International Journal of Computer Science Issues

IJCSI

International Journal of Computer Science Issues

Volume 7, Issue 4, No 2, July 2010 ISSN (Online): 1694-0784 ISSN (Print): 1694-0814

© IJCSI PUBLICATION www.IJCSI.org

IJCSI proceedings are currently indexed by:

© IJCSI PUBLICATION 2010 www.IJCSI.org

IJCSI Publicity Board 2010

Dr. Borislav D Dimitrov Department of General Practice, Royal College of Surgeons in Ireland Dublin, Ireland

Dr. Vishal Goyal Department of Computer Science, Punjabi University Patiala, India

Mr. Nehinbe Joshua University of Essex Colchester, Essex, UK

Mr. Vassilis Papataxiarhis Department of Informatics and Telecommunications National and Kapodistrian University of Athens, Athens, Greece

EDITORIAL In this fourth edition of 2010, we bring forward issues from various dynamic computer science areas ranging from system performance, computer vision, artificial intelligence, ontologies, software engineering, multimedia, pattern recognition, information retrieval, databases, security and networking among others. Considering the growing interest of academics worldwide to publish in IJCSI, we invite universities and institutions to partner with us to further encourage open-access publications. As always we thank all our reviewers for providing constructive comments on papers sent to them for review. This helps enormously in improving the quality of papers published in this issue. Apart from availability of the full-texts from the journal website, all published papers are deposited in open-access repositories to make access easier and ensure continuous availability of its proceedings. We are pleased to present IJCSI Volume 7, Issue 4, July 2010, split in nine numbers (IJCSI Vol. 7, Issue 4, No. 2). Out of the 179 paper submissions, 57 papers were retained for publication. The acceptance rate for this issue is 31.84%.

We wish you a happy reading!

IJCSI Editorial Board July 2010 Issue ISSN (Print): 1694-0814 ISSN (Online): 1694-0784 © IJCSI Publications www.IJCSI.org

IJCSI Editorial Board 2010

Dr Tristan Vanrullen Chief Editor LPL, Laboratoire Parole et Langage - CNRS - Aix en Provence, France LABRI, Laboratoire Bordelais de Recherche en Informatique - INRIA - Bordeaux, France LEEE, Laboratoire d'Esthétique et Expérimentations de l'Espace - Université d'Auvergne, France

Dr Constantino Malagôn Associate Professor Nebrija University Spain

Dr Lamia Fourati Chaari Associate Professor Multimedia and Informatics Higher Institute in SFAX Tunisia

Dr Mokhtar Beldjehem Professor Sainte-Anne University Halifax, NS, Canada

Dr Pascal Chatonnay Assistant Professor MaÎtre de Conférences Laboratoire d'Informatique de l'Université de Franche-Comté Université de Franche-Comté France

Dr Karim Mohammed Rezaul Centre for Applied Internet Research (CAIR) Glyndwr University Wrexham, United Kingdom

Dr Yee-Ming Chen Professor Department of Industrial Engineering and Management Yuan Ze University Taiwan

Dr Vishal Goyal Assistant Professor Department of Computer Science Punjabi University Patiala, India

Dr Dalbir Singh Faculty of Information Science And Technology National University of Malaysia Malaysia

Dr Natarajan Meghanathan Assistant Professor REU Program Director Department of Computer Science Jackson State University Jackson, USA

Dr Deepak Laxmi Narasimha Department of Software Engineering, Faculty of Computer Science and Information Technology, University of Malaya, Kuala Lumpur, Malaysia

Dr Navneet Agrawal Assistant Professor Department of ECE, College of Technology & Engineering, MPUAT, Udaipur 313001 Rajasthan, India

Dr T. V. Prasad Professor Department of Computer Science and Engineering, Lingaya's University Faridabad, Haryana, India

Prof N. Jaisankar Assistant Professor School of Computing Sciences, VIT University Vellore, Tamilnadu, India

IJCSI Reviewers Committee 2010  Mr. Markus Schatten, University of Zagreb, Faculty of Organization and Informatics, Croatia  Mr. Vassilis Papataxiarhis, Department of Informatics and Telecommunications, National and Kapodistrian University of Athens, Athens, Greece  Dr Modestos Stavrakis, University of the Aegean, Greece  Dr Fadi KHALIL, LAAS -- CNRS Laboratory, France  Dr Dimitar Trajanov, Faculty of Electrical Engineering and Information technologies, ss. Cyril and Methodius Univesity - Skopje, Macedonia  Dr Jinping Yuan, College of Information System and Management,National Univ. of Defense Tech., China  Dr Alexis Lazanas, Ministry of Education, Greece  Dr Stavroula Mougiakakou, University of Bern, ARTORG Center for Biomedical Engineering Research, Switzerland  Dr Cyril de Runz, CReSTIC-SIC, IUT de Reims, University of Reims, France  Mr. Pramodkumar P. Gupta, Dept of Bioinformatics, Dr D Y Patil University, India  Dr Alireza Fereidunian, School of ECE, University of Tehran, Iran  Mr. Fred Viezens, Otto-Von-Guericke-University Magdeburg, Germany  Dr. Richard G. Bush, Lawrence Technological University, United States  Dr. Ola Osunkoya, Information Security Architect, USA  Mr. Kotsokostas N.Antonios, TEI Piraeus, Hellas  Prof Steven Totosy de Zepetnek, U of Halle-Wittenberg & Purdue U & National Sun Yat-sen U, Germany, USA, Taiwan  Mr. M Arif Siddiqui, Najran University, Saudi Arabia  Ms. Ilknur Icke, The Graduate Center, City University of New York, USA  Prof Miroslav Baca, Faculty of Organization and Informatics, University of Zagreb, Croatia  Dr. Elvia Ruiz Beltrán, Instituto Tecnológico de Aguascalientes, Mexico  Mr. Moustafa Banbouk, Engineer du Telecom, UAE  Mr. Kevin P. Monaghan, Wayne State University, Detroit, Michigan, USA  Ms. Moira Stephens, University of Sydney, Australia  Ms. Maryam Feily, National Advanced IPv6 Centre of Excellence (NAV6) , Universiti Sains Malaysia (USM), Malaysia  Dr. Constantine YIALOURIS, Informatics Laboratory Agricultural University of Athens, Greece  Mrs. Angeles Abella, U. de Montreal, Canada  Dr. Patrizio Arrigo, CNR ISMAC, italy  Mr. Anirban Mukhopadhyay, B.P.Poddar Institute of Management & Technology, India  Mr. Dinesh Kumar, DAV Institute of Engineering & Technology, India  Mr. Jorge L. Hernandez-Ardieta, INDRA SISTEMAS / University Carlos III of Madrid, Spain  Mr. AliReza Shahrestani, University of Malaya (UM), National Advanced IPv6 Centre of Excellence (NAv6), Malaysia  Mr. Blagoj Ristevski, Faculty of Administration and Information Systems Management - Bitola, Republic of Macedonia  Mr. Mauricio Egidio Cantão, Department of Computer Science / University of São Paulo, Brazil  Mr. Jules Ruis, Fractal Consultancy, The Netherlands

