Sunteți pe pagina 1din 6

Texture and Clustering-based Skin Disease Classification

Pushpa Kotipalli1, Pradeep Mullangi1, and Y. Srinivasa Rao2


ECE Department, Shri Vishnu Engineering College for Women, Andhra Pradesh, India
Instrument Technology Department, A.U. College of Engineering, Andhra Pradesh, India

Abstract. The main objective of this paper is to classify the skin diseases using
image classification methods. For examining texture of image, a statistical method gray-level co-occurrence matrix (GLCM) was used. GLCM considers the
spatial relationship of pixels and characterize texture of an image by calculating how often pairs of pixel with specific values and specified spatial relationship occur in an image. The presented work here is focused on extraction of GLCM features inclusive of contrast, correlation, homogeneity and
energy. Fuzzy C-means Clustering along with GLCM is proposed for reducing
the time taken for skin disease classification. With simulation results, it is
shown that the proposed method is more efficient than GLCM alone method.
Keywords: Image Classification, GLCM, Fuzzy C-means clustering, and Skin


Skin disorders are among the most common diseases in both developing and industrialized countries. People living with skin disease experience stressful life as skin
disease affects their confidence and self-esteem in so many different ways. In 2013,
with prevalence rate of 10 percent, the population affected across India from skin
disease is estimated at nearly 15.1 crore. The situation is further worsened by the low
availability of dermatologists in India [1]. In this scenario, medical imaging plays an
important role for quick decision making in skin disease identification. Medical imaging is used for revealing internal structures hidden by the skin as well as to diagnose
and treat disease. This paper discusses detection of skin diseases using texture and
clustering based image classification.

Literature Survey

Sparavigna [2] proposed a texture based method in which differences in color and
coarseness of skin are quantitatively evaluated by using statistical approach to the
pattern recognition. Abbadi [3] developed a skin disease recognition method
based on GLCM and skin color. In this paper, GLCM is used for measuring texture

features such as correlation, contrast, energy and homogeneity. Then Fuzzy C-means
clustering technique is used with texture features extracted by GLCM.
The remainder of this paper is organized as follows. Proposed method is presented
in Section 2. Extraction of image features using GLCM is discussed in Section 3.
Fuzzy C-means clustering is introduced in Section 4. Section 5 presents simulation
results and Section 6 concludes the paper.

Proposed Method

Methodology of proposed method is shown in the Flow chart (Fig. 1).

Fig. 1. Flow chart of proposed method

This method is developed for classification of skin diseases by analyzing textures

obtained from a collection of images using features based on GLCM. This method has
two phases: (a) Training phase and (b) Classification phase.
We considered extraction of four texture features of an image: Correlation, Contrast, Homogeneity and Energy. In training phase, the texture features are extracted

from a given set of training images. In classification phase, the given test image is
segmented and then the above mentioned texture features are extracted. Then, Fuzzy
C-means clustering is applied on the texture features obtained from training images.
Then Chi. Square distances are calculated for skin disease classification.
Computing GLCM and extracting texture features of the image are discussed in the
section that follows.

Gray Level Co-occurrence Matrix

GLCM contains the second-order statistical information of spatial relationship of

pixels of an image. Texture feature calculations use the contents of the GLCM to give
a measure of the variation in intensity at the pixel of interest [4]. In computing GLCM
of an image two parameters Offset and Distance d between pixels are considered.
Here, offset represents the direction of pairing pixels. For example, with 0 offset and
d = 1 represents pairing horizontally adjacent elements. In this paper d = 1 and four
offsets 0, 45, 90, and 135 are considered. Offsets 45, 90 and 135 are obtained
by pairing the right diagonal elements, vertically upward elements and left diagonal
elements respectively.

GLCM of an Image

Let us consider a 4x4 section of an image having 4 grey levels as shown in Fig. 2(a).
Image data and GLCM with 0 offset are shown in Fig. 2(b) & Fig.2(c) respectively.

Fig. 2. (a) Small Section of image having 4 grey levels, (b) Image data and (c) GLCM of the
image with 0 offset and d = 1

Let G0, G45, G90, and G135 represent GLCMs with offsets 0, 45, 90, and 135
respectively. To get the symmetrical GLCM, original GLCM is added to its transpose.
Then, normalized symmetrical GLCM is obtained by dividing it with sum its elements. Normalized symmetrical GLCMs N0, N45, N90, and N135 respectively are

1 0
N 0o
22 0


0 0 1
4 3 0
3 4 0

0 0 4

N 45o

1 2

18 1

2 1 0
2 1 2
1 2 0

2 0 2

N 90o

1 3

24 1

3 1 0

2 2 2
1 2
2 2 0
20 0

2 0 4

2 0 1
0 2 1
2 2 2

1 2 2

Then, texture features like Correlation, Contrast, Energy and Homogeneity of the
given image are extracted from normalized symmetrical GLCMs.

