Binus University Repository

Soewito, M.Sc., Ph.D, Benfano

Attendance System on Android Smartphone
Benfano Soewito
Graduate Program
Bina Nusantara University
Jakarta, Indonesia
Email: benfano@gmail.com
Ford Lumban Gaol
Graduate Program
Bina Nusantara University
Jakarta, Indonesia
Echo Simanjuntak
Graduate Program
Bina Nusantara University
Jakarta, Indonesia
Fergyanto E. Gunawan
Graduate Program
Bina Nusantara University
Jakarta, Indonesia
Abstract—Attendance system which is currently excist still has
weaknesses. The first is the long queues in front of the attendace
machine at the time to come to work and leave work. The second
is cheating, employees can ask her/his friend to do attendance
process. The third is mostly attendance system has not been
connected with the payment system in human resouces software
or in the finance department. The fourth, employees who work
outside the office can not do attendance process. In this paper, we
introduced an attendance system based finger print technology
and GPS using a smartphone integrated with payment system
that will eliminate all the problems above. Our research also
based on prediction that in the next few year all the smart phone
will have a fingerprint scanner.
I. INTRODUCTION
Attendance system has been known since ancient time.
In the beginning, attendance system using only paper and
performed manually by calling the name of an existing name
listed and marked as present or not present along with the notes
[1]. This way is not efficient, it would be very time consuming
if we have to call one by one all the people who exist on the
list. In line with technogical developments in the field of clock
then created the attendance system used machine that more
efficient than manual attendance system. The first attendance
machine was very simple, the employees just inserting the
attendance paper or called timesheet into the machine, and
the time will be printed on the timesheet [2]. Basically, this
attendance machine consists of a manual card slot or hole to
insert the timesheet employees and an analog clock showing
the current time. The time when the employee insert the card
into the slot is what will be printed on the timesheet. This
attendance system has several disadvantages:
1) In this system, the data on timesheet has to process
manually. Human Resources Division should note
and record employee attendance manually that it can
be take a lot of time [3].
2) Inefficient due to spend a lot of paper. The use of this
manual attendance machine card is not environmen-
tally friendly and efficient because of the paper used
as employee timesheet [4].
3) Vulnerable fraud. Using this attendance system has a
vulnerable fraud because the fact is, many fraudulent
employee timesheet entrust them to his friend so that
when it comes late or absent, their cards will remain
in the list of attendance [5], [6], [7], [8].
4) Long queue. In the morning when employees will
come to work as well as at the time employees
leaving work will be long queue in front of the
attendance machine.
The second generation of attendance system is attendance
machine digital which is the attendance machine that uses a
digital method for the record employees attendance introduced
in 1970. In general, digital attendance machine has a few
buttons (number and alphabet) and a digital display that shows
the time or the text displayed [9]. Along with the machine,
there are number and alphabet keys to allow employees to
enter a password or barcodenya. Cheating employees are still
likely to occur in the use of digital attendance machine because
they may share the PIN with friends to do attendance. And also
still there is a long queue in front of the digital machine.
Lately, it appears attendance machine that using biometric.
The biometric attendance is machine attendance machine that
uses biometrics to authenticate employees when the doing
the attendance process [10], [11]. Biometrics is a special
technology that is commonly used in medical to identify the
man by detect certain characteristics of human beings. The
characteristics of which were identified using a biometric
system may be in the form of fingerprints, the shape and
characteristics of the face, the eyes, and the human voice. The
weakness of this system are:
1) Prices tend to be expensive. By using sophisticated
biometric technology, attendance machines tend to
be expensive because of the components / materials
specifically for detecting human facial characteristics.
2) Error at the time of identifying the employee. Em-
ployee identification using biometric systems tend
to encounter some problem or error. For example,
the facial characteristics are processed slowly and
can not even be identified by the machine when the
employee has been registered. Another example is the
characteristic sound, an employee who is impaired in
