IT convergence and security 2017 volume 1

Lecture Notes in Electrical Engineering 449

Kuinam J. Kim
Hyuncheol Kim
Nakhoon Baek
Editors

IT Convergence
and Security
2017
Volume 1


Lecture Notes in Electrical Engineering
Volume 449

Board of Series editors
Leopoldo Angrisani, Napoli, Italy
Marco Arteaga, Coyoacán, México
Samarjit Chakraborty, München, Germany
Jiming Chen, Hangzhou, P.R. China
Tan Kay Chen, Singapore, Singapore
Rüdiger Dillmann, Karlsruhe, Germany
Haibin Duan, Beijing, China
Gianluigi Ferrari, Parma, Italy
Manuel Ferre, Madrid, Spain
Sandra Hirche, München, Germany
Faryar Jabbari, Irvine, USA
Janusz Kacprzyk, Warsaw, Poland
Alaa Khamis, New Cairo City, Egypt
Torsten Kroeger, Stanford, USA

Tan Cher Ming, Singapore, Singapore
Wolfgang Minker, Ulm, Germany
Pradeep Misra, Dayton, USA
Sebastian Möller, Berlin, Germany
Subhas Mukhopadyay, Palmerston, New Zealand
Cun-Zheng Ning, Tempe, USA
Toyoaki Nishida, Sakyo-ku, Japan
Bijaya Ketan Panigrahi, New Delhi, India
Federica Pascucci, Roma, Italy
Tariq Samad, Minneapolis, USA
Gan Woon Seng, Nanyang Avenue, Singapore
Germano Veiga, Porto, Portugal
Haitao Wu, Beijing, China
Junjie James Zhang, Charlotte, USA


About this Series
“Lecture Notes in Electrical Engineering (LNEE)” is a book series which reports
the latest research and developments in Electrical Engineering, namely:
•
•
•
•
•

Communication, Networks, and Information Theory
Computer Engineering
Signal, Image, Speech and Information Processing
Circuits and Systems
Bioengineering

LNEE publishes authored monographs and contributed volumes which present
cutting edge research information as well as new perspectives on classical fields,
while maintaining Springer’s high standards of academic excellence. Also
considered for publication are lecture materials, proceedings, and other related
materials of exceptionally high quality and interest. The subject matter should be
original and timely, reporting the latest research and developments in all areas of
electrical engineering.
The audience for the books in LNEE consists of advanced level students,
researchers, and industry professionals working at the forefront of their fields. Much
like Springer’s other Lecture Notes series, LNEE will be distributed through
Springer’s print and electronic publishing channels.

More information about this series at />

Kuinam J. Kim Hyuncheol Kim
Nakhoon Baek
•

Editors

IT Convergence
and Security 2017
Volume 1

123


Editors
Kuinam J. Kim

iCatse, B-3001, Intellige 2
Kyonggi University
Seongnam-si, Kyonggi-do
Korea (Republic of)

Nakhoon Baek
School of Computer Science
and Engineering
Kyungpook National University
Daegu
Korea (Republic of)

Hyuncheol Kim
Computer Science
Namseoul University
Cheonan, Chungcheongnam-do
Korea (Republic of)

ISSN 1876-1100
ISSN 1876-1119 (electronic)
Lecture Notes in Electrical Engineering
ISBN 978-981-10-6450-0
ISBN 978-981-10-6451-7 (eBook)
DOI 10.1007/978-981-10-6451-7
Library of Congress Control Number: 2017951408
© Springer Nature Singapore Pte Ltd. 2018
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar

methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature
The registered company is Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore


Preface

This LNEE volume contains the papers presented at the iCatse International
Conference on IT Convergence and Security (ICITCS 2017) which was held in
Seoul, South Korea, during September 25 to 28, 2017.
The conferences received over 200 paper submissions from various countries.
After a rigorous peer-reviewed process, 69 full-length articles were accepted for
presentation at the conference. This corresponds to an acceptance rate that was very
low and is intended for maintaining the high standards of the conference
proceedings.
ICITCS2017 will provide an excellent international conference for sharing
knowledge and results in IT Convergence and Security. The aim of the conference
is to provide a platform to the researchers and practitioners from both academia and
industry to meet the share cutting-edge development in the field.
The primary goal of the conference is to exchange, share and distribute the latest

research and theories from our international community. The conference will be
held every year to make it an ideal platform for people to share views and experiences in IT Convergence and Security-related fields.
On behalf of the Organizing Committee, we would like to thank Springer for
publishing the proceedings of ICITCS2017. We also would like to express our
gratitude to the ‘Program Committee and Reviewers’ for providing extra help in the
review process. The quality of a refereed volume depends mainly on the expertise
and dedication of the reviewers. We are indebted to the Program Committee
members for their guidance and coordination in organizing the review process and
to the authors for contributing their research results to the conference.
Our sincere thanks go to the Institute of Creative Advanced Technology,
Engineering and Science for designing the conference Web page and also spending
countless days in preparing the final program in time for printing. We would also

v


vi

Preface

like to thank our organization committee for their hard work in sorting our
manuscripts from our authors.
We look forward to seeing all of you next year’s conference.
Kuinam J. Kim
Nakhoon Baek
Hyuncheol Kim
Editors of ICITCS2017