 Mr. Mohammad Iftekhar Husain, University at Buffalo, USA  Dr. Deepak Laxmi Narasimha, Department of Software Engineering, Faculty of Computer Science and Information Technology, University of Malaya, Malaysia  Dr. Paola Di Maio, DMEM University of Strathclyde, UK  Dr. Bhanu Pratap Singh, Institute of Instrumentation Engineering, Kurukshetra University Kurukshetra, India  Mr. Sana Ullah, Inha University, South Korea  Mr. Cornelis Pieter Pieters, Condast, The Netherlands  Dr. Amogh Kavimandan, The MathWorks Inc., USA  Dr. Zhinan Zhou, Samsung Telecommunications America, USA  Mr. Alberto de Santos Sierra, Universidad Politécnica de Madrid, Spain  Dr. Md. Atiqur Rahman Ahad, Department of Applied Physics, Electronics & Communication Engineering (APECE), University of Dhaka, Bangladesh  Dr. Charalampos Bratsas, Lab of Medical Informatics, Medical Faculty, Aristotle University, Thessaloniki, Greece  Ms. Alexia Dini Kounoudes, Cyprus University of Technology, Cyprus  Mr. Anthony Gesase, University of Dar es salaam Computing Centre, Tanzania  Dr. Jorge A. Ruiz-Vanoye, Universidad Juárez Autónoma de Tabasco, Mexico  Dr. Alejandro Fuentes Penna, Universidad Popular Autónoma del Estado de Puebla, México  Dr. Ocotlán Díaz-Parra, Universidad Juárez Autónoma de Tabasco, México  Mrs. Nantia Iakovidou, Aristotle University of Thessaloniki, Greece  Mr. Vinay Chopra, DAV Institute of Engineering & Technology, Jalandhar  Ms. Carmen Lastres, Universidad Politécnica de Madrid - Centre for Smart Environments, Spain  Dr. Sanja Lazarova-Molnar, United Arab Emirates University, UAE  Mr. Srikrishna Nudurumati, Imaging & Printing Group R&D Hub, Hewlett-Packard, India  Dr. Olivier Nocent, CReSTIC/SIC, University of Reims, France  Mr. Burak Cizmeci, Isik University, Turkey  Dr. Carlos Jaime Barrios Hernandez, LIG (Laboratory Of Informatics of Grenoble), France  Mr. Md. Rabiul Islam, Rajshahi university of Engineering & Technology (RUET), Bangladesh  Dr. LAKHOUA Mohamed Najeh, ISSAT - Laboratory of Analysis and Control of Systems, Tunisia  Dr. Alessandro Lavacchi, Department of Chemistry - University of Firenze, Italy  Mr. Mungwe, University of Oldenburg, Germany  Mr. Somnath Tagore, Dr D Y Patil University, India  Ms. Xueqin Wang, ATCS, USA  Dr. Borislav D Dimitrov, Department of General Practice, Royal College of Surgeons in Ireland, Dublin, Ireland  Dr. Fondjo Fotou Franklin, Langston University, USA  Dr. Vishal Goyal, Department of Computer Science, Punjabi University, Patiala, India  Mr. Thomas J. Clancy, ACM, United States  Dr. Ahmed Nabih Zaki Rashed, Dr. in Electronic Engineering, Faculty of Electronic Engineering, menouf 32951, Electronics and Electrical Communication Engineering Department, Menoufia university, EGYPT, EGYPT  Dr. Rushed Kanawati, LIPN, France  Mr. Koteshwar Rao, K G Reddy College Of ENGG.&TECH,CHILKUR, RR DIST.,AP, India

 Mr. M. Nagesh Kumar, Department of Electronics and Communication, J.S.S. research foundation, Mysore University, Mysore-6, India  Dr. Ibrahim Noha, Grenoble Informatics Laboratory, France  Mr. Muhammad Yasir Qadri, University of Essex, UK  Mr. Annadurai .P, KMCPGS, Lawspet, Pondicherry, India, (Aff. Pondicherry Univeristy, India  Mr. E Munivel , CEDTI (Govt. of India), India  Dr. Chitra Ganesh Desai, University of Pune, India  Mr. Syed, Analytical Services & Materials, Inc., USA  Dr. Mashud Kabir, Department of Computer Science, University of Tuebingen, Germany  Mrs. Payal N. Raj, Veer South Gujarat University, India  Mrs. Priti Maheshwary, Maulana Azad National Institute of Technology, Bhopal, India  Mr. Mahesh Goyani, S.P. University, India, India  Mr. Vinay Verma, Defence Avionics Research Establishment, DRDO, India  Dr. George A. Papakostas, Democritus University of Thrace, Greece  Mr. Abhijit Sanjiv Kulkarni, DARE, DRDO, India  Mr. Kavi Kumar Khedo, University of Mauritius, Mauritius  Dr. B. Sivaselvan, Indian Institute of Information Technology, Design & Manufacturing, Kancheepuram, IIT Madras Campus, India  Dr. Partha Pratim Bhattacharya, Greater Kolkata College of Engineering and Management, West Bengal University of Technology, India  Mr. Manish Maheshwari, Makhanlal C University of Journalism & Communication, India  Dr. Siddhartha Kumar Khaitan, Iowa State University, USA  Dr. Mandhapati Raju, General Motors Inc, USA  Dr. M.Iqbal Saripan, Universiti Putra Malaysia, Malaysia  Mr. Ahmad Shukri Mohd Noor, University Malaysia Terengganu, Malaysia  Mr. Selvakuberan K, TATA Consultancy Services, India  Dr. Smita Rajpal, Institute of Technology and Management, Gurgaon, India  Mr. Rakesh Kachroo, Tata Consultancy Services, India  Mr. Raman Kumar, National Institute of Technology, Jalandhar, Punjab., India  Mr. Nitesh Sureja, S.P.University, India  Dr. M. Emre Celebi, Louisiana State University, Shreveport, USA  Dr. Aung Kyaw Oo, Defence Services Academy, Myanmar  Mr. Sanjay P. Patel, Sankalchand Patel College of Engineering, Visnagar, Gujarat, India  Dr. Pascal Fallavollita, Queens University, Canada  Mr. Jitendra Agrawal, Rajiv Gandhi Technological University, Bhopal, MP, India  Mr. Ismael Rafael Ponce Medellín, Cenidet (Centro Nacional de Investigación y Desarrollo Tecnológico), Mexico  Mr. Supheakmungkol SARIN, Waseda University, Japan  Mr. Shoukat Ullah, Govt. Post Graduate College Bannu, Pakistan  Dr. Vivian Augustine, Telecom Zimbabwe, Zimbabwe  Mrs. Mutalli Vatila, Offshore Business Philipines, Philipines  Dr. Emanuele Goldoni, University of Pavia, Dept. of Electronics, TLC & Networking Lab, Italy  Mr. Pankaj Kumar, SAMA, India  Dr. Himanshu Aggarwal, Punjabi University,Patiala, India  Dr. Vauvert Guillaume, Europages, France

 Prof Yee Ming Chen, Department of Industrial Engineering and Management, Yuan Ze University, Taiwan  Dr. Constantino Malagón, Nebrija University, Spain  Prof Kanwalvir Singh Dhindsa, B.B.S.B.Engg.College, Fatehgarh Sahib (Punjab), India  Mr. Angkoon Phinyomark, Prince of Singkla University, Thailand  Ms. Nital H. Mistry, Veer Narmad South Gujarat University, Surat, India  Dr. M.R.Sumalatha, Anna University, India  Mr. Somesh Kumar Dewangan, Disha Institute of Management and Technology, India  Mr. Raman Maini, Punjabi University, Patiala(Punjab)-147002, India  Dr. Abdelkader Outtagarts, Alcatel-Lucent Bell-Labs, France  Prof Dr. Abdul Wahid, AKG Engg. College, Ghaziabad, India  Mr. Prabu Mohandas, Anna University/Adhiyamaan College of Engineering, india  Dr. Manish Kumar Jindal, Panjab University Regional Centre, Muktsar, India  Prof Mydhili K Nair, M S Ramaiah Institute of Technnology, Bangalore, India  Dr. C. Suresh Gnana Dhas, VelTech MultiTech Dr.Rangarajan Dr.Sagunthala Engineering College,Chennai,Tamilnadu, India  Prof Akash Rajak, Krishna Institute of Engineering and Technology, Ghaziabad, India  Mr. Ajay Kumar Shrivastava, Krishna Institute of Engineering & Technology, Ghaziabad, India  Mr. Deo Prakash, SMVD University, Kakryal(J&K), India  Dr. Vu Thanh Nguyen, University of Information Technology HoChiMinh City, VietNam  Prof Deo Prakash, SMVD University (A Technical University open on I.I.T. Pattern) Kakryal (J&K), India  Dr. Navneet Agrawal, Dept. of ECE, College of Technology & Engineering, MPUAT, Udaipur 313001 Rajasthan, India  Mr. Sufal Das, Sikkim Manipal Institute of Technology, India  Mr. Anil Kumar, Sikkim Manipal Institute of Technology, India  Dr. B. Prasanalakshmi, King Saud University, Saudi Arabia.  Dr. K D Verma, S.V. (P.G.) College, Aligarh, India  Mr. Mohd Nazri Ismail, System and Networking Department, University of Kuala Lumpur (UniKL), Malaysia  Dr. Nguyen Tuan Dang, University of Information Technology, Vietnam National University Ho Chi Minh city, Vietnam  Dr. Abdul Aziz, University of Central Punjab, Pakistan  Dr. P. Vasudeva Reddy, Andhra University, India  Mrs. Savvas A. Chatzichristofis, Democritus University of Thrace, Greece  Mr. Marcio Dorn, Federal University of Rio Grande do Sul - UFRGS Institute of Informatics, Brazil  Mr. Luca Mazzola, University of Lugano, Switzerland  Mr. Nadeem Mahmood, Department of Computer Science, University of Karachi, Pakistan  Mr. Hafeez Ullah Amin, Kohat University of Science & Technology, Pakistan  Dr. Professor Vikram Singh, Ch. Devi Lal University, Sirsa (Haryana), India  Mr. M. Azath, Calicut/Mets School of Enginerring, India  Dr. J. Hanumanthappa, DoS in CS, University of Mysore, India  Dr. Shahanawaj Ahamad, Department of Computer Science, King Saud University, Saudi Arabia  Dr. K. Duraiswamy, K. S. Rangasamy College of Technology, India  Prof. Dr Mazlina Esa, Universiti Teknologi Malaysia, Malaysia