his voice sound like a cough or lost will be difficult
to be identified by the attendance machine.
In this paper, we introduced a attendance system using fin-
gerprint and GPS (Global Positioning System) on smartphone
or mobile devices. Fingerprint attendance machine is one type
of biometric attendance machine that uses fingerprint detection
methods through employee to record employee attendance
list. This type of fingerprint becoming known and used since
1997 [12]. Inventors and creators of technology realize that
attendance machine fingerprint is one part of the human body
is unique and different from each other. In fact, even identical
twins have different types and shapes of different fingerprints.
This is the trigger for the emergence of the idea of integrating
the fingerprint attendance machine.
The advantages of using fingerprint and GPS on smart-
phone for attendance system are:
1) Its use is practical and simple. Employees can in-
stantly do attendance process and prove its presence
by placing one finger or all fingers on the screen that
has been available on the attendance machine.
2) Attendance data collection more accurate. Because
the media used to record employee absences is their
fingerprints, then the accuracy of attendance will be
guaranteed. No more employees who like to cheat by
ask the friend to do attendance process like in the
attendance machine cards because they themselves
should be authenticated.
3) Capacity that many users / unlimited. Fingerprint
attendance machine can accommodate the data to
the tens of thousands and even infinite because there
should be no production of paper / card ID.
But attendance system using this fingerprint can not elimi-
nate the problem of long queues at entry work and also at the
time out of work. Therefore, we introduced a attendance sys-
tem based finger print technology and GPS using a smartphone
that will eliminate the problem of long queues. Our research
also based on prediction that in a few year all the smart phone
will have a fingerprint scanner.
II. LITERATURE REVIEW
Fingerprint usually in the form of horizontal lines and
vertical or a combination of both and there is also a form
of arches. All humans in the world created by fingerprints are
different from each other. Therefore, every fingerprint is used
to identify every human being. No fingerprints are identical in
this world even if in between two twin sisters [13].
In table I show the comparison of biometrics factor eval-
uation such as FAR (False Accept Rate), FRR (False Accept
Rate), and EER (Equal Error Rate). FAR is the percentage of
unauthorized users are authenticated as a legitimate user. The
lower the FAR, the better the accuracy of the biometric method
[8]. FAR is the percentage of unauthorized users identified as
legitimate users. The lower the FAR, the better the accuracy
of the biometric method. FAR can only be determined based
on data collected by a sample from another person who is not
an actual person. FAR also directly related to the threshold
used in biometric identification process, the lower the FAR, the
threshold used will be higher. The higher threshold probable
cause an unauthorized person to access the system becoming
less and less, but also will be has impact to user comfortable,
because the user who has the right to enter into the system
will be more difficult when they have to perform biometric
identification. The relationship between FAR, FRR, error and
sensitivity can be seen in figure 1.
FRR is the percentage of legitimate or authorized users
who rejected in the process of authentication [8]. FRR can be
determined from each user either authorized users accessing
the system or not by analyzing data collected from each user
in a certain time frame, could be in weeks, months or years.
There are a variety of reasons why the false rejection such
as biometric image quality is poor, the user is in a hurry to
authenticate so that the system is difficult to identify the user.
FRR is also dependent on the threshold value, where the higher
TABLE I. BIOMETRIC FACTOR EVALUATION
Biometric EER FAR FRR
Face NA 1% 10%
Fingerprint 2% 2% 2%
Hand Geometry 1% 2% 2%
Iris 0.01% 0.94% 0.99%
Keystrokes 1.8% 7% 0.1%
Voice 6% 2% 10%
the threshold value, then the FRR will also increase and vice
versa.
EER is the level where FAR and FRR have the same value.
The lower the EER, the more accurate the system. EER is
most appropriate description to describe the error rate of an
algorithm because, as described above, the lower the EER,
meaning the lower error that will occur on an algorithm.
Fig. 1. The relationship between FAR, FRR, error and sensitivity.
In this paper we selected the fingerprint method to inter-
grate in attendance system. This is because the fingerprint has
EER about 2% that better than voice. We did not select the Iris
