Computers and Softwareeprints.binus.ac.id/33649/1/Hand Motion Gesture for... · Hand Motion Gesture for Human-Computer Interaction Using Support Vector Machine and Hidden Markov Model

International Review on Computers and Software

(IRECOS) Contents: Hand Motion Gesture for Human-Computer Interaction Using Support Vector Machine and Hidden Markov Model by Suharjito, Fitra B. Adinugraha

374

Fuzzy Expert System for Classifying Pests and Diseases of Paddy Using Bee Colony Algorithms by Yovita Tunardi, Suharjito

381

A Framework for Social Media and Text-Based Content Analysis for Event Management Purposes by Qusai Abuein, Mohammed Q. Shatnawi, Muneer Bani Yassein, Radwan Batiha

388

Optimized Implementation of H.264/AVC Motion Estimation on a Mixed Architecture Using SynDEx-Mix by Oussama Feki, Thierry Grandpierre, Nouri Masmoudi, Mohamed Akil

395

Integration Electronic Patients’ Records with Open Life Sciences Datasets Using Semantic Web Tools by Bassam Najeeb, Bassel Al Khatib

403

Enhancing Relay Selection Scheme for Connecting VANETs to Internet Over IEEE 802.11p in Congested and Fading Environment Scenarios by D. Abada, A. Massaq, A. Boulouz

410

Key Frames Based Video Authentication Using Fragile Watermarking and Singular Value Decomposition by Assma Azeroual, Karim Afdel

420

Localization of Mobile Stations from ONE Base Station in GSM Systems by Khalid G. Samarah

427

Word Extraction from Arabic Handwritten Documents Based on Statistical Measures by Ayman Al-Dmour, Raed Abu Zitar

436

A New Compression Scheme Based on Adaptive Vector Quantization and Singular Value Decomposition by Imene Soussi, Mohamed Ouslim

445

Real-Time Electroencephalography-Based Emotion Recognition System by Riyanarto Sarno, Muhammad Nadzeri Munawar, Brilian T. Nugraha

456

ISSN 1828-6003Vol. 11 N. 5

May 2016

REPRINT

International Review on Computers and Software (IRECOS)

Editor-in-Chief: Prof. Marios Angelides

Brunel University London Department of Electronic and Computer Engineering Uxbridge, Middlesex - UB8 3PH, U.K.

Co-Editor: Dr. Harry Agius Brunel University London Department of Electronic and Computer Engineering Uxbridge, Middlesex - UB8 3PH, U.K.

Editorial Board:

Mikio Aoyama (Japan) Pascal Lorenz (France) Francoise Balmas (France) Marlin H. Mickle (U.S.A.) Vijay Bhatkar (India) Ali Movaghar (Iran) Arndt Bode (Germany) Dimitris Nikolos (Greece) Rajkumar Buyya (Australia) Mohamed Ould-Khaoua (U.K.) Wojciech Cellary (Poland) Witold Pedrycz (Canada) Bernard Courtois (France) Dana Petcu (Romania) Andre Ponce de Carvalho (Brazil) Erich Schikuta (Austria) David Dagan Feng (Australia) Arun K. Somani (U.S.A.) Peng Gong (U.S.A.) Miroslav Švéda (Czech) Defa Hu (China) Daniel Thalmann (Switzerland) Michael N. Huhns (U.S.A.) Luis Javier García Villalba (Spain) Ismail Khalil (Austria) Brijesh Verma (Australia) Catalina M. Lladó (Spain) Lipo Wang (Singapore)