 Dr. P. Vasant, Power Control Optimization (Global), Malaysia  Dr. Taner Tuncer, Firat University, Turkey  Dr. Norrozila Sulaiman, University Malaysia Pahang, Malaysia  Prof. S K Gupta, BCET, Guradspur, India  Dr. Latha Parameswaran, Amrita Vishwa Vidyapeetham, India  Mr. M. Azath, Anna University, India  Dr. P. Suresh Varma, Adikavi Nannaya University, India  Prof. V. N. Kamalesh, JSS Academy of Technical Education, India  Dr. D Gunaseelan, Ibri College of Technology, Oman  Mr. Sanjay Kumar Anand, CDAC, India  Mr. Akshat Verma, CDAC, India  Mrs. Fazeela Tunnisa, Najran University, Kingdom of Saudi Arabia  Mr. Hasan Asil, Islamic Azad University Tabriz Branch (Azarshahr), Iran  Prof. Dr Sajal Kabiraj, Fr. C Rodrigues Institute of Management Studies (Affiliated to University of Mumbai, India), India  Mr. Syed Fawad Mustafa, GAC Center, Shandong University, China  Dr. Natarajan Meghanathan, Jackson State University, Jackson, MS, USA  Prof. Selvakani Kandeeban, Francis Xavier Engineering College, India  Mr. Tohid Sedghi, Urmia University, Iran  Dr. S. Sasikumar, PSNA College of Engg and Tech, Dindigul, India  Dr. Anupam Shukla, Indian Institute of Information Technology and Management Gwalior, India  Mr. Rahul Kala, Indian Institute of Inforamtion Technology and Management Gwalior, India  Dr. A V Nikolov, National University of Lesotho, Lesotho  Mr. Kamal Sarkar, Department of Computer Science and Engineering, Jadavpur University, India  Dr. Mokhled S. AlTarawneh, Computer Engineering Dept., Faculty of Engineering, Mutah University, Jordan, Jordan  Prof. Sattar J Aboud, Iraqi Council of Representatives, Iraq-Baghdad  Dr. Prasant Kumar Pattnaik, Department of CSE, KIST, India  Dr. Mohammed Amoon, King Saud University, Saudi Arabia  Dr. Tsvetanka Georgieva, Department of Information Technologies, St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria  Dr. Eva Volna, University of Ostrava, Czech Republic  Mr. Ujjal Marjit, University of Kalyani, West-Bengal, India  Dr. Prasant Kumar Pattnaik, KIST,Bhubaneswar,India, India  Dr. Guezouri Mustapha, Department of Electronics, Faculty of Electrical Engineering, University of Science and Technology (USTO), Oran, Algeria  Mr. Maniyar Shiraz Ahmed, Najran University, Najran, Saudi Arabia  Dr. Sreedhar Reddy, JNTU, SSIETW, Hyderabad, India  Mr. Bala Dhandayuthapani Veerasamy, Mekelle University, Ethiopa  Mr. Arash Habibi Lashkari, University of Malaya (UM), Malaysia  Mr. Rajesh Prasad, LDC Institute of Technical Studies, Allahabad, India  Ms. Habib Izadkhah, Tabriz University, Iran  Dr. Lokesh Kumar Sharma, Chhattisgarh Swami Vivekanand Technical University Bhilai, India  Mr. Kuldeep Yadav, IIIT Delhi, India  Dr. Naoufel Kraiem, Institut Superieur d'Informatique, Tunisia

 Prof. Frank Ortmeier, Otto-von-Guericke-Universitaet Magdeburg, Germany  Mr. Ashraf Aljammal, USM, Malaysia  Mrs. Amandeep Kaur, Department of Computer Science, Punjabi University, Patiala, Punjab, India  Mr. Babak Basharirad, University Technology of Malaysia, Malaysia  Mr. Avinash singh, Kiet Ghaziabad, India  Dr. Miguel Vargas-Lombardo, Technological University of Panama, Panama  Dr. Tuncay Sevindik, Firat University, Turkey  Ms. Pavai Kandavelu, Anna University Chennai, India  Mr. Ravish Khichar, Global Institute of Technology, India  Mr Aos Alaa Zaidan Ansaef, Multimedia University, Cyberjaya, Malaysia  Dr. Awadhesh Kumar Sharma, Dept. of CSE, MMM Engg College, Gorakhpur-273010, UP, India  Mr. Qasim Siddique, FUIEMS, Pakistan  Dr. Le Hoang Thai, University of Science, Vietnam National University - Ho Chi Minh City, Vietnam  Dr. Saravanan C, NIT, Durgapur, India  Dr. Vijay Kumar Mago, DAV College, Jalandhar, India  Dr. Do Van Nhon, University of Information Technology, Vietnam  Mr. Georgios Kioumourtzis, University of Patras, Greece  Mr. Amol D.Potgantwar, SITRC Nasik, India  Mr. Lesedi Melton Masisi, Council for Scientific and Industrial Research, South Africa  Dr. Karthik.S, Department of Computer Science & Engineering, SNS College of Technology, India  Mr. Nafiz Imtiaz Bin Hamid, Department of Electrical and Electronic Engineering, Islamic University of Technology (IUT), Bangladesh  Mr. Muhammad Imran Khan, Universiti Teknologi PETRONAS, Malaysia  Dr. Abdul Kareem M. Radhi, Information Engineering - Nahrin University, Iraq  Dr. Mohd Nazri Ismail, University of Kuala Lumpur, Malaysia  Dr. Manuj Darbari, BBDNITM, Institute of Technology, A-649, Indira Nagar, Lucknow 226016, India  Ms. Izerrouken, INP-IRIT, France  Mr. Nitin Ashokrao Naik, Dept. of Computer Science, Yeshwant Mahavidyalaya, Nanded, India  Mr. Nikhil Raj, National Institute of Technology, Kurukshetra, India  Prof. Maher Ben Jemaa, National School of Engineers of Sfax, Tunisia  Prof. Rajeshwar Singh, BRCM College of Engineering and Technology, Bahal Bhiwani, Haryana, India  Mr. Gaurav Kumar, Department of Computer Applications, Chitkara Institute of Engineering and Technology, Rajpura, Punjab, India  Mr. Ajeet Kumar Pandey, Indian Institute of Technology, Kharagpur, India  Mr. Rajiv Phougat, IBM Corporation, USA  Mrs. Aysha V, College of Applied Science Pattuvam affiliated with Kannur University, India  Dr. Debotosh Bhattacharjee, Department of Computer Science and Engineering, Jadavpur University, Kolkata-700032, India  Dr. Neelam Srivastava, Institute of engineering & Technology, Lucknow, India  Prof. Sweta Verma, Galgotia's College of Engineering & Technology, Greater Noida, India  Mr. Harminder Singh BIndra, MIMIT, INDIA  Dr. Lokesh Kumar Sharma, Chhattisgarh Swami Vivekanand Technical University, Bhilai, India  Mr. Tarun Kumar, U.P. Technical University/Radha Govinend Engg. College, India  Mr. Tirthraj Rai, Jawahar Lal Nehru University, New Delhi, India