method because of two reason. Firstly, in the next five year, in
each smart phone will be has the fingerprint scanner as well as
the camera. Secondly, the equipments to do attendance system
with iris technology can not be perform on smart phone or
other mobile devices.
III. SMARTPHONE ATTENDANCE SYSTEM
There are several phases in this research. The first phase
is to collect various issues contained in attendance system that
already exist today. Various problems include the limitations
of existing attendance system either manually or electronically.
After we collecting the problems and limitations, then we
design an attendance system that is accurate, efficient, and
dynamic. The second phase is to perform a variety of literature
related to attendance systems, biometric methods as user
identifier concerned, and mobile systems to be used as a
platform for the attendance system. The reference collected
by searching online or offline, online journals or obtained by
finding the appropriate studies related objects on the internet,
while the offline method gathered from various media such as
books. The third phase is preparations on a research tool, this
preparation includes checking the version and type of operating
system, check whether the device is being used to support
of the hardware fingerprint scanner. Checking of availablelity
fingerprint scanner is also done, it is useful to determine the
next step in which if the device already has a fingerprint
scanner then no mounting hardware fingerprint scanner again
and vice-versa. The fourth phase is learn the existing payroll
system today and its relationship with the attendance system,
this is done in order to know the extent of integration and
the effectiveness of the existing integration. After all the
observations and analyzes carried out, then will proceed with
the developing mobile applications based attendance system
using the android operating system. Then develop the payroll
application and integration of these systems with mobile atten-
dance system that has been created previously. The phase is
to perform simulations with predefined scenarios to determine
the success of the implementation of the attendance system
integrated with the payroll system. Based on the results of the
simulation will be made conclusions about the effectiveness of
the attendance system and accuracy of the data.
Fig. 2. Architecture of smartphone attendance system.
In figure 2 shows our proposed architecture system.
There are five componets in our attendance system: Smart
phone, Parser, Application Server, Administrator Computer
and Printer. Basically the research method followed the archi-
tecture system in 2. We begin developed application on android
smartphone using java. Then developed the application for
admin. The last step was developed parser that can connected
all the components in the system.
IV. RESULTS
In this research, we build a coding for attendance system
to be run on android smartphone, that will be integrated with
payment system and human resources application. We choose
android smartphone because the number of android user is very
huge in the world [14] and [15].
A. Attendance Application on Android Device
This application runs on the android device associated with
employee payroll application on the admin side through an
internet connection. This application will takes the informa-
tion about user, such as the user position through the GPS
coordinate that available on the android device [16] and [17],
the user’s fingerprint that will matched with a database on the
server, and retrieve the date and time when the user is doing
absent. With this method, it will improve the accuracy of the
attendance system because it takes a position, the date, and
time in real time. With this particular method, the employee
who works in project out of building also can do absent and
also there will not long queue anymore. The use of user
identification by fingerprint method will also reduce fraud
because the user can not ask friends to do attendance. There
are several menu on the main screen of applications, such as
About Application and Registration, which is classified as a
second menu bar menu. Then there are two menu again which
is a main menu: Attendance and Payroll & Schedule created
using the button.
Fig. 3. Finger print scanner on smartphone attendance system.
When users want to use this application, then user has
to register the device and user information to the employee
payroll application on the admin side, if the user has not
been registered, the registration menu can choose to display
the registration form. In order to send the information about
the finger print, user has to send the finger print through
the application on the android device as shown in figure 3.
Users must put his finger on a fingerprint sensor and press