The International Review on Computers and Software (IRECOS) is a publication of the Praise Worthy Prize S.r.l.. The Review is published monthly, appearing on the last day of every month. Published and Printed in Italy by Praise Worthy Prize S.r.l., Naples, May 31, 2016. Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved. This journal and the individual contributions contained in it are protected under copyright by Praise Worthy Prize S.r.l. and the following terms and conditions apply to their use: Single photocopies of single articles may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale and all forms of document delivery. Permission may be sought directly from Praise Worthy Prize S.r.l. at the e-mail address: [email protected] Permission of the Publisher is required to store or use electronically any material contained in this journal, including any article or part of an article. Except as outlined above, no part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. E-mail address permission request: [email protected] Responsibility for the contents rests upon the authors and not upon the Praise Worthy Prize S.r.l.. Statement and opinions expressed in the articles and communications are those of the individual contributors and not the statements and opinions of Praise Worthy Prize S.r.l.. Praise Worthy Prize S.r.l. assumes no responsibility or liability for any damage or injury to persons or property arising out of the use of any materials, instructions, methods or ideas contained herein. Praise Worthy Prize S.r.l. expressly disclaims any implied warranties of merchantability or fitness for a particular purpose. If expert assistance is required, the service of a competent professional person should be sought.

REPRINT

International Review on Computers and Software (I.RE.CO.S.), Vol. 11, N. 5

ISSN 1828-6003 May 2016

Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved DOI: 10.15866/irecos.v11i5.8641

374

Hand Motion Gesture for Human-Computer Interaction

Using Support Vector Machine and Hidden Markov Model

Suharjito, Fitra B. Adinugraha

Abstract – Hand gesture recognition for human-computer interaction has become very popular

in recent times. The main problem of this technology is how the system can recognize the presence

of a gesture in a streaming video. In this paper, we propose a model that can recognize hand

motion gesture in avideo stream using Support Vector Machine and Hidden-Markov model.

Support Vector Machine has the advantage of generalizing classification. On the other hand,

Hidden Markov Model is a statistical model that is capable of modeling Spatial-temporal time

series. This system is divided into two main processes. First, this system recognizes hand posture

using SMV (static gesture recognition) and generates sequence observation which is used for the

second process later. The second process is recognizing dynamic hand gesture with the sequence

observation from the static gesture. The implementation shows that static hand gesture recognition

achieves average accuracy at 91% using testing dataset. Meanwhile,dynamically isolated hand

gesture recognition gets average accuracy at 89% using testing dataset. We also have tested the

system with continuous dynamic gesture using video stream. The system can recognize the gesture

very well with accuracy of 83%. This achievement shows that the model can be used in human

computer-interaction with specific supports such as Vector Machine and Hidden Markov Model

parameter. Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Gesture Recognition, Hand Gesture Recognition, Human Computer Interaction

I. Introduction

In daily life, gesture becomes one of main elements of

non-verbal communication. We use gesture to convey

some meaningful information. Gesture recognition is the

process in which gestures made by a user is made known

to the system. On the other hand, human-computer

interaction using motion gesture provides something

more natural, innovative and user-friendly [1]. Gesture

recognition has a wide range of applications, such as

virtual reality, augmented reality, sign language

recognition, video games, and medical application [2].

Recognition using the hand can be done using the

glove-based technique, device accelerator-based [3] and

visual-based technique [2]. While glove-based and

accelerator device sensing are very effective tools for

capturing hand motion, they are very expensive,

unnatural, and difficult to calibrate and setup the

procedure. On the other hand, the vision-based technique

provides natural gesture. It is inexpensive but still has

many challenges to improve the accuracy and speed [2].

One of the problems to achieve accuracy is how we

can extract the right feature that represents the hand

shape. Hand gesture recognition can be classified into

two types, static and dynamic. The static gesture is a

configuration or pose of the hand represented by a single

image while the dynamic gesture is represented by more

than one image [2]. In this paper, we focus on problems

regarding both of them.

After we have acquired static hand gesture, we use the

result to create a feature for the dynamic hand gesture. In

the static hand gesture, we use Support Vector Machine

(SVM) to classify single hand pose image with another

pose. Then, Hidden Markov Model (HMM) is used to

model dynamic hand gesture from the sequence of pose

hand image.

II. Related Work

In the last decades, there has been some research that

performs hand gesture recognition. Some of them

employ device based recognition like glove-based

gesture recognition [4], This glove-based gesture

recognition is very efficient in recognizing gesture

because the glove contains embedded sensors that can

detect motion very well. However, this method is

uncomfortable for the hand because this glove must be

attached to the wire. The other method using device-

based is MEMS accelerator recognition [3]. In this

method, the author tries to recognize motion gesture

through a device that can detect acceleration of hand

motion. Similar tothe glove-based gesture recognition,

this method is also efficient. However, we cannot

recognize the shape of the hand.