 Mr. Akhilesh Tiwari, Madhav Institute of Technology & Science, India  Mr. Dakshina Ranjan Kisku, Dr. B. C. Roy Engineering College, WBUT, India  Ms. Anu Suneja, Maharshi Markandeshwar University, Mullana, Haryana, India  Mr. Munish Kumar Jindal, Punjabi University Regional Centre, Jaito (Faridkot), India  Dr. Ashraf Bany Mohammed, Management Information Systems Department, Faculty of Administrative and Financial Sciences, Petra University, Jordan  Mrs. Jyoti Jain, R.G.P.V. Bhopal, India  Dr. Lamia Chaari, SFAX University, Tunisia  Mr. Akhter Raza Syed, Department of Computer Science, University of Karachi, Pakistan  Prof. Khubaib Ahmed Qureshi, Information Technology Department, HIMS, Hamdard University, Pakistan  Prof. Boubker Sbihi, Ecole des Sciences de L'Information, Morocco  Dr. S. M. Riazul Islam, Inha University, South Korea  Prof. Lokhande S.N., S.R.T.M.University, Nanded (MH), India  Dr. Vijay H Mankar, Dept. of Electronics, Govt. Polytechnic, Nagpur, India  Dr. M. Sreedhar Reddy, JNTU, Hyderabad, SSIETW, India  Mr. Ojesanmi Olusegun, Ajayi Crowther University, Oyo, Nigeria  Ms. Mamta Juneja, RBIEBT, PTU, India  Dr. Ekta Walia Bhullar, Maharishi Markandeshwar University, Mullana Ambala (Haryana), India  Prof. Chandra Mohan, John Bosco Engineering College, India  Mr. Nitin A. Naik, Yeshwant Mahavidyalaya, Nanded, India  Mr. Sunil Kashibarao Nayak, Bahirji Smarak Mahavidyalaya, Basmathnagar Dist-Hingoli., India  Prof. Rakesh.L, Vijetha Institute of Technology, Bangalore, India  Mr B. M. Patil, Indian Institute of Technology, Roorkee, Uttarakhand, India  Mr. Thipendra Pal Singh, Sharda University, K.P. III, Greater Noida, Uttar Pradesh, India  Prof. Chandra Mohan, John Bosco Engg College, India  Mr. Hadi Saboohi, University of Malaya - Faculty of Computer Science and Information Technology, Malaysia  Dr. R. Baskaran, Anna University, India  Dr. Wichian Sittiprapaporn, Mahasarakham University College of Music, Thailand  Mr. Lai Khin Wee, Universiti Teknologi Malaysia, Malaysia  Dr. Kamaljit I. Lakhtaria, Atmiya Institute of Technology, India  Mrs. Inderpreet Kaur, PTU, Jalandhar, India  Mr. Iqbaldeep Kaur, PTU / RBIEBT, India  Mrs. Vasudha Bahl, Maharaja Agrasen Institute of Technology, Delhi, India  Prof. Vinay Uttamrao Kale, P.R.M. Institute of Technology & Research, Badnera, Amravati, Maharashtra, India  Mr. Suhas J Manangi, Microsoft, India  Ms. Anna Kuzio, Adam Mickiewicz University, School of English, Poland  Dr. Debojyoti Mitra, Sir Padampat Singhania University, India  Prof. Rachit Garg, Department of Computer Science, L K College, India  Mrs. Manjula K A, Kannur University, India  Mr. Rakesh Kumar, Indian Institute of Technology Roorkee, India

TABLE OF CONTENTS

1. A New Public-Key Encryption Scheme Based on Non-Expansion Visual Cryptography and Boolean Operation Abdullah M. Jaafar and Azman Samsudin

Pg 1-10

2. Architecture for Automated Tagging and Clustering of Song Files According to Mood Puneet Singh, Ashutosh Kapoor, Vishal Kaushik and Hima Bindu Maringanti

Pg 11-17

3. An Improved k-Nearest Neighbor Classification Using Genetic Algorithm N. Suguna and K. Thanushkodi

Pg 18-21

4. Empirical Evaluation of Suitable Segmentation Algorithms for IR Images G. Padmavathi, P. Subashini and A. Sumi

Pg 22-29

5. Numerical Analysis of the DQPSK Modulation Formats Implementation With 40 Gbits/s Hadjira Badaoui, Yann Frignac, Petros Ramantanis, Badr Eddine Benkelfat and Mohammed Feham

Pg 30-36

6. N-Dimensional Self Organizing Petrinets for Urban Traffic Modeling Manuj Darbari, Vivek Kumar Singh, Rishi Asthana and Savitur Prakash

Pg 37-40

7. Domain Driven Data Mining – Application to Business Adeyemi Adejuwon and Amir Mosavi

Pg 41-44

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 ISSN (Online): 1694-0784 ISSN (Print): 1694-0814

1

A New Public-Key Encryption Scheme Based on Non-Expansion Visual Cryptography and Boolean Operation Abdullah M. Jaafar and Azman Samsudin School of Computer Sciences, Universiti Sains Malaysia 11800 Penang, Malaysia

Abstract Currently, most of the existing public-key encryption schemes are based on complex algorithms with heavy computations. In 1994, Naor and Shamir proposed a simple cryptography method for digital images called visual cryptography. Existing visual cryptography primitives can be considered as a special kind of secret-key cryptography that does not require heavy computations for encrypting and decrypting an image. In this paper, we propose a new public-key encryption scheme for image based on non-expansion visual cryptography and Boolean operation. The proposed scheme uses only Boolean operations and therefore requires comparatively lower computations. Keywords: Public-key encryption, V isual cryp tography, P ixel expansion, Boolean operation.

1. Introduction Information security in the present era is becoming very important in communication and data storage. Data transferred from one party to another over an insecure channel (e.g., Internet) can be protected by cryptography. The encrypting technologies of traditional and modern cryptography are usually used to avoid the message from being disclosed [1-5]. Public-key cryptography usually uses complex mathematical computations to scramble the message [1, 2, 6]. In 1976, Diffie and Hellman [7] introduced the first concept of public-key (asymmetric-key) cryptography to solve the key exchange problem. Public-key cryptography is one of the greatest contributions in the history of cryptography. Nowadays, public-key cryptography is practically utilized in everyday life to attain privacy, authenticity, integrity, and non-repudiation [6, 8, 9]. One of the main branches and applications of the public-key cryptography is a public-key encryption scheme which allows two parties to communicate securely over an insecure channel without having prior knowledge of each other to establish a shared secret key. Unlike secret-key (symmetric-key) encryption scheme, public-key encryption scheme does not use the same key to encrypt and decrypt a message. Instead, each one of the two parties has two different keys but related mathematically, the

public key known to everyone and the private key known only to receiver of the message. There are some popular public-key encryption algorithms, for example, RSA, ElGamal, and ECC. The security of the most public-key encryption algorithms is based on discrete logarithms in finite groups or integer factorization [6, 9, 10]. However, with all benefits and advantages of public-key encryption schemes, these schemes require a great deal of complex computations to generate the keys and to encrypt and decrypt confidential information; hence computing devices (computers) are fundamental for generation of the keys and for both encryption and decryption [1]. Conducting such computations without the assistance of a computing device (e.g., computer) is a difficult task, if not impossible [11]. Under this situation, we propose a new public-key encryption scheme with easy encryption/decryption algorithms and a comparatively low computation complexity [12]. In 1994, Naor and Shamir [13] suggested an emerging cryptography method, namely, visual cryptography (VC), which is very easy to use and perfectly safe. The encryption process is performed by simple and low computational device, whereas the decryption process is performed directly by human visual system without any complex computations. VC can be used where computers are scarce or access to them is not possible [2, 13-18]. VC divides a secret image into n transparent shares and it uses the human visual system to recover the secret image by superimposing and aligning carefully all or some of n transparent shares according to visual cryptography scheme used [2]. Because Boolean AND operation is simple, quick, and very adaptive to be implemented to an image cryptography system, the proposed public-key encryption scheme in this paper is also based on this operation. In this paper, a novel public-key encryption scheme is proposed, which is based on non-expansion visual cryptography and Boolean operation, to overcome the problem of complex computations as in most of the existing public-key encryption schemes. The proposed scheme begins to establish a shared visual secret key between two communicating parties such as Alice and

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org Bob. After that, this shared visual secret key will be used in our proposed scheme during encryption and decryption processes. The remaining of this paper is as follows. Section 2 gives a brief background on the conventional public-key encryption scheme and the visual cryptography. Section 3 describes and explains our proposed method. We discuss the security analysis and computational complexity of our proposed method in Section 4. Section 5 gives the experimental results. Finally, the conclusion is given in Section 6.

2. Background 2.1 Conventional public-key encryption scheme In the process of encryption, the sender encrypts his or her confidential information in such a way that only the intended recipient can decrypt the confidential information. Using public-key encryption mechanism all communications between two parties over an unprotected channel involve only public keys, and without need for exchanging any private (secret) key [3]. Public-key encryption depends on a different two keys but mathematically linked, the first key is the public key which is put in a public, used for encryption; and the second key is the private (secret) key which is kept secret, used for decryption. In addition, it is computationally difficult to derive the private key from the public key [6, 10, 19, 20]. Fig. 1 shows the concept of public-key encryption technique. For this system, suppose that the receiver, Bob, has private key and public key are (PRR) and (PUR) respectively. Receiver’s public key (PUR) is publicly known, used for encryption; and receiver’s private key (PRR) is kept secret, used for decryption. Suppose that the sender, Alice, wants to send an original secret message (SM) to the receiver (Bob), Alice (the sender) will encrypt her secret message (SM) using Bob’s public key (PUR) to get the encrypted secret message, which is known as cipher message (C), and sends it to Bob (the receiver). The receiver (Bob) can decrypt the cipher message (C) by using only his private key (PRR).