Organizing Committee

General Chairs
Hyung Woo Park
Nikolai Joukov
Nakhoon Baek
HyeunCheol Kim

KISTI, Republic of Korea
New York University and modelizeIT Inc, USA
Kyungpook National University, Republic
of Korea
NamSeoul University, Republic of Korea

Steering Committee
Nikolai Joukov
Borko Furht
Bezalel Gavish
Kin Fun Li
Kuinam J. Kim
Naruemon
Wattanapongsakorn
Xiaoxia Huang
Dato’ Ahmad Mujahid
Ahmad Zaidi

New York University and modelizeIT Inc, USA
Florida Atlantic University, USA
Southern Methodist University, USA
University of Victoria, Canada
Kyonggi University, Republic of Korea

King Mongkut’s University of Technology
Thonburi, Thailand
University of Science and Technology Beijing,
China
National Defence University of Malaysia,
Malaysia

Program Chair
Kuinam J. Kim

Kyonggi University, Republic of Korea

vii


viii

Organizing Committee

Publicity Chairs
Miroslav Bureš
Dan (Dong-Seong) Kim
Sanggyoon Oh
Xiaoxia Huang

Czech Technical University, Czech Republic
University of Canterbury, New Zealand
BPU Holdings Corp, Republic of Korea
University of Science and Technology Beijing,
China

Financial Chair
Donghwi Lee

Dongshin University, Republic of Korea

Publication Chairs
Minki Noh
Hongseok Jeon

KISTI, Republic of Korea
ETRI, Republic of Korea

Organizers and Supporters
Institute of Creative Advanced Technologies, Science and Engineering
Korea Industrial Security Forum
Korean Convergence Security Association
University of Utah, Department of Biomedical Informatics, USA
River Publishers, Netherlands
Czech Technical University, Czech Republic
Chonnam National University, Republic of Korea
University of Science and Technology Beijing, China
King Mongkut’s University of Technology Thonburi, Thailand
ETRI, Republic of Korea
KISTI, Republic of Korea
Kyungpook National University, Republic of Korea
Seoul Metropolitan Government

Program Committee
Bhagyashree S R

Richard Chbeir
Nandan Mishra

ATME College of Engineering, Mysore,
Karnataka, India
Université Pau & Pays Adour (UPPA), France
Cognizant Technology Solutions, USA


Organizing Committee

Reza Malekian
Sharmistha Chatterjee
Shimpei Matsumoto
Sharifah Md Yasin
C. Christober Asir Rajan
Chin-Chen Chang
Danilo Pelusi
Necmi Taspinar
Alvaro Suarez
Wail Mardini
Josep Domingo-Ferrer
Yaxin Bi
Jie Zhang
Miroslav N. Velev
Johann M. Marquez-Barja
Nicholas Race
Gaurav Sharma
Yanling Wei
Mohd Fairuz Iskandar

Othman
Harikumar Rajaguru
Chittaranjan Pradhan
Frank Werner
Suranga Hettiarachchi
Sa’adah Hassan
Frantisek Capkovic
Oscar Mortagua Pereira
Filippo Gaudenzi
Virgilio Cruz Machado
Pao-Ann Hsiung
M. Iqbal Saripan
Lorenz Pascal
Helmi Zulhaidi Mohd Shafri
Harekrishna Misra
Nuno Miguel Castanheira
Almeida
Bandit Suksawat
Jitender Grover
Kwangjin Park
Ahmad Kamran Malik

ix

University of Pretoria, South Africa
Florida Atlantic University, USA
Hiroshima Institute of Technology, Japan
University Putra Malaysia, Malaysia
Pondicherry Engineering College, India
Feng Chia University, Taiwan

University of Teramo, Italy
Erciyes University, Kayseri, Turkey
University of Las Palmas de G.C., Spain
Jordan University, Jordan
Universitat Rovira i Virgili, Spain
Ulster University at Jordanstown, UK
Newcastle University, UK
Aries Design Automation, USA
CONNECT Research Centre, Trinity College
Dublin, Ireland
Lancaster University, UK
Université libre de Bruxelles, Belgium
Technical University of Berlin, Germany
Universiti Teknikal Malaysia Melaka (UTeM),
Malaysia
Bannari Amman Institute of Technology,
Sathyamangalam, India
KIIT University, India
Otto-von-Guericke University Magdeburg,
Germany
Indiana University Southeast, USA
Universiti Putra, Malaysia
Institute of Informatics, Slovak Academy
of Sciences, Slovakia
University of Aveiro, Portugal
Università degli Studi di Milano, Italy
Universidade Nova de Lisboa-UNIDEMI,
Portugal
National Chung Cheng University, Taiwan
Universiti Putra Malaysia, Malaysia