a button that sends the fingerprint scan results to the server,
then server will check whether the device and fingerprint are
already registered previously or not. After registration process
done, then user can do attendance process by put the finger on
the fingerprint scanner.
B. Application for admin payroll
At this payroll application, the admin can take over atten-
dance management of employees. Admin can change or create
information about shift employees, add or edit employees,
and admin can also display data involving payroll timesheet
management and perform logtime to the employee through
this application. In figure 4 showed screenshot of timesheet at
the admin site. We develop many option or menu and create
database for admin such as employee personal information,
logtimes, shift schedule, group employees, overtime payroll,
and timesheet. We discuss all of them in journal separetely
with this paper.
Fig. 4. Screenshot of timesheet at the admin application.
V. CONCLUSION
Fingerprint as one way to do idenifikasi against someone
can be collaborated with GPS and mobile technology to create
an accurate attendance system applications. In this paper we
introduced the attendance system utilize fingerprint and GPS
on smartphone that can solve the problems in attendance
system. Our attendance system can avoid long queue in front
of attendance machine because the process of attendance in our
system used smartphone or other mobile devices. The problem
of cheating can be reduced because we used fingerprint tech-
nology. Our attendance system also was integrated with the
application of human resources, the human resources staff do
not need to calculate manually for overtime salary and payroll.
The last, the employees who work on project out of office can
do absent, because our system will collect the GPS coordinate
beside the date and time.
ACKNOWLEDGMENT
The authors would like to thank Directorate General of
Higher Education, Republic of Indonesia for their support in
making this research possible. This research was supported by
Competitive Research Grant scheme no. 112/K3/KM/2015.
REFERENCES
[1] Ashbourn, J. Biometrics: Advanced Identity Verification: The Complete
Guide. Springer-Verlag, London, Springer.2000
[2] Wayman, J.L. and L. Alyea. Picking the Best Biometric for Your
Applications. National Biometric Test Center Collected Works. National
Biometric Test Center: San Jose. p. 269-275, 2000
[3] Bredin, K. Human Resource Management in Project-Based Organi-
sations. 2008 - Challenges, Changes, and Capabilities. Doctoral the-
sis, Department of Management and Engineering, Linkping University,
Linkping.
[4] M.S Cardon and C.E. Stevens. Managing Human Resources in Small
Organizations: What do we know?. Human Resource Management
Review, vol 14 (2004), pp. 295-323.
[5] Prabhakar, S., S. Pankanti, and A.K. Jain. Biometrics Recognition:
Security and Privacy Concerns. IEEE Security & Privacy. 2003.1(2): p.
33-42.
[6] Lisa Myers, An Exploration of Voice Biometrics. SANS Institute InfoSec
Reading Room. 2014
[7] A. Tiwana. Web Security. Digital Press An imprint of Butterworth-
Heinemann. 1999.
[8] C.P. Pfleeger. Security in Computing. Second edition. Prentice Hall PTR.
1997.
[9] Zhang, D. Automated Biometrics: Technologies and Systems. Norwell,
MA: Kluwer Academic Publishers. 2000.
[10] Allan, A. Biometric Authentication. Perspective. Gartner Research,
2002, p. 1-31.
[11] Harris, A.J. and D.C. YEN. Biometric authentication: assuring access
to information. Information Management andComputer Security, 2002.10
(1): p. 12-19.
[12] Dugelay, J.L. Recent Advantages in Biometric Person Authentication.
ICASSP International Conference on Acoustics, Speech and Signal
Processing, Orlando, Florida, USA, 2002.
[13] Woodward, J.D. Army Biometric Applications: Identifying and Ad-
dressing Sociocultural Concerns. 2001.
[14] Chris Velazco. 2011. Android Still Most Popular Smartphone OS,
iOS Holds Steady In Second Place. (on-line), Techcrunch web.
http://techcrunch.com/2011/11/03/android-still-most-popular-os-ios-
holds-steady-in-second-place/. 13 Januari 2013.
[15] Tyler Lee. 2011. AuthenTec releases SDK for fingerprint sensor
applications on Android phones. (on-line), Ubergizmo web.
http://www.ubergizmo.com/2011/08/authentec-releases-sdk-for-
fingerprint-sensor-applications-on-android-phones/. 13 Januari 2013.
[16] Darcey Lauren and Conder Shane. 2011. Enable a Fingerprint Scanner
for Android on the Motorola ATRIX 4G. (on-line), developer.com
web. http://www.developer.com/ws/android/devices/enable-a-fingerprint-
scanner-for-android-on-the-motorola-atrix-4g.html. 13 Januari 2013.
[17] Sandi Fajar Rodiyansyah. 2011. Ar-
sitektur Sistem Operasi Android.(online),
http://educnology.web.id/opensource/rodiyansyah/arsitektur-sistem-
operasi-android, 14 Januari 2013.