The RSA public-key encryption algorithm [21] is the first practical encryption scheme based on the concept of the public-key cryptography. There are many public-key encryption algorithms published after the RSA public-key encryption, such as ElGamal public-key encryption [22] and elliptic curve public-key encryption [23] and others.

2.2 Visual cryptography Visual Cryptography is a special kind of cryptography which encrypts visual information (i.e., pictures, printed text, handwritten notes, etc.) into n transparent images (shares) so that humans can perform the decryption visually without the assistance of computers [13, 24]. Each one of n transparent images is an indistinguishable from random noise, thus they can be transmitted or distributed over an unprotected communication channel (i.e., Internet). In other words, because the shares appear as random binary patterns, the attackers cannot sense any hints about a secret image from individual shares [2]. The secret information can be decrypted from the shares directly by the human visual system when all or any majority of the shares are stacked together so that the subpixels are carefully aligned. On the other hand, any minority number of stacked shares or every share individually cannot leak any hint about the secret information, even if computers are available [13, 25, 26]. The basic model of visual cryptography, addressed by Naor and Shamir, divides every pixel in an original image into 2×2 black and white subpixels in the two shares on the basis of the rules in Table 1. As in Table 1, a white pixel is encrypted into two identical blocks in the two shares, and a black pixel is encrypted into two complementary blocks in the two shares. Each block is 2×2 black and white subpixels [13, 26, 27]. Take Fig. 2 for example. It shows the result obtained according to Table 1. The original secret image (a) is encrypted into two transparent shares (b) and (c). We can get the recovered secret image (d) when these two transparent shares (b) and (c) are superimposed together and carefully aligned. Table 1: Naor and Shamir’s (2, 2) visual cryptography scheme of black and white pixels (adapted from [2, 11, 13, 27-30])

Pixel of the secret image Sender (Alice)

Receiver (Bob)

Share 1

Pub lic Key = PU R Private Key = PRR Bob sen ds his pub lic key (PU R) to Alice Secret message (SM)

Share 2 Decry ption algorithm

En cryption algorithm Alice sends the cip her message (C) to Bob

C = Encrypt (SM, PU R)

SM = Decrypt (C, PRR)

Fig. 1 The concept of public-key encryption technique

Secret message (SM)

2

Stacked results (Share 1+ Share 2)

White pixel

Black pixel

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org

(a)

(b)

3

of the two shares (without pixel expansion). The secret image is recovered by stacking and aligning carefully the pixels of the two shares, where every pixel in share 1 is superimposed on the corresponding pixel in share 2; this is performed through the OR operation on the two transparent shares [32]. Fig 3 shows the result obtained according to Table 2. In the following section, ProbVC models will be used to construct our proposed public-key encryption scheme. Table 2: Ito et al.’s (2, 2) ProbVC scheme of black and white pixels (adapted from [18, 32])

Pixel of the secret image

Share 1

Share 2

Recovered results

Probability 0.5 0.5 0.5

(d)

(c)

0.5

Fig. 2 The concept of Naor and Shamir’s (2, 2) visual cryptography scheme with four subpixels: (a) The original secret image, (b) The first share, (c) The second share, (d) The recovered image by superimposing (b) and (c)

Most studies of visual cryptography schemes are based on the technique of pixel expansion; therefore, the resultant shares of encrypted secret image by this technique are expanded several times of the original size thereby causing many problems such as image distortion, use of more memory space, and difficulty in carrying shares. To overcome the above-mentioned problems, Ito et al. [18] and Yang [31] proposed non-expansion visual cryptography or so-called probabilistic visual cryptography (ProbVC) model for black and white images, namely, they merged the conventional visual cryptography with the concept of the probability and without pixel expansion. In their models the sizes of the original image, shares (shadow images), and the recovered image are the same. Each pixel in the original secret image is represented as a black or white pixel in the shares and the original secret image can be distinguished by superimposing these shares together. ProbVC models directly use the ready-made two n×n Boolean basis matrices S0 and S1 to generate the shares. To encrypt a pixel from the secret image in ProbVC models, one randomly selects a column in S0 or S1 according to the color of the pixel (white or black), and assigns i-th row of the selected column to i-th share (corresponding share). Ito | to | et al. [18] defined a new parameter represent the contrast of the recovered image, where and are the probabilities with which a black pixel on the recovered image is created from a white and black pixel on the secret image, respectively. Table 2 shows Ito et al.’s (2, 2) ProbVC scheme that a pixel on a black and white secret image is mapped into a corresponding pixel in each

(a)

(b)

(d)

(c)

Fig. 3 An example of (2, 2) ProbVC scheme without pixel expansion: (a) The original secret image, (b) The first share, (c) The second share, (d) The recovered image by superimposing (b) and (c)

3. The proposed method In this section, we propose a new approach to public-key encryption scheme based on a non-expansion visual cryptography and Boolean AND operation. Instead of generating and computing large and long random integer values as in the most of the existing public-key encryption schemes, our scheme generates shadow images (shares) and manipulates them by using simple Boolean OR and

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org AND operations. The Boolean OR operation can be performed by the human visual system on the transparent shares (i.e., superimposing the shares) as shown in [18, 31], whereas the Boolean AND operation can be performed by any simple low-end computer. Table 3 gives the truth tables of the OR and the AND operations for binary inputs. Table 3: The truth tables of OR and AND Boolean operations for binary inputs

0 1

1 1

0 0

, , ,

1, … , 1, … ,

; ;

1, … , 1, … ,

. .

The expression means that the ij-th element cij , where aij and bij are the of matrix C is equal to ij-th elements of matrix A and matrix B, respectively. Similarly, the expression means that the ij-th , where aij element dij of matrix D is equal to and bij are the ij-th elements of matrix A and matrix B, respectively. The proposed scheme also involves the binary inner binary matrices, denoted product of two as which is computed by using simple Boolean OR and AND operations as follows:

1

1, … ,

;

Description

G PU

An integer number with 2 A visual public share (common shadow image) A black and white secret image selected by the first party (Alice) for generating G visual private keys A black and white secret image selected by the second party (Bob) for generating G visual private keys First party’s visual private keys, where 1, … , 2 Second party’s visual private keys, where 1, … , 2 an inverse matrix (key) of second party’s visual private key PRBG+1 First party’s first visual public key First party’s second visual public key Second party’s first visual public key Second party’s second visual public key First party’s first intermediate shares in the first stage of the construction procedure, where 1, … , Second party’s first intermediate shares in the first stage of the construction procedure, where 1, … , First party’s second intermediate shares in the first stage of the construction procedure, where 1, … , Second party’s second intermediate shares in the first stage of the construction procedure, where 1, … , First party’s first intermediate shares in the second stage of the construction procedure, where 1, … , Second party’s first intermediate shares in the second stage of the construction procedure, where 1, … , First party’s second intermediate shares in the second stage of the construction procedure, where 1, … , Second party’s second intermediate shares in the second stage of the construction procedure, where 1, … , First party’s third intermediate share in the second stage of the construction procedure Second party’s third intermediate share in the second stage of the construction procedure First party’s shared visual secret key Second party’s shared visual secret key Ciphered image

IMA IMB

PUA1 PUA2 PUB1 PUB2 CAi CBi DAj DBj EAi EBi

,

where

Notation

PRBi

The OR and the AND of two binary matrices can be described by the following formulas: ,

Table 4: The notations

PRAi

0 1

1, … ,

.

The expression means that the ij-th element eij of matrix E is equal to , where a and b are the binary elements of matrix A and matrix B, respectively. The proposed scheme consists of the notations used, the shared visual secret key generation phase, the encryption phase, the decryption phase, and brief comparison with the conventional public-key encryption schemes. The proposed scheme will begin to establish a shared visual secret key between two communicating parties such as Alice and Bob. Next, we will use this shared visual secret key at encryption and decryption phases.