University of Haute Alsace, France
Universiti Putra Malaysia, Malaysia
Institute of Rural Management Anand, India
Polytechnic of Leiria, Portugal
King Mongkut’s University, Thailand
IIIT Hyderabad, India
Wonkwang University, Korea
COMSATS Institute of IT, Pakistan


x

Shitala Prasad
Hao Han
Anooj P.K.
Hyo Jong Lee
D’Arco Paolo
Suresh Subramoniam
Abdolhossein Sarrafzadeh
Stelvio Cimato
Ivan Mezei
Terje Jensen
Selma Regina Martins
Oliveira
Firdous Kausar
M. Shamim Kaiser
Maria Leonilde Rocha Varela
Nadeem Javaid
Urmila Shrawankar
Yongjin Yeom

Olivier Blazy
Bikram Das
Edelberto Franco Silva
Wing Kwong
Dae-Kyoo Kim
Nickolas S. Sapidis
Eric J. Addeo
T. Ramayah
Yiliu Liu
Shang-Ming Zhou
Anastasios Doulamis
Baojun Ma
Fatemeh Almasi
Mohamad Afendee Mohamed
Jun Peng
Nestor Michael C. Tiglao
Mohd Faizal Abdollah
Alessandro Bianchi
Reza Barkhi
Mohammad Osman Tokhi
Prabhat K. Mahanti
Chia-Chu Chiang
Tan Syh Yuan
Qiang (Shawn) Cheng
Michal Choras

Organizing Committee

NTU Singapore, Singapore
The University of Tokyo, Japan

Al Musanna College of Technology, Oman
Chonbuk National University,Korea
University of Salerno, Italy
CET School of Management, India
Unitec Institute of Technology, New Zealand
University of Milan, Italy
University of Novi Sad, Serbia
Telenor, Norway
Federal Fluminense University, Brazil
Imam Ibm Saud University, Saudi Arabia
Jahangirnagar University, Bangladesh
University of Minho, Portugal
COMSATS Institute of Information Technology,
Pakistan
RTM Nagpur University, India
Kookmin University, Korea
Université de Limoges, France
NIT Agartala, India
Universidade Federal de Juiz de Fora, Brazil
Hofstra University, USA
Oakland University, USA
University of Western Macedonia, Greece
DeVry University, USA
Universiti Sains Malaysia, Malaysia
Norwegian University, Norway
Swansea University, UK
National Technical University, Greece
Beijing University, China
Ecole Centrale de Nantes, France
Universiti Sultan Zainal Abidin, Malaysia

University of Texas, USA
University of the Philippines Diliman,
Philippines
University Technical Malaysia Melaka, Malaysia
University of Bari, Italy
Virginia Tech, USA
London South Bank University, UK
University of New Brunswick, Canada
University of Arkansas at Little Rock, USA
Multimedia University, Malaysia
Southern Illinois University, USA
University of Science and Technology, Korea


Organizing Committee

El-Sayed M. El-Alfy
Abdelmajid Khelil
James Braman
Rajesh Bodade
Nasser-Eddine Rikli
Zeyar Aung
Schahram Dustdar
Ya Bin Dang
Marco Aiello
Chau Yuen
Yoshinobu Tamura
Nor Asilah Wati Abdul
Hamid
Pavel Loskot

Rika Ampuh Hadiguna
Hui-Ching Hsieh
Javid Taheri
Fu-Chien Kao
Siana Halim
Goi Bok Min
Shamim H Ripon
Munir Majdalawieh
Hyunsung Kim
Ahmed A. Abdelwahab
Vana Kalogeraki
Joan Ballantine
Jianbin Qiu
Mohammed Awadh Ahmed
Ben Mubarak
Mehmet Celenk
Shakeel Ahmed
Sherali Zeadally
Seung Yeob Nam
Tarig Mohamed Hassan
Vishwas Ruamurthy
Ankit Chaudhary
Mohammad Faiz Liew
Abdullah
Francesco Lo Presti
Muhammad Usman
Kurt Kurt Tutschku

xi

King Fahd University, Saudi Arabia
Landshut University, Germany
The Community College of Baltimore County,
USA
Defence College of Telecommunication
Engineering, India
King Saud University, Saudi Arabia
Khalifa University, United Arab Emirates
TU Wien, Austria
IBM Research, China
University of Groningen, Netherlands
Singapore University, Singapore
Tokyo City University, Japan
Universiti Putra Malaysia, Malaysia
Swansea University, UK
Andalas University, Indonesia
Hsing Wu University, Taiwan
Karlstad University, Sweden
Da-Yeh University, Taiwan
Petra Christian University, Indonesia
Universiti Tunku Abdul Rahman, Malaysia
East West University, USA
George Mason University, USA
Kyungil University, Korea
Qassim University, Saudi Arabia
Athens University, Greece
Ulster University, UK
Harbin Institute of Technology, China
Infrastructure University Kuala Lumpur,
Malaysia