SVM is apopular model used for classification and

regression. In gesture recognition, SVM is commonly

paired with another feature extraction or model like HU

REPRINT


Copyright © 2016 Praise Worthy Prize S.r.l. - All rights reserved International Review on Computers and Software, Vol. 11, N. 5

375

moments feature extraction. It has been used to classify

hand shape for static hand gesture [5].This pairing

method has an advantage of extracting moment of the

hand efficiently, such as hand orientation, angle, center

moment. The other method that has been paired with

SVM is Fourier descriptor used by Gamal et al. [2].

This method has some advantages, including HU

moments with additional fast processing because Fourier

descriptor only uses a limited number of feature for each

image. SVM is also often used to be combined with

another classification model as proposed by Demidova.

He combines SVM and fuzzy clustering algorithms [6].

Then, AbAzziz, who combines Fast Artificial Immune

and SVM for maximum load margin improvement [7].

The other method like HMM has been used to develop

Arabic continuous hand gesture from color image

sequence by applying trajectory of a single hand [8].

Malgireddy et al. [9] develop dynamic hand gesture.

They suggest a model explicitly (gesture grammar) and

use it to learn a model for each gesture. This method

divides gesture to sub-gestures and then uses the sub-

gestures to recognize the start point and end point of

gesture. With this method, Malgireddy uses HMM that

has been modified to use multiclass model and compare

HCRF. Hsiang-Yueh [10] and Hui-Shyong Yeo [11]

suggestthe use of YCrCb color space and convexity

defect character point to extract fingertip position and

perform robust dynamic hand gesture recognition with

simple algorithms. SVM HMM is also commonly used to

be combined with other classification models, like

proposed by Regan and Srivatsa.

They present a hybrid Canny Edge Detection and

Weibull Probability Density Function based on

Hierarchical Dirichlet Process HMM [12]. In this paper,

we focus on combining the base on these effective

methods. First, we suggest combining these advantages

from the previous research, such as using sub-gesture

from Malgireddy [9] with little modification.

We only use a single image to detect static gesture and

then perform the static gesture usingSVM. Second, the

used method is performing extract feature using

convexity defect method from Hsiang-Yueh to get robust

finger detection. Then, we also use HMM to recognize

continuous dynamics hand motion gesture. This process

will be explained in the methodology.

III. Methodology

In this section, we will present the proposed model of

human-computer interaction using hand motion gesture

recognition. The system is divided into four processes.

There are preprocessing, feature extraction, hand static

gesture and hand dynamic gesture recognition. Fig. 2

shows the proposed model.

III.1. Preprocessing

The purpose of this process is to prepare images for

the next phase.

We segment image color using YCrCb color space

information first and select the appropriate threshold

algorithm with additional threshold configuration to be

calibrated in the current environment. The default

threshold in the Eq. (1) [11]. Later, we will remove the

noise using dilate and mean filtering to smooth the

images. This process left the image of binary color:

77 ≤ Cb ≤ 127 and 133 ≤ Cr ≤ 173

where Y Cb Cr,, [0,255]

(1)

(a)

(c)

(b)

(a) (b)

(a)

(c)

(b)

(c)

Figs. 1. Hand Segmentation,(a) original image, (b) filter color image

(c) smooth image

The next step iscalculating contour of the hand

forfurther use in feature extraction.

Capture

Video

Image Filtering(YCrCb, MEDIAN,

COUNTOUR)

Feature Extraction(CONVEX HULL, CONVEXITY

DEFECTS, FINGERTIP)

Hand Classification(SVM)

Motion Classification(HMM)

Module Training

Motion Classification(HMM)

Module Training

Hand Classification(SVM)

GesturePrediction

(Left Click)

Models

Topology HMM

(A, B, )

Models SVM

(alpha, bias, SV, Kernel)

Class Label

(1-5-5-2-1-1-2)

Feature Vector1:50 2:32 3:40 ...