3.1 The notations Table 4 summarizes notations used in this paper.

4

FAj FBj HA HB SVSKA SVSKB CI

3.2 Shared visual secret key generation phase This phase consists of the initialization and construction procedure as follows.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org A) Initialization  The first party (Alice) and the second party (Bob) agree on a public integer G greater than or equal to 2 and a visual public key PU in the form of n×n pixels, where PU is a non-invertible and a non-identity matrix.  The first party (Alice) chooses secretly a black and white image IMA with size n×n pixels, and uses (k, k) ProbVC scheme as shown in [18, 31] (here k is equal to G) to encrypt IMA into G visual private keys, ,…, , where each one of them is in the form of n×n pixels.  The first party (Alice) builds the visual private key PRAG+1 in the form of n×n pixels, where PRAG+1 is not an identity matrix.  The second party (Bob) chooses secretly a black and white image IMB with size n×n pixels, and uses (k, k) ProbVC scheme as shown in [18, 31] (here k is equal to G) to encrypt IMB into G visual private keys, ,…, , where each one of them is in the form of n×n pixels.  The second party (Bob) builds the visual private key , where each one of PRBG+1 and its inverse them in the form of n×n pixels and PRBG+1 is not an identity matrix. B) Construction procedure: It consists of the following two stages: (a) First stage of the construction procedure: First party (Alice) produces her first visual public key PUA1 and the second party (Bob) produces his two visual public keys PUB1 and PUB2. Below are the details of the first stage of the construction procedure, which is performed simultaneously by Alice and Bob. First stage (generated by first party, Alice): Step 1: Generate the first visual public key PUA1 as follows. First, construct the first intermediate shares ,…, of G as follows: 1, … ,

(1)

Second, construct the second intermediate ,…, of G as follows: shares

5

Third, superimpose the second intermediate ,…, of G for getting the shares first visual public key PUA1 as follows: (3) Step 2: Send PUA1 to the second party (Bob). First stage (generated by second party, Bob): Step 1: Generate the first visual public key PUB1 as follows. First, construct the first intermediate shares ,…, of G as follows: 1, … ,

(4)

Second, construct the second intermediate ,…, of G as follows: shares 1, … ,

(5)

Third, superimpose the second intermediate ,…, of G for getting the first shares visual public key PUB1 as follows: (6) Step 2: Compute the second visual public key PUB2 as follows. (7) Note that PUB2 is the Boolean product of Bob’s visual private matrix PRBG+1 and the visual public matrix PU. Step 3: Send PUB1 and PUB2 to the first party (Alice). (b) Second stage of the construction procedure: The first party (Alice) generates the shared visual secret key SVSKA. The second party (Bob) generates the second visual public key PUB2 and then generates the shared visual secret key SVSKB, (note that SVSKA=SVSKB). Below are the details of the second stage of the construction procedure for the first party and the second party. Second stage (generated by first party, Alice):

1, … ,

(2) Step 3: Receive the second party’s (Bob’s) visual public keys PUB1 and PUB2.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org Step 4: Compute the second visual public key PUA2 as follows. (8) Note that PUA2 is the Boolean product of Bob’s second visual public matrix PUB2 and Alice’s visual private matrix PRAG+1. Step 5: Send PUA2 to the second party (Bob). Step 6: Generate the shared visual secret key SVSKA as follows.

6

1, … ,

(14)


(15)

Third, superimpose the second intermediate ,…, of G for getting the shares third visual intermediate share HB as follows: (16)

First, construct the first intermediate shares ,…, of G as follows: 1, … ,

(9)

Fourth, compute the visual private key PRBG+2 as follows: (17)


(10)

Third, superimpose the second intermediate ,…, of G for getting the shares third intermediate share HA as follows: (11)

Note that PRBG+2 is the Boolean product of the inverse of the visual private key PRBG+1, , and Alice’s second visual namely, public key PUA2. Fifth, superimpose the third intermediate share HB and the visual private key PRBG+2 for getting the shared visual secret key SVSKB:

Fourth, compute the visual private key PRAG+2 as follows: (12) Note that PRAG+2 is the Boolean product of the visual public matrix PU and Alice’s visual private matrix PRAG+1. Fifth, superimpose the third intermediate share HA and the visual private key PRAG+2 for getting the shared visual secret key SVSKA: (13)

(18)

3.3 Encryption phase Suppose that one of the two parties has a black and white secret image and he/she wants to send it to another party. Because the first party’s shared visual secret key SVSKA is equal to the second party’s shared visual secret key SVSKB (i.e., SVSK = SV SKA = SVSKB), the shared visual secret key SVSK can serve as an encryption key in the sender party and as a decryption key in the receiver party. Suppose that the first party (such as Alice) has a black and white secret image SI with size n×n pixels and she wants to send it to the second party (such as Bob). The sender (Alice) must do the following steps:

Second stage (generated by second party, Bob): Step 4: Receive the first party’s (Alice’s) visual public keys PUA1 and PUA2. Step 5: Generate the shared visual secret key SVSKB as follows. First, construct the first intermediate shares ,…, of G as follows:

Step 1: Use the (2, 2) ProbVC scheme as shown in Subsection 2.2, for encoding (encryption) the secret image SI into two shares (shadow images), where each share is in the form n×n pixels. The first share should be equal to the shared visual secret key SVSKA (which serves as an encryption key) which has been established previously in the shared visual secret key generation phase and the

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org second share will be the ciphered image CI which computes from the original secret image SI and the shared visual secret key SVSKA as shown in the following formula. , P

VC

(19)

Step 2: Send the ciphered image CI to the receiver, Bob.

Our proposed scheme has some advantages and benefits compared with conventional public-key encryption schemes. Table 5 gives a summary of the comparison. Table 5: Brief comparison between conventional public-key encryption schemes and the proposed scheme

Algorithm

Requirement

RSA

Modular exponentiation arithmetic Simple Boolean arithmetic

ElGamal The proposed scheme

3.3 Decryption phase

7

Secret information

Complex computation

Numbers in finite fields

High

Shadow images

Low

The receiver (Bob) must do the following steps: Step 1: Receive the ciphered image CI from the sender, Alice.

4. Security analysis and computational complexity

Step 2: Superimpose the shared visual secret key SVSKB (which serves as a decryption key) and the ciphered image CI, align them carefully, for recovering the secret image SI as follows.

4.1 Security analysis

(20) Note that the recovered secret image SI is in the form of n×n pixels. Fig. 4 gives the basic idea of the proposed public-key encryption scheme. The first party (Alice)

Open network

The second party (Bob)

Alice and Bob agree on an integer G and a visual public key PU

PUA1

Shared Visual Secret Key Generation Algorithm

PUA2 PUB1

Shared Visual Secret Key Generation Algorithm

PUB2

SVSKA

Encryption key

SVSKB

Decryption key

Ciphered image (2, 2) ProbVC Scheme (Encryption)

SI

Secret image

CI

Superposition of the two shares (Decryption)

SI

Recovered image

Fig. 4 The basic idea of the proposed public-key encryption scheme

3.4 Comparison with the conventional public-key encryption schemes

Because the ciphered image and the visual public keys (i.e., CI, PUA1, PUA2, PUB1, PUB2) are open to public, the attackers may try to generate the shared visual secret key from CI, PUA1, PUA2, PUB1, and PUB2 in order to decrypt the ciphered image CI, that is, to recover the original secret image. The security of the proposed encryption algorithm is based on the security of the shared visual secret key (SVSK = SVSKA = SVSKB) which depends on the solving of the problem of Boolean algebra arithmetic, where it is not possible to obtain the shared visual secret key from the public information (i.e., CI, PUA1, PUA2, PUB1, and PUB2). This is especially true when using a large integer number G, which leads to the use of huge probabilities and a large number of Boolean operations. In addition, it is practically impossible to find the inverse of PUA2 or PUB2 from the visual public key PU because PU is a non-invertible matrix as we conditioned above. Therefore, an attacker will face a hard time to obtain the shared visual secret key and to discover the secret image SI. In encryption phase, the sender (Alice) can encrypt the secret image SI by the (2, 2) ProbVC scheme into two shares (shadow images). The first share must be equal to the shared visual secret key SVSK (SVSK has already been established between the two parties and here it serves as an encryption key) while the second share is the ciphered image CI. The sender sends CI to the receiver. In decryption phase, the receiver can recover the secret image SI by stacking the shared visual secret key SVSK (here SVSK serves as a decryption key) and the ciphered image CI, but CI alone cannot disclose any information about the original secret image. In addition, if the ciphered image CI is changed and forged by an attacker, the stacked image will be unclear and the secret is still unidentified. Therefore, our proposed scheme is secure.