Ohio University, USA
King Faisal University, Saudi Arabia
University of Kentucky, USA
Yeungnam University, Korea
University of Khartoum, Sudan
Visvesvaraya Technological University, India
Northwest Missouri State University, USA
University Tun Hussein Onn, Malaysia
University of Rome Tor Vergata, Italy
National University of Sciences and Technology
(NUST), Pakistan
Blekinge Institute of Technology, Sweden


xii

Ivan Ganchev
Mohammad M. Banat
David Naccache
Kittisak Jermsittiparsert
Pierluigi Siano
Hiroaki Kikuchi
Ireneusz Czarnowski
Lingfeng Wang
Somlak Wannarumon
Kielarova
Chang Wu Yu
Kennedy Njenga
Kok-Seng Wong
Ray C.C. Cheung

Stephanie Teufel
Nader F. Mir
Zongyang Zhang
Alexandar Djordjevich
Chew Sue Ping
Saeed Iqbal Khattak
Chuangyin Dang
Riccardo Martoglia
Qin Xin
Andreas Dewald
Rubing Huang
Sangseo Parko
Mainguenaud Michel
Selma Regina Martins
Oliveira
Enrique Romero-Cadaval
Noraini Che Pa
Minghai Jiao
Ruay-Shiung Chang
Afizan Azman
Yusmadi Yah Jusoh
Daniel B.-W. Chen
Wuxu Peng
Noridayu Manshor
Alberto Núñez Covarrubias

Organizing Committee

University of Limerick, Ireland/University
of Plovdiv “Paisii Hilendarski”

Jordan University, Jordan
Ecole normale supérieure, France
Rangsit University, Thailand
University of Salerno, Italy
Meiji University, Japan
Gdynia Maritime University, Poland
University of Wisconsin-Milwaukee, USA
Naresuan University, Thailand
Chung Hua University, Taiwan
University of Johannesburg,
Republic of South Africa
Soongsil University, Korea
City University of Hong Kong, China
University of Fribourg, Switzerland
San Jose State University, California
Beihang University, China
City University of Hong Kong, China
National Defense University of Malaysia,
Malaysia
University of Central Punjab, Pakistan
City University of Hong Kong, China
FIM, University of Modena and Reggio Emilia,
Italy
University of the Faroe Islands, Faroe Islands,
Denmark
ERNW Research GmbH, Germany
Jiangsu University, China
Korea
Insitut National des sciences Appliquées Rouen,
France

Universidade Federal Fluminense, Brazil
University of Extremadura, Spain
Universiti Putra Malaysia (UPM), Malaysia
Northeastern University, USA
National Taipei University of Business, Taiwan
Multimedia University, Malaysia
Universiti Putra Malaysia, Malaysia
Monash University, Australia
Texas State University, USA
Universiti Putra Malaysia, Malaysia
Universidad Complutense de Madrid, Spain


Organizing Committee

Stephen Flowerday
Anton Setzer
Jinlei Jiang
Lorna Uden
Wei-Ming Lin
Lutfiye Durak-Ata
Srinivas Sethi
Edward Chlebus
Siti Rahayu Selamat
Nur Izura Udzir
Jinhong Kim
Michel Toulouse
Vicente Traver Salcedo
Hardeep Singh
Jiqiang Lu

Juntae Kim
Kuo-Hui Yeh
Ljiljana Trajkovic
Kouichi Sakurai
Jay Kishigami
Dachuan Huang
Ankit Mundra
Hanumanthappa J.
Muhammad Zafrul Hasan
Christian Prehofer
Lim Tong Ming
Yuhuan Du
Subrata Acharya
Warusia Yassin
Fevzi Belli

xiii

University of Fort Hare, Republic of South Africa
Swansea University, UK
Tsinghua University, China
Staffordshire University, UK
University of Texas at San Antonio, USA
Istanbul Technical University, Turkey
IGIT Sarang, India
Illinois Institute of Technology, USA
Universiti Teknikal Malaysia Melaka, Malaysia
Universiti Putra Malaysia, Malaysia
Seoil University, Korea
Vietnamese-German University, Vietnam

Universitat Politècnica de València, Spain
Ferozepur College of Engg & Technology
(FCET) India, India
Institute for Infocomm Research, Singapore
Dongguk University, Korea
National Dong Hwa University, China
Simon Fraser University, Canada
Kyushu Univ., Japan
Muroran Institute of Technology, Japan
Snap Inc., USA
Department of IT, School of Computing and IT,
Manipal University Jaipur, India
University of Mysore, India
Texas A&M International University, USA
An-Institut der Technischen Universitaet
Muenchen, Germany
Sunway University, Malaysia
Software Engineer, Dropbox, San Francisco,
USA
Towson University, USA
Universiti Teknikal Malaysia Melaka, Malaysia
Univ. Paderborn, Germany