English

Attendance system on Android smartphone

Suharjito, Suharjito

Adaptive Neuro-Fuzzy Inference System
for Texture Image Classiﬁcation
B. Ari Kuncoro
Master of Information Technology
Bina Nusantara University
Jakarta, Indonesia
Email: b.kuncoro [at] binus.ac.id
Suharjito
Master of Information Technology
Bina Nusantara University
Jakarta, Indonesia
Email: suharjito [at] binus.ac.id
Abstract—One of the most important problems in pattern
recognition is texture-based image classiﬁcation. In this pa-
per, the combination of Discrete Cosine Transform (DCT) and
Gray Level Co-Occurrence Matrix (GLCM) methods for feature
extraction was proposed. The attributes extracted from DCT
method were mean and variance, while the attributes extracted
from GLCM method were energy and entropy. Adaptive Neuro-
Fuzzy Inference System (ANFIS) was used as a classiﬁer. The
classiﬁer model was trained using 50% of the texture images
and remaining images were used for testing. Four classes of
texture images were downloaded from KTH-TIPS (Textures
under varying Illumination, Pose and Scale) image database,
three of which were used in each experiments thus there were four
data combination. The best data testing accuracy result towards
textures of crumpled aluminium foil, corduroy, and orange peel is
98.3%, which is 1.6% better than one hidden-layer feed forward
neural network classiﬁer. In average, testing accuracy result of
ANFIS excelled one hidden-layer feed forward neural network
with 93.7% over 90.4%.
Index Terms—Texture Images; Discrete Cosine Transform,
Gray Level Co-Occurence Matrix, Adaptive Neuro-Fuzzy Infer-
ence System
I. INTRODUCTION
Texture has been studied for a long time as a primitive
visual cue and is known as low-level image features [1]. It is
useful in recognition and interpretation, because as image, it
consists of related and interrelated elements [2]. To extract the
features from the texture images, several steps must be applied.
Many researchers reported various methods in how to extract
features based on textures. They include statistical methods
of GLCM, energy ﬁlters and edgeness factor, Gabor ﬁlters,
and Wavelet transform [3], [4]. There is no strict guideline to
determine what features are most suitable for speciﬁc data,
thus experimental methods such as empirical and heuristic
methods are commonly used.
Upon extracting features, the classiﬁcation is applied. There
were various techniques reported and developed for classi-
ﬁcation by many researchers such as feed-forward neural
networks, extreme learning machine, and ANFIS. The ap-
Fig. 1. Samples of Dataset
plications of this development are also various such as the
classiﬁcation of fruits (e.g. oranges) and rocks [5]. Those
applications are useful to help industries expedite classiﬁcation
process.
In this study, the used features extraction methods are based
on DCT and GLCM. In DCT, statistical components of mean
and variance of DCT matrix were used, while in GLCM, the
energy and entropy of GLCM matrix were used. The classiﬁer
used in this study is ANFIS. To classify images based on
texture and to know the accuracy of the model is the objective
of this work.
II. DATASET
The image dataset can be downloaded at
http://www.nada.kth.se/cvap/databases/kth-tips/download.html
[6]. Four-class texture image samples consist of alumunium
foil, corduroy, orange peel, and cotton are illustrated in Fig.
1. Each class contains 80 texture images, thus there were 320
2015 International Conference on Automation, Cognitive Science, Optics, Micro Electro-Mechanical System, and Information Technology
(ICACOMIT), Bandung, Indonesia, October 29–30, 2015
978-1-4673-7408-8/15/$31.00©2015 IEEE  
196
texture images collected. Using combinational method, each
of experiments used three out of four classes or 240 images.
III. METHODOLOGY
The methodology of this study is illustrated in Fig. 2. First,
320 gray-scale images were collected in one folder. Then the
images were resized into 64×64 pixels. On each images, DCT
and GLCM blocks were applied thus DCT and GLCM matrix
were produced. From DCT matrix, mean and variance were
calculated as attribute 1 and attribute 2, while from GLCM,
energy and entropy were extracted as attribute 3 and attribute
4. Upon extracting, 3 out of 4 classes were combined into
four different combination of dataset. The ﬁrst experiment
utilized 80 aluminium foil, 80 corduroy, and 80 orange peel
texture. The second until the forth experiment involved cotton
texture using combinational method. As a result there were
four experiments in total. The combination of experiments
based on the data involved is written as follows.
• Combination 1: aluminium foil, corduroy, and orange
peel
• Combination 2: aluminium foil, corduroy, and cotton
• Combination 3: aluminium foil, cotton, and orange peel
• Combination 4: corduroy, cotton, and orange peel
Each combination of dataset was divided equally into 50%
data training and 50% data testing. Data training was used for
ANFIS training and to build the model of classiﬁer, while
data testing was used for testing the model. The last step
was computing the confusion matrix and accuracy. Detail
description is written in the following subsection.
A. Discrete Cosine Transform (DCT)
The ﬁrst step in DCT is usually to convert an RGB image
into a gray level image. Since in this dataset each image has
been in gray level image, no conversion is needed. The next
step is resizing the images into N ×N sub-blocks where the
size of N depends on size of image. The 2D DCT can be
written in terms of pixel values [2] [7]. The resized image
matrix is represented by f(i, j) for i, j = 0, 1, ..., N − 1 and
the frequency-domain transform coefﬁcients can be obtained
from the general equation N × M below. To simplify the
computation, square-size images were used. Thus in this case,
M equals to N.
F (u, v) =
(
2
N
) 1
2
(
2
M
) 1
2
N−1∑
i=0
M−1∑
j=0
Λ(i)Λ(j)
cos
[ πu
2N
(2i+ 1)
]
cos
[ πv
2M
(2j + 1)
]
f(i, j)
where Λ(i) =
{ 1√
2
for ε = 0
1 otherwise
(1)
Fig. 2. Flowchart of Image Classiﬁcation using ANFIS
The size of matrix F (u, v) is similar to f(i, j). If the image
size is 64×64, then the matrix size of DCT is also 64×64. In
DCT matrix, the upper left corner of the matrix represents the
low frequency part [8]. As the used image size in this study
is 64×64 and the used DCT matrix part is the low frequency,
thus the ﬁrst 8×8 components was collected to calculate mean
and variance. Those parameters were used as as attribute 1 and
attribute 2 respectively.
B. Gray Level Co-Occurrence Matrix (GCLM)
Gray Level Co-Occurrence Matrix (GLCM) has proved to
be a popular statistical method of extracting textural feature
from images. According to co-occurrence matrix, Haralick
deﬁnes fourteen textural features measured from the probabil-
ity matrix to extract the characteristics of texture statistics of
remote sensing images [9]. Mohanaiah [4] performed extrac-
tion using these important features, Angular Second Moment
(energy), Correlation, Entropy, and the Inverse Difference Mo-
ment. In this work, feature of Entropy and Energy from GLCM
Matrix were used as attribute 3 and attribute 4 respectively.
 