4.2 Computational complexity Image encryption includes two steps: first, obtain the shared visual secret key, and then create the ciphered image CI. The shared visual secret key includes constructing 2G of the intermediate shares, D1,…, DG and F1,…, FG, and constructing two multiplications of two binary matrices. The time complexity of constructing 2G of the intermediate shares is O(n2G) + O(n2G) = O(n2G) and the time complexity of constructing two multiplications of two binary matrices is O(n3) + O(n3) = O(n3) if neglecting the constant and the multiplication of two binary matrices is carried out naively. Therefore, the total time complexity of the shared visual secret key is , or O(n3) + either O(n3) + O(n2G) = O(n3) when 2 2 , excluding the time needed O(n G) = O(n G) when to generate 2G+2 distinct random shares, where the size of the share is equal to n×n pixels. Generating the shared visual secret key requires 6G-1 of OR operations (here OR operations mean superimposing the shares), and at most two multiplications of two binary matrices (keys) in the sender party (here multiplication means performing the OR and the AND operations of two binary matrices as shown previously in Section 3). The time complexity of the ciphered image CI is O(n2). Therefore, the total time complexity for image encryption is either O(n3) + O(n2) = , or O(n2G) + O(n2) = O(n2G) when O(n3) when . Image decryption includes two steps: first, obtain the shared visual secret key and then superimpose the shared visual secret key and the ciphered image CI for reconstructing the secret image. The time complexity for reconstructing the secret image is equal to the time complexity of the shared visual secret key which is, as we already mentioned in this Subsection, either O(n3) when , excluding the time needed or O(n2G) when to generate 2G+2 distinct random shares. Reconstructing the image requires 6G of OR operations (here OR operations mean superimposing the shares), and at most two multiplications of two binary matrices (keys) in the receiver party (here multiplication means performing the OR and the AND operations of two binary matrices as shown previously in Section 3). Suppose that an attacker wants to recover our public-key encryption scheme, two steps are required:  Suppose that there is no a computer; Table 6 shows how much time is needed to reconstruct the visual public-key encryption scheme, performed by a human who does not use any computational devices (i.e., done manually). We will assume using some different sizes of shares and G is equal to 2, 64, and 128. Also, we assume that a person performs one operation per minute. From Table 6, first, we can see that the time required increases with the increasing size of share while using the value of G less than or equal to n. Second, we can also see that the time

8

required increases with the increasing value of G (here G is greater than or equal to n) while using the same size of share. For example, from same table, the time required to reconstruct our scheme is more than half year when the share size is 64×64 pixels and G is less than or equal to 64. The time required to reconstruct our scheme is more than one year when the share size is 64×64 pixels and 128, and so on. Therefore, we can say that our scheme is secure when performing the computation manually and using a proper size of the share with a proper value of G.

1

2

3

4

5

6

7

Table 6: The Time spent to reconstruct our scheme manually Share Number of Time Description size G Boolean required* (pixels) operation Shared visual 8 min 2 23 secret key 2×2 8 Encryption 64 2 4.26 h Decryption 128 29 8.59 h Shared visual 6 2 2 1.06 h secret key 4×4 Encryption 64 210 17.06 h Decryption 128 211 1.42 d Shared visual 2 29 8.59 h secret key 8×8 12 Encryption 64 2 2.84 d Decryption 128 213 5.68 d Shared visual 12 2 2 2.84 d secret key 16×16 Encryption 64 214 11.37 d Decryption 128 215 22.75 d Shared visual 22.75 d 2 215 secret key 32×32 16 Encryption 64 2 1.51 mth Decryption 128 217 3.03 mth Shared visual 18 2 2 6.06 mth secret key 64×64 18 Encryption 64 2 6.06 mth Decryption 128 219 1.01 yr Shared visual 2 221 4.04 yr secret key 128×128 21 Encryption 64 2 4.04 yr Decryption 128 221 4.04 yr * min = minutes, h = hours, d = days, mth = months, yr = years

 Suppose that there is a computer; Tables 7 shows how much time is spent to recover the shared visual secret key when performed by a human who uses a computational device (i.e., a computer). We will assume a computer that executes one billion instructions per second. We will also assume using shadow images (shares) of large sizes and G is equal to 256, 4096, and 32768. From Table 7, first, we can see that the time required increases with increasing the size of share while using the value of G is less than or equal to n. Second, from the same table, we can also see that the time required increases with increasing the value of G (here G is greater than or equal to n) while using shares of the same size.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org Table 7: The time spent to reconstruct our scheme by a computer Number of Share size Time Description G Boolean (pixels) required* operation Shared visual 16 ms 256 224 secret key 256×256 Encryption 4096 228 268 ms Decryption 32768 231 2 sec Shared visual 134 ms 256 227 secret key 512×512 Encryption 4096 230 1 sec Decryption 32768 233 8 sec Shared visual 256 230 1 sec secret key 1024×1024 32 Encryption 4096 2 4 sec Decryption 32768 235 34 sec Shared visual 256 233 8 sec secret key 2048×2048 Encryption 4096 234 17 sec Decryption 32768 237 2.29 min Shared visual 1.14 min 256 236 secret key 4096×4096 36 Encryption 4096 2 1.14 min Decryption 32768 239 9.16 min Shared visual 39 9.16 min 256 2 secret key 8192×8192 Encryption 4096 239 9.16 min Decryption 32768 241 36.65 min Shared visual 256 242 1.22 h secret key 16384×16384 42 Encryption 4096 2 1.22 h Decryption 32768 243 2.44 h

9

party. On the side of the receiver party, because the sender’s shared visual secret key is equal to the receiver’s shared visual secret key, the receiver’s shared visual secret key SVSK serves as a decryption key. Fig. 5(i) shows that the secret image’s message can be recovered by stacking the receiver’s shared visual secret key SVSK and the ciphered image CI. PUA1

PU

(a)

(b) PUB1

(d)

PUA2

(c) SVSK

PUB2

(e)

(f) CI

* ms = milliseconds, sec = seconds, min = minutes, h = hours

5. Working example of the proposed scheme (g)

In this section, we will present an example of applying our proposed scheme to a black and white secret image. Fig. 5 shows one of the experimental results. In the initialization phase of establishing the shared visual secret key SVSK, the first party (Alice) and the second party (Bob) agree on an integer G with 2 and a common shadow image PU. For simplicity we assume they choose 16 and a common shadow image PU with size 512×512 pixels as shown in Fig. 5(a). Also, in the same phase, each party constructs his/her G+1 visual private keys by using the (k, k) ProbVC scheme as shown in [18, 31]. Each party generates his/her visual public keys (i.e., PUA1 and PUA2 for Alice and PUB1 and PUB2 for Bob) and sends them to the other party as shown in Fig. 5(b)-(e), and then they establish the shared visual secret key (SVSK = SVSKA = SVSKB), as shown in Fig. 5(f). In the encryption phase we assume that the sender takes a message of 512×512 pixels image as shown in Fig. 5(g) as the secret image SI. The secret image SI encrypts into two 512×512 shadow images (shares) by using the (2, 2) ProbVC scheme as shown in Subsection 2.2, where the first share is equal to the sender’s shared visual secret key SVSK (SVSK serves as an encryption key) and the second share is the ciphered image CI as shown in Fig. 5(h) which will be sent to the receiver

(h)

(i)

Fig. 5 Experimental results: (a) Common shadow image, (b)-(c) Alice’s visual public keys, (d)-(e) Bob’s visual public keys, (f) Shared visual secret key, (g) Secret image, (h) Ciphered image, and (i) Recovered image.