1:52 2:31 3:42 ...

Filterd ImageImage Frame

Fig. 2. Proposed model of hand motion gesture recognition

III.2. Feature Extraction Using Convexity Defect

Our proposed feature extraction method evolved from

previous research regarding hand tracking and gesture

recognition system [10], [11]. The result of preprocessing

is the contour of the hand. The contour is a list of the

point that represents a curve in an image.

REPRINT



376

First, we calculate polygon represented the

approximate to build contour [11]. Second, it is to extract

interior of the hand contour to gain more information

about hand shape image. This process can be done by

calculating palm center, and extracting convex hull and

convexity defect. Palm Center is an area determined as

the maximum inscribed circle inside the contour.

We calculate the maximum inscribed circle using

Voronoi diagram [13]. The maximum point of inscribed

circle lies on the point generated from the Voronoi

diagram. Next, the convexity defect will be illustrated in

Fig. 3. In this paper, we only use several convexity defect

points that match the following criteria:

- Depth point (Pd)and start point (Ps) must be above

the center of maximum inscribed circle (Ca)

- Each start point (Pa) and end point (Pe) must be

above Depth Point (Pd)

- Depth of each defect (ld) must be longer than palm

center radius (ra)

Fig. 3. Convex hull and convexity defect extracted from [11] hand

tracking and recognition system for human-computer interaction using

low-cost hardware

The next step is calculating or extracting the feature

from convexity defect that matches the criteria. In this

paper, we use two feature vectors.

The first feature is taken from the angle between depth

point (Pd) and end point (Pe) of each convexity defects

(θa), whilethe second one is taken from the depth

distance (distance between depth point and end point).

III.3. Static Gesture Recognition Using SVM

Support vector machine is a supervised classification

model that is often used to classify and regress. The basic

algorithm SVM is to separate the feature of training data

with a line or field separator, which is called hyperplane.

The function of the hyperplane is to maximize the

distance between the two pieces of data [2].

Basically, SVM is used to classify the two pieces of

data or classes. However, the application of SVM applied

is to classify more than two classes using method 1-vs-1

and 1-vs-all strategy 1-vs-1 by comparing one class with

another class so the class that has the highest value will

become the solution.

This method has andisadvantage. When the class that

will be classified multiplies, the time to classify will

increase, that can affect the performance of the

application. Moreover, the strategy of one-vs-all SVM

does createa model in which the model will detrain with

all class of existing dataset so that the decision will

depend on the model [14].The other ability of the SVM is

providing a transformation dataset from dimensional data

for other higher dimensional data defined by kernel

function [6].

Now, we have a set of feature hand shape images.

Then, the next step is to run the SVM classifier model

and get the best parameter to classify the images. SVM is

a supervised machine learning method used for

classification and regression.

The SVM algorithm separates the training data in

feature space by a hyperplane, which maximizes the

margin between two data. Hence, it is also known as

maximum margin classifiers [2]. SVM is based on the

principle of structural risk minimization (SRM). The

SRM induction principle has two main objectives. First,

it is to control the empirical risk on the training data, and

the second is to control the capacity of the decision

functions used to obtain this risk value [14].

There is an immediate problem arising from the

SVM’s original hyperplane formulation. It is not very

obvious how to make the model applicable to more than

two classes. Several approaches have been proposed to

overcome this limitation. Two of them are known as the

1-vs-1 and 1-vs-all strategies for multiple class

classification. For the decision problem over classes, 1-

vs-all requires the creation of classifiers.

Each is trained to distinguish one class from the

others. The decision is taken with a winner-takes-all

approach. However, there is no clear indication that this

approach results in an excellent decision. In the 1-vs-1

strategy, the problem is divided into c (c – 1) / 2 sub-

problems considering there are only two classes at a time.

At the decision phase, each machine casts a vote for

one of the classes and the label with the highest number

of votes wins.

This leaves the problem of evaluating an increased

number of machines every time a new instance is

classified – which can become troublesome or

prohibitive in time sensitive applications easily [15].