Contents

Machine Learning and Deep Learning
Image-Based Content Retrieval via Class-Based
Histogram Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
John Kundert-Gibbs

3

Smart Content Recognition from Images Using a Mixture
of Convolutional Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Tee Connie, Mundher Al-Shabi, and Michael Goh

11

Failure Part Mining Using an Association Rules Mining
by FP-Growth and Apriori Algorithms: Case of ATM
Maintenance in Thailand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Nachirat Rachburee, Jedsada Arunrerk, and Wattana Punlumjeak
Improving Classification of Imbalanced Student Dataset
Using Ensemble Method of Voting, Bagging, and Adaboost
with Under-Sampling Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Wattana Punlumjeak, Sitti Rugtanom, Samatachai Jantarat,
and Nachirat Rachburee
Reduction of Overfitting in Diabetes Prediction Using Deep
Learning Neural Network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Akm Ashiquzzaman, Abdul Kawsar Tushar, Md. Rashedul Islam,
Dongkoo Shon, Kichang Im, Jeong-Ho Park, Dong-Sun Lim,
and Jongmyon Kim
An Improved SVM-T-RFE Based on Intensity-Dependent
Normalization for Feature Selection in Gene Expression
of Big-Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chayoung Kim and Hye-young Kim

19

27

35

44

xv


xvi

Contents

Vehicle Counting System Based on Vehicle Type Classification
Using Deep Learning Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Suryanti Awang and Nik Mohamad Aizuddin Nik Azmi

52

Metadata Discovery of Heterogeneous Biomedical Datasets
Using Token-Based Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Jingran Wen, Ramkiran Gouripeddi, and Julio C. Facelli

60

Heavy Rainfall Forecasting Model Using Artificial Neural Network
for Flood Prone Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Junaida Sulaiman and Siti Hajar Wahab

68

Communication and Signal Processing
I-Vector Extraction Using Speaker Relevancy for Short Duration
Speaker Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Woo Hyun Kang, Won Ik Cho, Se Young Jang, Hyeon Seung Lee,
and Nam Soo Kim

79

A Recommended Replacement Algorithm for the Scalable
Asynchronous Cache Consistency Scheme . . . . . . . . . . . . . . . . . . . . . . . .
Ramzi A. Haraty and Lama Hasan Nahas

88

Multiple Constraints Satisfaction-Based Reliable Localization
for Mobile Underwater Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . .
Guangyuan Wang, Yongji Ren, Xiaofeng Xu, and Xiaolei Liu

97

A Design of Kernel-Level Remote Memory Extension System. . . . . . . . . 104
Shinyoung Ahn, Eunji Lim, Wan Choi, Sungwon Kang,
and Hyuncheol Kim
A Comparison of Model Validation Techniques for Audio-Visual
Speech Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Thum Wei Seong, Mohd Zamri Ibrahim, Nurul Wahidah Binti Arshad,
and D.J. Mulvaney
Multi-focus Image Fusion Based on Non-subsampled Shearlet
Transform and Sparse Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

Weiguo Wan and Hyo Jong Lee
Implementation of Large-Scale Network Flow Collection System
and Flow Analysis in KREONET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Chanjin Park, Wonhyuk Lee, and Hyuncheol Kim


Contents

xvii

Computer Vision and Applications
A Novel BP Neural Network Based System for Face Detection . . . . . . . . 137
Shuhui Cao, Zhihao Yu, Xiao Lin, Linhua Jiang, and Dongfang Zhao
A Distributed CBIR System Based on Improved SURF
on Apache Spark. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Tingting Huang, Zhihao Yu, Xiao Lin, Linhua Jiang, and Dongfang Zhao
Fish Species Recognition Based on CNN Using Annotated Image . . . . . . 156
Tsubasa Miyazono and Takeshi Saitoh
Head Pose Estimation Using Convolutional Neural Network. . . . . . . . . . 164
Seungsu Lee and Takeshi Saitoh
Towards Robust Face Sketch Synthesis with Style
Transfer Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Philip Chikontwe and Hyo Jong Lee
Object Segmentation with Neural Network Combined GrabCut . . . . . . . 180
Yong-Gyun Choi and Sukho Lee
From Voxels to Ellipsoids: Application to Pore Space
Geometrical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Alain Tresor Kemgue and Olivier Monga
Investigation of Dimensionality Reduction in a Finger Vein
Verification System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