Texture measures from GLCM

Most of the texture calculations are weighted averages of the normalized symmetrical
GLCM cell contents.
Contrast (Con): Contrast is a measure of intensity or gray-level variations between
the reference pixel and its neighbor.


i , j 0

i, j

(i j ) 2

where pi , j represents element (i,,j) of the normalized symmetrical GLCM, K

represents number of gray levels in the image and (i-j)2 represents the weight.
Correlation (Cor):. It presents how a reference pixel is related to its neighbor. It has
values from -1to1.-1 refers to uncorrelated & 1 is perfectly correlated.
K (i )( j ) p
i, j

i j

i , j 0

i , j and i , j represent the means and standard deviations of pi , j respectively.

Homogeneity (H): Homogeneity returns a value that measures the closeness of the
distribution of elements in the GLCM to the GLCM diagonal.
pi , j
i , j 11 | i j |
Energy (E): It measures the textural uniformity that is pixel pair repetitions. It detects
disorders in textures. It simply returns the sum of squared elements in the GLCM.

i , j 1


i, j

GLCM Based Classification

Let (N0, N45, N90, N135) Tn represent set of normalized symmetrical GLCMs for the
training image Tn; n = 1, 2,, N. For each set of normalized symmetrical GLCMs,
above mentioned four texture features are computed and final values of texture features are obtained by taking average of individual feature set of each image [5]. Let
(Cor, Con, H, E)Tn represent final values of texture features of training image Tn; n =
1, 2,,N. In the similar way average texture feature values of Test image are computed. Then the skin disease classification is done by computing Chi.square distance
between train and test images.

Fuzzy Logic and GLCM based Skin Disease Classification

GLCM is implemented on all the training images, from which texture features of each
and every training image are obtained. Then the fuzzy logic (C - means clustering) is
implemented on those values of the training image set. Fuzzy C-Means (FCM) clustering is an iterative process. First, the initial fuzzy partition matrix is generated and
the initial fuzzy cluster centers are calculated. In each step of the iteration, the cluster
centers and the membership grade point are updated and the objective function is
minimized to find the best location for the clusters. The process stops when the maximum number of iterations is reached [6]. In this way cluster centers make set of N
values of each feature to one value which in turn reduces computational complexity
drastically. Then, classification is done based on similarity between cluster centers of
training classes and the test image features.

Simulation Results

Images of skin diseases Angioedema, Actinic Keratosis and Eczema are collected
from They are cropped to 100x100 pixels size and converted to
gray-scale. Then, they are divided to training and test image sets.

Implementation of fuzzy logic

Firstly Fuzzy logic is implemented on the class Angioedema and the results are as
shown in Fig. 3, where 1(bolded) represents the cluster center. From the Fig. 3, it can
be observed that we get a single value for each feature. FCM clustering is applied on
the other two classes Keratosis and Eczema and the results are given in Table 1.

Fig. 3.

Fuzzy C-means clustering on Angioedema

Table 1.

Features of three classes of skin diseases after clustering

Skin disease class

Actinic Keratosis





Skin Disease Classification

Skin disease classification is done using Chi. square distance.

Chi Square distance, d

(Sn M m )2

n 1 S n M m

m 1,2,...

Sn, Mm and N represent nth training image, mth test image and number of training
images respectively. N is large for GLCM method where as N = 1 for the proposed
method. The procedure is repeated for all the three classes. Then the class, whose
distance comes out to be minimum, is said to be the class the test image belongs to.

Efficiency Calculation

% Efficiency=100 - %Error where %Error=No. of false detections/No. of test images.

In simulations we obtain % Efficiency with GLCM method as 79% whereas % Efficiency with GLCM and Fuzzy C-means clustering method as 84%.


Three skin disease classes Angioedema, Actinic Keratosis and Eczema are considered
for skin disease classification. Texture features are extracted using normalized symmetrical GLCMs. Fuzzy C-means clustering is applied on the texture features of three
classes of training data sets. Then Chi. Square distance is used for image classification. Proposed GLCM and FCM method reduces the time consumed in skin disease
classification and also, it exhibits high efficiency as compared to GLCM method.
1. Ankit Goyal, Around 19 Crore Indians Likely to Suffer from Skin Diseases by 2015
notes Frost & Sullivan, article, May 03, 2014.
2. A.Sparavigna, R.Marazzato, An image processing Analysis of skin textures, Skin research and technology, Vol.16, Issue.2, pp. 161-167, 2010.
3. Nidhal K. Al Abbadi et al., Psoriasis Detection using skin color and texture features,
Journal of Computer Science, Vol.6, Issue.6, pp. 648-652, 2010.
4. R.M. Haralick, Statistical and Structural Approaches to Texture, Proc. IEEE, vol. 67,
no. 5, pp. 786-804, 1979.
5. Ranjan Parekh, Using Texture Analysis for Medical Diagnosis, Media in medicine,
IEEE Computer Society, 2012.
6. Subhagata, A Comparative study of Fuzzy C-means algorithm and Entropy-based Fuzzy
clustering algorithms, Computing and Informatics, Vol. 30, pp. 701-720, 2011.