In this paper, we train and test four kernel types: RBF,

Linear, Polynomial, and sigmoid. Then, we propose ten

poses of static hand gesture as illustrated in Fig. 4. We

use multiclass classifier of SVM to classify ten classes of

static hand gesture. Fig. 4 shows static hand gesture

representing sequence observation label. This static

gesture is not used for human-computer interaction.

III.4. Dynamic Gesture Recognition using HMM

This step requires the recognition ofdynamic hand

gesture since the process considered is the dynamic

process and needs to be handled with the learning model

that can be accommodated with temporal time.

REPRINT



377

Fig. 4. 10 static hand gesture pose (later this pose used for sequence

observation label for dynamic hand gesture)

We propose HMM model [16] to accommodate this

dynamic motion gesture model. Hidden Markov Model is

one of the models based on the statistic. HMM is often

used in applications that intersect with time. HMM

possesses three main parameters, λ = (Π, A, B) [17],

where Π represents the vector beginning, A is the matrix

of transition while B is the matrix emission. Furthermore,

HMM has three major problems, which are Evaluation,

Decoding, and Training. However, these three problems

can be solved using the method of Forward-Backward

algorithm, Viterbi, and Baum-Welch [17].

In the application, HMM is often implemented using

three topologies. There are main Fully Connected

(Ergodic model) where all the states are connected as a

whole between the states; (Left-Right model) where each

state can only be returned to the state itself and in the

next state; (Parallel-Left-Right model) where each state

can only be repopulated themselves or towards the next

state. Shown in Figs. 6, the input of the model is the

result of sequence classification static hand gesture from

SVM. Figs. 5 show dynamic hand gestures in this

system.

Left Click

Right Click

Swipe

Right

Swipe

Left

Grap

(a)

(b)

(c)

(d)

(e)

Figs. 5. 5 Dynamic hand motion gesture that replacing mouse

functionality (random sample images from frame 1 until the end of

videos), (a) gesture for “Left Click”,(b) gesture for “Right Click”,(c)

gesture for “Swipe Left”,(d) gesture for “Swipe Right”

(e) gesture for “Grap”

(a)

(b)

(c)

Figs. 6. (a) ergodic, (b) left-right, (c) parallel-left-right [18] extracted

froma tutorial on hidden markov model

In this process, we train HMM model with HMM type

Ergodic and left-right then for each type of HMM.We

train it usingthe total hidden state from 2 until 9.

IV. Result and Discussion

Our proposed system is implemented in .net using

EmguCV wrapper library, libsvm, and accord hmm.

Fig. 7 shows the system implementation screenshot.

We use the standard webcam with 240 × 320 pixels

resolution. Moreover, we use conditional indoor lighting

environment. In this process, we use three scenarios to

test the system.

The first scenario that we find is the matching of SVM

parameter C and γ with the value from -3 to 3 for C

parameter and -15 to 3 for γ parameter.

We use four kernels (RBF, Linear, Polynomial, and

Sigmoid) in here.

Fig. 7. Hand motion gesture recognition system

REPRINT



378

The second scenario we see is the matching of HMM

parameter topology (Ergodic, left-right, and parallel-left-

right) and we test the number of the hidden state with the

value from 2-9. In this scenario, we will test it using

isolated hand motion gesture.

The third scenario is that we will test the best

parameter of SVM and HMM with continuous hand

motion gesture. Then, we calculate the accuracy with the

following equation:

(2)

We use two datasets that were created by our camera.

The first dataset contains 50 images used to train and 50

images to test each static gesture.

In this dataset, every image has various orientation

and scale. The second dataset contains 30 videos used to

train and 30 videos to test each isolated dynamic gesture

recognition.

On the other hand, we have two videos to test the

continuous dynamic hand gesture. One video has five

gestures that are performed sequentially, and another

video contains five gestures done randomly. In other

words, our dataset contains 1000 images and 302 videos.

Table I shows that kernel RBF with parameter C = 3 and

γ = -1 achieves maximum accuracy training at 100% and

93.4% for the testing dataset.