Ei Wei Ting, M.Z. Ibrahim, and D.J. Mulvaney
Palm Vein Recognition Using Scale Invariant Feature Transform
with RANSAC Mismatching Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Shi Chuan Soh, M.Z. Ibrahim, Marlina Binti Yakno, and D.J. Mulvaney
Speed Limit Traffic Sign Classification Using Multiple
Features Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Aryuanto Soetedjo and I. Komang Somawirata
Future Network Technology
Big Streaming Data Sampling and Optimization . . . . . . . . . . . . . . . . . . . 221
Abhilash Kancharala, Nohjin Park, Jongyeop Kim, and Nohpill Park
Artificial Intelligence and Robotics
Fuzzy Model for the Average Delay Time on a Road Ending
with a Traffic Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Zsolt Csaba Johanyák and Rafael Pedro Alvarez Gil


xviii

Contents

Characteristics of Magnetorheological Fluids Applied to Prosthesis
for Lower Limbs with Active Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Oscar Arteaga, Diego Camacho, Segundo M. Espín, Maria I. Erazo,
Victor H. Andaluz, M. Mounir Bou-Ali, Joanes Berasategi,
Alvaro Velasco, and Erick Mera
Multi-Objective Shape Optimization in Generative Design:
Art Deco Double Clip Brooch Jewelry Design . . . . . . . . . . . . . . . . . . . . . 248
Sunisa Sansri and Somlak Wannarumon Kielarova
Adaptation of the Bioloid Humanoid as an Auxiliary in the Treatment
of Autistic Children. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256

Luis Proaño, Vicente Morales, Danny Pérez, Víctor H. Andaluz,
Fabián Baño, Ricardo Espín, Kelvin Pérez, Esteban Puma,
Jimmy Sangolquiza, and Cesar A. Naranjo
Autonomous Assistance System for People with Amyotrophic
Lateral Sclerosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Alex Santana G., Orfait Ortiz C, Julio F. Acosta, and Víctor H. Andaluz
Coordinated Control of a Omnidirectional Double
Mobile Manipulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Jessica S. Ortiz, María F. Molina, Víctor H. Andaluz, José Varela,
and Vicente Morales
Heterogeneous Cooperation for Autonomous Navigation
Between Terrestrial and Aerial Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Jessica S. Ortiz, Cristhian F. Zapata, Alex D. Vega, Alex Santana G.,
and Víctor H. Andaluz
Linear Algebra Applied to Kinematic Control
of Mobile Manipulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
Víctor H. Andaluz, Edison R. Sásig, William D. Chicaiza,
and Paola M. Velasco
Software Engineering and Knowledge Engineering
Enterprise Requirements Management Knowledge Towards
Digital Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Shuichiro Yamamoto
Qualitative Requirements Analysis Process in Organization
Goal-Oriented Requirements Engineering (OGORE)
for E-Commerce Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
Fransiskus Adikara, Sandfreni, Ari Anggarani, and Ernawati
An Improvement of Unknown-Item Search for OPAC
Using Ontology and Academic Information . . . . . . . . . . . . . . . . . . . . . . . 325
Peerasak Intarapaiboon

Contents

xix

Activities in Software Project Management Class: An Experience
from Flipped Classrooms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Sakgasit Ramingwong and Lachana Ramingwong
Solo Scrum in Bureaucratic Organization: A Case Study
from Thailand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Lachana Ramingwong, Sakgasit Ramingwong, and Pensiri Kusalaporn
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349


Machine Learning and Deep Learning


Image-Based Content Retrieval
via Class-Based Histogram Comparisons
John Kundert-Gibbs(&)
The Institute for Artificial Intelligence,
The University of Georgia, Athens, GA 30604, USA


Abstract. Content-based image retrieval has proved to be a fundamental
research challenge for disciplines like search and computer vision. Though
many approaches have been proposed in the past, most of them suffer from poor
image representation and comparison methods, returning images that match the
query image rather poorly when judged by a human. The recent rebirth of deep
learning neural networks has been a boon to CBIR, producing much higher

quality results, yet there are still issues with many recent uses of deep learning.
Our method, which makes use of a pre-trained deep net, compares class-based
histograms between the known image database and query images. This method
produces results that are significantly better than baseline methods we test
against. In addition, we modify the base network in two ways and then use a
weighted voting system to decide on images to display. These modifications
further improve image recall quality.
Keywords: Deep learning
Image based recall Á CBIR

Á Image retrieval Á Content-Based image retrieval Á
Á IBR Á Information retrieval Á Computer vision

1 Introduction
In the last few years, users’ desire to find more images like the one they are currently
viewing has increased dramatically. From personal photo libraries to personal and
business searches, vast numbers of image consumers are interested in finding images
that “look like” the one they are viewing at the moment. As the quantity of stored
images has expanded to a number far beyond what any team of humans could examine,
classify, and catalogue, we have turned to machines running Artificial Intelligence
searches to do the work for us.
The industry terms for recovering images that are visually and semantically similar
to the search image are Content-Based Image Recall (CBIR) or Image-Based Recall
(IBR). The major IBR breakthrough in the past few years has been the use of deep
convolutional neural networks. Even with major advances in IBR, however, the area is
an ongoing topic of research as results are not consistently appropriate. We propose a
new system that can outperform publically available IBR packages on a reasonable size
database of images. Our system utilizes a class histogram approach (described in
Sect. 3) to compare a query image to scores from an image database, producing quality
results rapidly. Though evaluating IBR can prove challenging as the results are

© Springer Nature Singapore Pte Ltd. 2018
K.J. Kim et al. (eds.), IT Convergence and Security 2017,
Lecture Notes in Electrical Engineering 449,
DOI 10.1007/978-981-10-6451-7_1


4

J. Kundert-Gibbs

generally qualitative, we can make some quantitative assessment as well. By comparing two off-the-shelf IBR solutions, as well as an un-retrained and a retrained
network using our method, we show that our system works better than the available
systems, and that further training increases the accuracy of our method.