197
1) Angular Second Moment (Energy): Angular Second
Moment or Energy is the sum of squares of entries in the
GLCM matrix. It is also known as homogeneity. The value of
which energy high when image has very good homogeneity
or when pixels are very similar [4].
Energy =
Ng−1∑
i=0
Ng−1∑
j=0
P (i, j)2 (2)
2) Entropy: Entropy shows the amount of information of
the image that is needed for the image compression. Entropy
measures the loss of information or message in a transmitted
signal and also measures the image information.
Entropy =
Ng−1∑
i=0
Ng−1∑
j=0
−P (i, j)× log(P (i, j)) (3)
In this paper, each class was represented by integer. For
combination 1, integer 1, 2, and 3 represent aluminium foil,
corduroy, and orange peel respectively. Since the feature
extraction process produced four features, each of the data
contains four attributes. As a result, the used data contains
four inputs and one output. The snippet of the data is as
follows.
880.729,4.68593e+06,2.98135,0.0891916,1
811.697,1.73683e+06,2.72428,0.13527,1
732.743,2.2623e+06,2.62406,0.119913,1
418.068,509726,2.0458,0.348333,1
424.64,689089,2.30269,0.239463,1
408.461,800735,2.57124,0.158994,1
...
246.253,2.61788e+06,1.78639,0.261709,2
248.174,2.52921e+06,2.14585,0.183963,2
281.978,3.66744e+06,2.4358,0.110385,2
284.576,3.38465e+06,1.57276,0.260701,2
277.482,2.82053e+06,1.78907,0.234655,2
323.808,3.50561e+06,1.97454,0.183467,2
282.32,3.3365e+06,1.54988,0.3172,2
...
399.822,6.94773e+06,1.48722,0.302955,3
378.079,6.5303e+06,1.51542,0.26855,3
370.851,4.96315e+06,1.5838,0.322928,3
368.65,6.50668e+06,1.31369,0.321808,3
354.542,5.24536e+06,0.844406,0.66578,3
374.87,5.52941e+06,1.35618,0.422251,3
398.339,6.67421e+06,1.43632,0.30473
...
C. Adaptive Neuro-Fuzzy Inference System
The deﬁnition of ANFIS is derived from the combination of
Artiﬁcial Neural Network and Fuzzy Inference System (FIS) in
Fig. 3. ANFIS Model Structure
Fig. 4. Membership Function of FCM Generated FIS
such a way that neural network learning algorithms are used
to determine the parameters of FIS. This system takes the
beneﬁts of both Artiﬁcial Neural network system and Fuzzy
Logic system and eliminates their limitations [10]. An even
more important aspect is that the system should always be
interpretable in terms of fuzzy if-then rules, because it is based
on the fuzzy system reﬂecting vague knowledge [7]. In this
study, the structure of ANFIS is illustrated in Fig. 3. It consists
of 4 inputs, 3 membership functions, 3 rules, and 1 output.
Using training data containing 4 inputs and 1 output, ANFIS
constructs a FIS whose membership function parameters are
tuned (adjusted) using Fuzzy C-Mean (FCM) clustering and
Sugeno structure type. The membership functions were built
using Gaussian as illustrated in Fig. 4.
The three rules used are written as follows:
Rule 1: If (In1 is In1cluster1) and (In2 is In2cluster1) and
(In3 is In3cluster1) and (In4 is In4cluster1) then (out1 is
out1cluster1)
Rule 2: If (In1 is In1cluster2) and (In2 is In2cluster2)
and (In3 is In3cluster2) and (In4 is In4cluster2) then (out1 is
out1cluster2)
Rule 3: If (In1 is In1cluster3) and (In2 is In2cluster3)
 