TABLE I

RESULT TEST SCENARIO 1 FOR SVM PARAMETER

Kernel Type C Value

(2x) γ Value (2x)

Training

Accuracy (%)

Testing

Accuracy (%)

RBF 3 -1 100 93.4

Linear 3 -15 99.4 89.4

Polynomial 0.5 1 76.6 79.4

Sigmoid 3 -3 97 83.8

In Table II topology parallel-left-right gets the

maximum accuracy from training HMM with the number

of hidden state = 4 and kernel SVM Polynomial.

On the other hand, Figs. 8, 9, 10, 11 show that the best

accuracy for hand motion gesture is HMM with topology

Parallel-Left-Right and with 2-4 number of hidden state.

The highest accuracy is achieved by using SVM

kernel polynomial with parameter C = 0.5 and γ = -1.

Then, Ergodic topology has the worst accuracy in all

SVM kernel.This is because the sequence observation

only needs various three sequences for each gesture,

whileErgodic topology, which has more transitions than

lef-right topology,was used to solve more complex

problem [17].

TABLE II

RESULT TEST SCENARIO 1USING RBF KERNEL SVM

HMM

Topology

Number of

Hidden State

Training

Accuracy (%)

Testing

Accuracy (%)

Ergodic 2 94 80

Left-Right 2 97 89.34

Parallel-Left-

Right 4 99 89.34

TABLE III

RESULT TEST SCENARIO 1 USING LINEAR KERNEL SVM

HMM

Topology

Number of

Hidden State

Training

Accuracy (%)

Testing

Accuracy (%)

Ergodic 2 51.34 38.67

Left-Right 2 94.67 83.34

Parallel-Left-Right 2 98 82

TABLE IV

RESULT TEST SCENARIO 1 USING POLYNOMIAL KERNEL SVM

HMM Topology Number of

Hidden State

Training

Accuracy (%)

Testing

Accuracy (%)

Ergodic 2 45.34 41.34

Left-Right 2 98 94

parallel-Left-Right 4 98 94.67

TABLE V

RESULT TEST SCENARIO 1 USING SIGMOID KERNEL SVM

HMM Topology Number of

Hidden State

Training

Accuracy (%)

Testing Accuracy

(%)

Ergodic 2 94 77.33333

Left-Right 2 96.66667 82.66667

Parallel-Left-Right 4 95.33333 82

Fig. 8 below shows the result of graph in finding the

best topology and number of hidden state in HMM.

The last scenario is to test the best model of SVM and

HMM for continuous dynamic hand gesture recognition.

The system shows a good result, that continuous hand

gesture with convergent dataset testing (videos that

contain only one gesture variance) can classify the

gesture with achievement up to 83%. Otherwise, testing

with videos that have random gesture achieves 63%. This

happens because convergent gesture videos have the

same sequence of image observation over all. Therefore,

HMM gets high probability to choose the same sequence

of image observation. If we test it with videos that have

the random sequence of gesture, the system gets false

recognition.

Fig. 8. HMM topology and number of hidden state testing result

for kernel RBF SVM


for kernel LINEAR SVM

REPRINT



379


for kernel POLYNOMIAL SVM


for kernel SIGMOID SVM

V. Conclusion and Future Works

This paper suggests a system for human-computer

interaction using hand motion gesture. This system

achieves an average accuracy of about 97%. The highest

accuracy of static hand gesture is obtained from SVM

with kernel RBF, C = 3 and γ = -1.

On the other hand, the highest accuracy for dynamic

hand gesture is from HMM with topology parallel-left-

right using kernel polynomial with parameter C = 0.5 and

γ = -1. It reaches 98% of accuracy.

By using HMM topology parallel-left-right with four

hidden states, the system presents 97% of accuracy.

In the future, our system will focus on recognition

gesture spotting using a depth camera to create effective

skin color segmentation.

References

[1] M. Panwar and P. Singh Mehra, “Hand gesture recognition for

human computer interaction,” 2011 Int. Conf. Image Inf. Process.,

no. Iciip, pp. 1–7, 2011.