2 Research in Image-Based Recall
Substantial work has been done on the topic of IBR for more than two decades. Most of
the traditional methods [1–5] require a large number of training instances. Until
recently most training sets were relatively small, thus IBR engines did not have much
to work with. Even with the advent of large image databases like Imagenet, and new
techniques like Support Vector Machines [6, 7] and active learning [8], results have
been only marginal. More recent approaches have made use of ensemble learning.
These ensemble schemes have been successful at improving classification accuracy
through bias or variance reduction, but they do not help reduce the number of samples
and the time required to learn a query concept. An approach based on Support Vector
Machines (SVMs) is proposed in [6], but this approach requires seeds to start, which is
not practically feasible, especially for large database queries.
Conventional IBR approaches usually choose rigid distance functions on some
extracted low-level features for their similarity search mode, such as Euclidean distance. However, a fixed rigid similarity/distance function may not be optimal for the
complex visual image retrieval tasks. As a result recently there has been a surge of
research into designing various distance/similarity measures on low-level features by

exploring machine learning techniques [9–12]. Distance metric learning for image
retrieval has been extensively studied [13–21]. In some instances like [16], class labels
are used to train DML.
Over the past half decade, a rich family of deep learning techniques has been
applied to the field of computer vision and machine learning. Just a few examples are
Deep Belief Networks [22], Boltzmann Machines [23], Restricted Boltzmann Machines [24], Deep Boltzmann Machines [25], and Deep Neural Networks [26, 33]. The
deep convolutional neural networks (CNNs) proposed in [27] got first place in the 2012
image classification task, ILSVRC-2012, proving the worth of this rejuvenated network
architecture. For our method, we make use of a pre-trained VGG- 16 model.

3 IBR Packages
While a number of IBR packages exist, we found two packages based on MATLAB
that are good experimental candidates because they utilize MATLAB as a basis and are
consistent in their underpinnings, using scripts that are open to examination. These two
IBR implementations serve to provide baseline results for comparison with our IBR
method, which is also implemented in MATLAB.
The first package examined is cbires, developed primarily by Joani Mitro. cbires
uses either k-nearest-neighbors (knn) or Support Vector Machines (SVM) plus feature
extraction to perform IBR [28]. The second package, CBIR, was developed by Amine


Image-Based Content Retrieval

5

Ben Khalifa and Faezeh Tafazzoli [29]. CBIR utilizes feature extraction which can
either be done locally or globally. Color and texture features can be extracted globally
or locally, and different distance measures can be invoked to compare images.
The method we have developed operates differently than the two baseline IBR
packages described above. Termed Class-Based Histogram, or CBH-IBR, this system

uses a pre-trained deep learning convolutional neural network—in this case trained on
the Imagenet database [30]—as the basis for image recall. In our case we use a network
trained via matconvnet [31]—a script package for MATLAB that is specifically
designed to create and train convolutional neural networks—that is set up to classify
the 1,000 categories of images that Imagenet contains. While this network,
imagenet-vgg-f.mat, which comes included with the matconvnet download, is intended
for use classifying a single output class, we note that the final layer (a fully connected
softmax probability layer) produces a 1,000 element vector that contains a probability
between 0.0 and 1.0 for each of the classes. We exploit this fact by running a
MATLAB script that records the full 1,000 element vector for each image in a resource
database (from which images are pulled to match the query image). These vectors
create a histogram of each of the 1,000 possible classes. When a query image is
submitted via another script, its class vector is calculated and then compared via RMSE
to each of the other images, as shown in the following formula.
Sbest ¼ min

images
X
X classes
i¼1

À

qj À bij

Á2

!
; q ¼ queryimage; b ¼ baseimages

j¼1

The n closest matches (smallest differential RMSE) in the resource database are chosen
and displayed, as is the error between the query and resource images.
Even for images that do not contain one of the 1,000 Imagenet classes, each
histogram turns out to be distinct and therefore can be used to retrieve similar images.
The fact that images not featuring an object from the Imagenet classes can be queried
and matched is exceptionally useful as it means this method can recall images it was
not trained to recognize at all. We have used our CBH method to test and recall many
query images that are not classified within the 1,000 image-net classes, and while these
images would not produce viable classifications via the network, they produce good
results for IBR.