198
Fig. 5. Plot Input data in 2-D
and (In3 is In3cluster3) and (In4 is In4cluster3) then (out1 is
out1cluster3)
D. Artiﬁcial Neural Network
In order to evaluate the performance of ANFIS, ANFIS
parent’s method called Artiﬁcial Neural Network (ANN) with
feed forward network type was used on the same data. The
structure used was one hidden-layer. The similar strategy in
how to divide the data training and data testing was also
applied.
IV. RESULTS AND DISCUSSION
To visualize the data based on generated attributes, 2-D
plotting data graph is required. It is depicted in Fig. 5 that color
of blue, green, and red represent aluminium foil, corduroy, and
orange peel textures respectively. Based on that ﬁgure, it is
seen that aluminium foil texture is well grouped compare to
the other two textures. Despite that it is still a good signal for
classiﬁcation purposes.
The data performance graph of combination 1 can be seen
in Fig. 6. It is quite impressive that from epoch 1 to 20,
root mean square error (RMSE) decreased signiﬁcantly from
0.07 to 0.023. Its value is unchanged when epoch reached 22.
Another performance indicated by confusion matrix is shown
in Fig. 7. This confusion matrix was built based on data testing.
The parameter of accuracy is used to measure the effectiveness
of the system which can be derived from equation 4 or simply
collect the value of confusion matrix with blue background
color located in the most lower right position. Figure 7 shows
that the accuracy of ANFIS for combination 1 data is 98.3%.
Accuracy =
TP + TN
TP + FN + TN + FP
× 100% (4)
Fig. 6. Epoch vs RMS Graph of Combination 1 ANFIS
Fig. 7. Confusion matrix against data testing of Combination 1
Where,
• True Positive (TP) represents the number of positive data
that is correctly labeled by a classiﬁer
• True Negative (TN) represents the number of negative
data that is correctly labeled by a classiﬁer
• False Positive (FP) represents the number of positive data
that is incorrectly labeled by a classiﬁer
• False Negative (FN) represents the number of negative
data that is incorrectly labeled by a classiﬁer
The performance result of ANFIS and one hidden-layer
ANN can be seen in Table I. It presents the accuracy based
on both of data training and data testing. For data training and
combination 1 data, ANFIS accuracy result is 99.2% while one
hidden-layer ANN accuracy result is 98.3%. For data testing
 