[2] H. M. Gamal, H. M. Abdul-Kader, and E. a. Sallam, “Hand

gesture recognition using fourier descriptors,” 2013 8th Int. Conf.

Comput. Eng. Syst., pp. 274–279, 2013.

[3] R. Xu, S. Zhou, and W. Li, “MEMS accelerometer based

nonspecific-user hand gesture recognition,” Sensors Journal,

IEEE, vol. 12, no. 5, pp. 1166–1173, 2012.

[4] P. Kumar, J. Verma, and S. Prasad, “Hand Data Glove: A

Wearable Real-Time Device for Human-Computer Interaction,”

Int. J. Adv. Sci. Technol., vol. 43, pp. 15–26, 2012.

[5] M. K. Bhuyan, D. Ajay Kumar, K. F. MacDorman, and Y.

Iwahori, “A novel set of features for continuous hand gesture

recognition,” J. Multimodal User Interfaces, 2014.

[6] Demidova, L., Sokolova, Y., Nikulchev, E., Use of Fuzzy

Clustering Algorithms Ensemble for SVM Classifier

Development, (2015) International Review on Modelling and

Simulations (IREMOS), 8 (4), pp. 446-457.

[7] Ab Aziz, N., Abdul Rahman, T., Zakaria, Z., Reactive Power

Planning for Maximum Load Margin Improvement Using Fast

Artificial Immune Support Vector Machine (FAISVM), (2014)

International Review of Automatic Control (IREACO), 7 (5), pp.

436-447.

[8] M. Elmezain, A. Al-Hamadi, J. Appenrodt, and B. Michaelis, “A

Hidden Markov Model-based continuous gesture recognition

system for hand motion trajectory,” 2008 19th Int. Conf. Pattern

Recognit., pp. 3–6, 2008.

[9] M. R. Malgireddy, J. J. Corso, S. Setlur, V. Govindaraju, and D.

Mandalapu, “A framework for hand gesture recognition and

spotting using sub-gesture modeling,” Proc. - Int. Conf. Pattern

Recognit., pp. 3780–3783, 2010.

[10] H. Y. Lai and H. J. Lai, “Real-Time Dynamic Hand Gesture

Recognition,” 2014 Int. Symp. Comput. Consum. Control, no. 1,

pp. 658–661, 2014.

[11] H. Yeo, B. Lee, and H. Lim, “Hand tracking and gesture

recognition system for human-computer interaction using low-

cost hardware,” Multimed Tools Applications 74(1) pp. 2687–

2715, 2015.

[12] Regan, D., Srivatsa, S., Mixed Pixel Wise Characterization Based

on HMM and Hyper spectral Image Gradient Enhancement for

Classification Using SVM-FSK, (2014) International Review on

Computers and Software (IRECOS), 9 (6), pp. 1017-1026.

[13] M. Schuster, “Largest Empty Circle Problem,”

https://www.cs.swarthmore.edu/~adanner/cs97/s08/papers/schuste

r.pdf pp. 28–37, 2008.

[14] L. Gericke, M. Wenzel, R. Gumienny, C. Willems, and C. Meinel,

“Handwriting recognition for a digital whiteboard collaboration

platform,” Proc. 2012 Int. Conf. Collab. Technol. Syst. CTS 2012,

pp. 226–233, 2012.

[15] C. R. Souza and E. B. Pizzolato, “Sign Language Recognition

with Support Vector Machines and Hidden Conditional Random

Fields: Going from Fingerspelling to Natural Articulated Words"

Machine Learning and Data Mining in Pattern Recognition,” vol.

7988, of the series Lecture Notes in Computer Science pp 84-98,

2013.

[16] A. Ramamoorthy, N. Vaswani, S. Chaudhury, and S. Banerjee,

“Recognition of dynamic hand gestures,” Pattern Recognition,

vol. 36, no. 9, pp. 2069–2081, 2003.

[17] M. Elmezain and A. Al-hamadi, “A Hidden Markov Model-Based

Isolated and Meaningful Hand Gesture Recognition,”Proceedings

Of World Academy Of Science, Engineering And Technology

(WASET) Vol. 31 July 2008 pp. 393–400, 2008.