4 Experimental Setup and Methodology
We utilize the F-measure to determine the quality of results in our experiment: we
count images that are “very close” to the query image, images that are “pretty close,”
and images that are “not at all close.” From these relatively straightforward metrics we
calculate the precision of our results, either using only the correct (very close) images,
or both correct and partially correct results. In Table 1, we provide the F-measure for
both the correct results and the correct + partially correct results.
We selected two image sets, the Caltech 256 data set [32] and the one included with
the CBIR package [29], and combined them into an image database of 29,970 images


6

J. Kundert-Gibbs

that fall within 271 classes (many of which are not Imagenet classes). These images
contain between 80 and 200 of each image class/descriptor (e.g., sailboat, horses, bear,

car).1 We then selected 50 images from google.com and duckduckgo.com as test query
images. The images are chosen to be reasonable images given the source image
database; in other words, images that are similar to a large number of images (at least
one class of 80+) within the source images. These images are isolated from the query
database and any training work, so that they remain completely outside the world that
the IBR packages had access to for training or querying.2 For each engine, after
adjusting to find optimum settings, we run a query for each of our 50 test images and
request 20 similar images be output. For each of the 50 search results (with 20 images
each) we count up the number of correct images, the number of partially correct, and
the number of incorrect results, and record them in a spreadsheet. F-measures are
computed for each image query as well as a single F-measure result for the entire 50
image query set for each query technique, shown in Table 1.
In our tests, our pretrained Imagenet network works very well, but still has room for
improvement. We thus tried numerous methods to retrain/refine the network, including
retraining via softmax log loss, top k error, mshinge, and our own modified version of
softmax log loss. While our hope was to find one method that outperformed the original
network in all cases, this did not occur. We thus created a voting method that utilizes
the best three retraining methods—the original network, the network retrained with
softmax log, and the network retrained via minimizing sum of squared errors on
CBH-IBR—creating a results vector of all three methods combined. We sort this new,
combined vector (3 times the length of each return vector, or 60 values in this case) and
take the top 20 results. While a few results are actually worse, most are the same or
improved, so this method produces the best overall F-measure, as presented in Table 1.

5 Results
While our results are somewhat qualitative, as we have to use human judgment to
determine how close IBR results are, we have come up with distinctly differentiated results
that are borne out by direct observation. Our IBR engine performs substantially better than
the baseline packages using the same source image database and the same query images.
The cbires recall engine performs the worst of the group, as evidenced both by its

total F-measure and by observing results. We attempted to improve the results, but
there are very few parameters that can be adjusted via its GUI. We tried both knn and
SVM methods, and found them about the same. Our results are for the knn method. As
Table 1 indicates, the results of cbires are less than adequate. CBIR performed marginally better after some tweaking. Even at the best settings we could find, however, the
image query results are also less than adequate, as indicated in Table 1.

1

2

As the Caltech data set contains many classes with more than 200 images, while others have as few
as 80, we removed any images beyond 200 for a class to reduce class imbalance.
cbires requires the query image be in the image database, so we had to place the query image in the
database before performing cbires searches.


Image-Based Content Retrieval

7

As opposed to the baseline methods, CBH-IBR produces high quality results, both
visually and via F-measure. Without retraining, the only tuning adjustment for this
system is whether to ignore small values in the histogram vector, and what the
threshold should be for ignoring small values. Empirically we determined that a value
of 0.01 works the best. This setting ignores the noise of any very small probabilities,
improving results dramatically. Interestingly, a large value for the threshold percentage
reduces the quality of the results, indicating that categories of classification with
smaller values significantly improve the engine’s ability to find similar images. Figure 1 shows two excellent results, while the left-hand side of Fig. 2 shows one that is
not wholly adequate.

Fig. 1. Using CBH-IBR to query images of a classic car (left) and a tiger (right).

Retraining the CBH-IBR method involves assigning classes to each image in the
database, and using loss algorithms to determine the quality of the result. Retraining
improves results in many cases, but also creates instances where the results are worse
than the original. Thus we have stacked the three best methods—the original network,
one retrained via softmax log loss, and one via a custom histogram/class method3—and
use lowest error scores from amongst the three methods to generate our 20 results. The
right-hand side of Fig. 2 shows the results of the same query after retraining. Note that
the two images in the second to last row are now images of mandolins, not incorrect
images. Though this stacked network method works the best of any we tested, it still
produces results that are less than perfect for some query images.
From visual examination, we produce F-measures for each search method, and for
both correct and correct + partially correct results. Table 1 shows the results of these
more quantitative measures and the numbers parallel our visual observations.
3

Our method minimizes the derivative of the sum of squared errors between class histograms (term 1)
added to class error, or
2 softmax log loss (term 2).
02
3
13
"
#
images
images
classes
À
Á

P
P
P
o
6
B6 e i 7
C7
Sderivative ¼ dzdy à 4À1:5
2 qj À bij À
@4classes
P o 5 À ci A5
i¼1

j¼1

i¼1

e

k¼1

.

k