp. 223- 243.

Mr. Dillip Ranjan Nayak received the

B.Tech. (CSE) degree from North Orissa
University. He completed his M.Tech.
(Computer Science)from NIT Rourkela. He
worked as Asst. Prof. of CSE in C.V.Raman
College of Engineering, Odisha for five years. Presently,
he is working as Asst. Prof in Govt. College of
Engineering, Kalahandi, under Biju Patnaik University of
Technology, Odisha, India. He has published 19 research
articles. His research interests include Image
Segmentation,Image Extracion and Medical Imaging.

Dr. Basanta Kumar Swain is presently

serving as a
Assistant Professor in Computer Science &
Engineering, at GCEK, Bhawanipatna
Odisha.
Formerly, he was working as a Lecturer at
post graduate level in department of CSA, Utkal
University, BBSR, Orissa. He has been doing his
research work in the domain of Speech Processing for
Oriya language since last 10 years. His area of interest
includes Speech Recognition, Speaker Identification,
Data Mining and Pattern Recognition. He has a number of
publications to his credit in international journals and
conferences.

Volume 6, Issue 5, September October 2017 Page 173