[18] L. R. Rabiner, “A Tutorial on Hidden Markov Models and

Selected Application in Speech Recognition.” Proceeding of the

IEEE Vol. 77 No. 2 February 1989 pp. 257 - 286. 1989.

Authors’ information

Magister in Information Technology, Binus Graduate Program, Bina

Nusantara University, Jakarta, Indonesia.

E-mails: [email protected]

[email protected]

Fitra B. Adinugraha received his Bachelor

degree in Informatics majoring in Computer

Science from Institute Technology Telkom

(ITT), Indonesia in 2011. Presently, he is

pursuing the Graduate Program of Information

Technology at Bina Nusantara University,

Indonesia. His research interest includes

Human Computer Interaction (HCI), Natural

User Interface (NUI), Mobile Applications,

Cloud Storage, and Enterprise Architecture Design.

REPRINT

mailto:[email protected]

mailto:[email protected]



380

Suharjito is the Head of Information

Technology Department in Binus Graduate

Program of Binus University. He received under

graduated degree in mathematics from The

Faculty of Mathematics and Natural Science in

GadjahMada University, Yogyakarta, Indonesia

in 1994. He received master degree in

information technology engineering from

Sepuluh November Institute of Technology, Surabaya, Indonesia in

2000. He received the PhD degree in system engineering from the

Bogor Agricultural University (IPB), Bogor, Indonesia in 2011.His

research interests are intelligent system, Fuzzy system, image

processing and software engineering.

REPRINT

International Review on Computers and Software (IRECOS)

Aims and scope

The International Review on Computers and Software (IRECOS) is a peer-reviewed journal that publishes original papers on all branches of the academic Computer Science and Engineering communities. Thematic areas include, but are not limited to: Computer Science Theory, Methods and Tools Software engineering, algorithms and complexity, computational logic, formal methods, heuristics, mathematics and models of computation, programming languages and semantics. Computer and Communications Networks and Systems Network and distributed architectures and protocols, traffic engineering, resource management and Quality of Service, network monitoring and traffic measurements, wireless networks, personal and body area networks, vehicular networks, content and service-centric networking, multimedia communications and standards, energy efficient/green networks, opportunistic and cognitive networks. Computational Intelligence, Machine Learning and Data Analytics Human computer interaction, computational science, pattern recognition, computer vision, speech processing, machine intelligence and reasoning, web science, databases, information retrieval, visualisation, current applications domains, e.g. Healthcare and BioInformatics, and emerging application domains, e.g. big data. Security in Computer Systems and Networks Computer systems security, hardware and embedded systems security, security protocol design and analysis, cryptography and cryptanalysis, intrusion detection systems and techniques, user authentication techniques and systems. Hardware Design Computer architectures, parallel architectures, operating systems and signal processing.

Instructions for submitting a paper

The journal publishes invited tutorials or critical reviews; original scientific research papers (regular papers), letters to the Editor and research notes which should also be original presenting proposals for a new research, reporting on research in progress or discussing the latest scientific results in advanced fields; short communications and discussions, book reviews, reports from meetings and special issues describing research in any of the above thematic areas. All papers will be subjected to a fast editorial process. Any paper will be published within two months from the submitted date, if it has been accepted.

Papers must be correctly formatted, in order to be published. An Author guidelines template file can be found at the following web address:

www.praiseworthyprize.org/jsm/?journal=irecos Manuscripts should be sent via e-mail as attachment in .doc and .pdf formats to:

[email protected]

The regular paper page length limit is defined at 15 formatted Review pages, including illustrations, references and author(s) biographies. Pages 16 and above are charged 10 euros per page and payment is a prerequisite for publication. Abstracting and Indexing Information:

Cambridge Scientific Abstracts (CSA/CIG) Academic Search Complete (EBSCO Information Services) Elsevier Bibliographic Database - SCOPUS Index Copernicus - IC Journal Master List 2012: ICV 6.45

Autorizzazione del Tribunale di Napoli n. 59 del 30/06/2006

REPRINT

1828-6003(201605)11:5;1-7

REPRINT