199
TABLE I
PERFORMANCE MEASUREMENT RESULT OF ANFIS AND ONE
HIDDEN-LAYER ANN
Accuracy Data ANFIS One HL ANN
Data Training
Combination 1 99.2% 98.3%
Combination 2 94.1% 91.7%
Combination 3 97.5% 92.5
Combination 4 90.8% 87.5%
Average 95.4% 92.5%
Data Testing
Combination 1 98.3% 96.7%
Combination 2 94.1% 89.2%
Combination 3 95% 88.2%
Combination 4 87.5% 87.5%
Average 93.7% 90.4%
and combination 1 data, accuracy against data testing is 98.3%.
It is 1.6% better than the accuracy against data testing based
on one hidden-layer ANN classiﬁer. In average of combination
1, 2, 3, and 4 data, ANFIS surpassed one hidden-layer ANN in
term of accuracy which is 95.4% over 92.5% for data training
and 93.7% over 90.4% for data testing.
V. CONCLUSION
The combination of DCT and GLCM methods is really
powerful to be used in features extraction of texture images.
Mean and variance of DCT, entropy and energy of GLCM are
the features that were extracted in this work. It was proven that
ANFIS works well to classify the texture images. ANFIS can
classify aluminium foil, corduroy, and orange peel textures
data with 98.3% data testing accuracy with only 22 epoch.
Overall, ANFIS performance surpasses the one hidden-layer
ANN performance in term of accuracy with 95.4% over 92.5%
data training accuracy and 93.7% over 90.4% data testing
accuracy.
ACKNOWLEDGMENT
The author would like to thank Dr. Diaz D. Santika of Bina
Nusantara University, Jakarta, Indonesia for his guidance and
Dr. Bayram Cetili Suleyman of Demirel University, Computer
Engineeering, Isparta, Turkey for his work in ANFIS MAT-
LAB program that was used and modiﬁed in this study.
REFERENCES
[1] D. Avci, M. Poyraz, and M. Leblebicioglu, “An expert system based
on discrete wavelet transform - anﬁs for acquisition and recognition
of invariant features from texture images,” in Signal Processing and
Communications Applications Conference (SIU), 2015 23th, May 2015,
pp. 1070–1073.
[2] P. H. Chandankhede, “Soft Computing Based Texture Classiﬁcation
with MATLAB Tool,” International Journal of Soft Computing and
Engineering, no. 2, pp. 475–480, 2012.
[3] L. A. Ruiz, A. Fdez-Sarrı´a, and J. A. Recio, “Texture Feature Extraction
for Classiﬁcation of Remote Sensing Data using Wavelet Decomposi-
tion: A Comparative Study,” in International Archives of Photogramme-
try and Remote Sensing, vol. XXXV, 2004, pp. 1682–1750.
[4] P. Mohanaiah, P. Sathyanarayana, and L. Gurukumar, “Image Texture
Feature Extraction Using GLCM Approach,” International Journal of
Scientiﬁc & Research Publication, vol. 3, no. 5, pp. 1–5, 2013.
[5] C.-H. Wu, J.-S. Lee, C.-Y. Shie, and M.-Y. Su, “Rock classiﬁcation
via a mobile device,” in Communications and Networking in China
(CHINACOM), 2012 7th International ICST Conference on, Aug 2012,
pp. 663–666.
[6] B. C. Mario Fritz, Eric Hayman and J.-O. Eklundh, “The KTH-
TIPS and KTH-TIPS2 Image Databases,” 2004. [Online]. Available:
http://www.nada.kth.se/cvap/databases/kth-tips/documentation.html
[7] G. Sorwar and A. Abraham, “DCT based texture classiﬁcation using
soft computing approach,” CoRR, vol. cs.AI/0405013, 2004. [Online].
Available: http://arxiv.org/abs/cs.AI/0405013
[8] W. Zhen and X. Zhe, “SIFT Feature Extraction Algorithm for Image in
DCT Domain,” Proceedings of the 2nd ISCCCA-13, pp. 261–264, 2013.
[9] P. Costianes and J. Plock, “Gray-level co-occurrence matrices as features
in edge enhanced images,” in Applied Imagery Pattern Recognition
Workshop (AIPR), 2010 IEEE 39th, Oct 2010, pp. 1–6.
[10] S. Goswami and L. Bhaiya, “A hybrid neuro-fuzzy approach for brain
abnormality detection using glcm based feature extraction,” in Emerg-
ing Trends in Communication, Control, Signal Processing Computing
Applications (C2SPCA), 2013 International Conference on, Oct 2013,
pp. 1–7.
 
200


Adaptive neuro-fuzzy inference system for texture image classification

Sasmoko, Sasmoko

Construct Validity in NeuroResearch

Girsang, Abba Suganda

Form Reviewer "Repairing the Inconsistent Fuzzy Preference Matrix Using Multiobjective PSO"

CANDRA, SEVENPRI

Form Reviewer "The use of IPA in bricks and clicks service quality (lesson learned from ICT shop)"

Thoha, Nurianna

Reviewer Measuring the Effect of Equality Matching Relationships on Lecturers’ Knowledge Sharing Behavior

Anggraeni, Adilla

Form Reviewer "Drivers of Brand Trust in Internet Retailing: The Case of Indonesia"

Form Reviewer "Intention to Use Self-Service Technology: Blitzmegaplex Jakarta"

Furinto, Asnan

Reviewer "Does Internationalisation Improve Organisational Performance? A Study on Higher Education Institutions in Indonesia"

Reviewer "The Influence Factors of the Innovation Adoption Behavior of Customer Website (case study: Student State University)"

Binus University

Not a member yet