6. Conclusions In this paper, we presented a new public-key encryption scheme based on non-expansion visual cryptography and Boolean operation. Our scheme allows one party to send a secret image to another party over the open network, even if many eavesdroppers listen. We used simple Boolean operations to construct our scheme, in which the secret image can encrypt and decrypt easily without complex computations. Therefore, our scheme can be useful in many applications. Our scheme gives reliable security especially when using a large value of the G and a large size of the share

References [1] Y. C. Hou, and S. F. Tu, "A Visual Cryptographic Technique for Chromatic Images Using Multi-pixel Encoding Method", Journal of Research and Practice in Information Technology, Vol. 37, No. 2, (2005), pp. 179-192.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org [2] Y. C. Hou, "Visual cryptography for color images", Pattern Recognition, Vol. 36, No. 7, (2003), pp. 1619-1629. [3] A. MS, "Public Key Cryptography-Applications Algorithms and Mathematical Explanations", Tata Elxsi Ltd, (2007). [4] I. Ozturk, and I. Sogukpinar, "Analysis and Comparison of Image Encryption Algorithms", International Journal of Information Technology, Vol. 1, No. 2, (2005), pp. 64-67. [5] C. Chan, and Y. Wu, "A Visual Information Encryption Scheme Based on Visual Cryptography and D-H Key Agreement Scheme", International Journal of Computer Science and Network Security, Vol. 8, No. 4, (2008), pp. 128-132. [6] W. Stallings, Cryptography and Network Security-Principles and Practices, Prentice Hall, Inc, 4th Ed., (2006). [7] W. Diffie, and M. Hellman, "New Directions in Cryptography", IEEE Transactions in Information Theory, Vol. IT-22, No. 6, (1976), pp. 644-654. [8] K.-Y. Chen, "The study and Implementations of Certificates in PKI", Ph.D. Thesis, Department of Electrical Engineering, National Cheng Kung University, Taiwan, (2004). [9] C.-S. Laih, and K. Y. Chen, "Generating visible RSA public keys for PKI", International Journal of Information Security, Vol. 2, No. 2, Springer-Verlag, Berlin, (2004), pp. 103-109. [10] J. J. Amador, and R. W. Green, "Symmetric-Key Block Cipher for Image and Text Cryptography", International Journal of Imaging and Technology, Vol. 15, No. 3, (2005), pp. 178-188. [11] I. Muecke, "Greyscale and Colour Visual Cryptography", M.Sc. thesis, Faculty of Computer Science, Dalhousie University, USA, (1999). [12] W. D. Shun, Z. Lei, M. Ning, and H. L. Sheng, "Secret Color Images Sharing Schemes Based on XOR Operation", Department of Computer Science and Technology, Tsinghua University, Beijing, China, (2005). [13] M. Naor, and A. Shamir, "Visual cryptography", Advances in Cryptology-EUROCRYPT’94, lecture Notes in Computer Science, Vol. 950, Springer-Verlag, Berlin, (1995), pp. 1-12. [14] M. Nakajima, and Y. Yamaguchi, "Extended Visual Cryptography for Natural Images", Department of Graphics and Computer Sciences, Graduate School of Arts and Sciences, University of Tokyo, (2002). [15] C. Lin, and W. Tsai, "Visual cryptography for gray-level images by dithering techniques", Pattern Recognition Letter, Vol. 24, No. 1-3, (2003), pp. 349-358. [16] D. Tsai, T. Chen, and G. Horng, "A cheating prevention scheme for binary visual cryptography with homogeneous secret images", Pattern Recognition, Vol. 40, No. 8, (2007), pp. 2356-2366. [17] C. Yang, and T. Chen, "Colored visual cryptography scheme based on additive color mixing", Pattern Recognition, Vol. 41, No. 10, (2008), pp. 3114-3129. [18] R. Ito, H. Kuwakado, and H. Tanaka, "Image size invariant visual cryptography", IEICE Trans. Fund., Vol. E82-A, No.10, (1999), pp. 2172-2177. [19] M. Hellman, "An Overview of Public Key Cryptography", Communications Magazine, Vol. 16, No. 6, (1978), pp. 2432. [20] Y. Xue, "Overview of Public-Key Cryptography", CS 291: Network Security, Department of Electrical Engineering and Computer Science, Vanderbilt University, (2006).

10

[21] R. A. Rivest, A. Shamir, and L. Adleman, "A method for Obtaining Digital Signatures and Public-Key Cryptosystems", Communications of the ACM, Vol. 21, No. 2, (1978), pp. 120-126. [22] T. ElGamal, "A Public-key Cryptosystem and a Signature Scheme Based on Discrete Logarithms", IEEE Transactions on Information Theory, Vol. IT-31, No. 4, (1985), pp.469472. [23] N. Koblitz, "Elliptic Curve Cryptosystems", Mathematics of Computation, Vol. 48, No. 177, (1987), pp. 203-209. [24] A. D. Bonis, and A. D. Santis, "Secret Sharing and Visual Cryptography Schemes", Proceedings of the IFIP TC11 16th International Conference on Information Security, (2001), pp. 123-138. [25] R. Youmaran, A. Adler, and A. Miri, "An Improved Visual Cryptography Scheme for Secret Hiding", 23rd Biennial Symposium on Communications, (2006), pp. 340-343. [26] Z. Zhou, "Advances on Digital Video and Visual Cryptography", Ph.D. thesis, Faculty of Electrical Engineering, Delaware University, (2004). [27] D. Q. Viet, K. Kurosawa, "Almost Ideal Contrast Visual Cryptography Scheme with Reversing", lecture Notes in Computer Science, Vol. 2964, Springer, (2004), pp. 353-365. [28] C. Hsu, and Y. Hou, "Visual cryptography and statistics based method for ownership identification of digital images", in: Proceedings of the International Conference on Signal Processing (ICSP’2004), Istanbul, Turkey, (2004), pp. 221224. [29] C.-S. Hsu, "A study of Visual Cryptography and Its Applications to Copyright protection Based on Goal programming and Statistics", Ph.D. Dissertation, Department of Information Management, National Central University, Taiwan, (2004). [30] H.-C. Lin, "New Digital Image Encryption/Decryption Algorithms and Hidden Visual Cryptography Algorithm", M.Sc. thesis, Institute of Communication Engineering, Tatung University, (2004). [31] C.-N. Yang, "New visual secret sharing schemes using probabilistic method", Pattern Recognition Letter, Vol. 25, No. 4, (2004), pp.481-494. [32] S.-F. Tu, "On the design of protection scheme for digital images and documents based on visual secret sharing and Steganography", Ph.D. Dissertation, Department of Information Management, National Central University, Taiwan, (2004). Abdullah M. Jaafar was born in Taiz, Republic of Yemen in March 7, 1977. He received the B.Sc. degree in Computer Science from Al-Mustansiriyah University, Baghdad, Republic of Iraq in 1999, and the M.Sc. degree in Computer Science from Iraqi Commission for Computers and Informatic, Institute for Post Graduate Studies in Informatic, Baghdad, Republic of Iraq in 2003. During 20042006, he worked as a lecturer at Taiz University in Republic of Yemen. Currently he is a Ph.D. student at the School of Computer Sciences, Universiti Sains Malaysia. Azman Samsudin is a lecturer at the School of Computer Sciences, Universiti Sains Malaysia. He received the B.Sc. degree in Computer Science from University of Rochester, USA, in 1989. He obtained his M.Sc. and Ph.D. degrees in Computer Science from University of Denver, USA, in 1993 and 1998, respectively. His current research interests are in the fields of Cryptography and Parallel Distributed Computing.


11

Architecture for Automated Tagging and Clustering of Song Files According to Mood Puneet SINGH1, Ashutosh KAPOOR2, Vishal KAUSHIK3 and Hima Bindu MARINGANTI 4 1,2,3,4

Jaypee Institute of Information Technology NOIDA, Uttar Pradesh, India

Abstract Music is one of the basic human needs for recreation and entertainment. As song files are digitalized now a days, and digital libraries are expanding continuously, which makes it difficult to recall a song. Thus need of a new classification system other than genre is very obvious and mood based classification system serves the purpose very well. In this paper we will present a well-defined architecture to classify songs into different mood-based categories, using audio content analysis, affective value of song lyrics to map a song onto a psychological-based emotion space and information from online sources. In audio content analysis we will use music features such as intensity, timbre and rhythm including their subfeatures to map music in a 2-Dimensional emotional space. In lyric based classification 1-Dimensional emotional space is used. Both the results are merged onto a 2-Dimensional emotional space, which will classify song into a particular mood category. Finally clusters of mood based song files are formed and arranged according to data acquired from various Internet sources.

Keywords: Music i nformation ret rieval, mood detection from music, song classification, mood models, music features, lyric processing.

1. Introduction Listening music is one of the oldest and the easiest way to entertain oneself and change one's mood. But as the digital libraries are increasing with myriad songs coming each year, it is becoming difficult for listeners to keep track of each song. There are several problems faced by listeners when they have the high freedom of choice as they favor upper 20% of the items against the other 80%; this is known as 80-20 rule (Pareto Principle). Thus there is a need for a system which retrieves and recommends as well as arranges music according to a simple yet important category i.e., mood. A song comprises of two very important components, music and lyrics. Thus the effect of both on the psychology is the basis of the work presented here. A vast research has already been done in the field of music information retrieval and relation between music and emotion; as a result there exists many different models of human emotions and their relation with music. The

objective of the architecture is to use best possible psychological model of emotion and to incorporate the findings of these studies in the best possible way so that music can be queried, browsed or explored by mood, rather than by artist, album or genre. The very basic approach of the work presented here are the features extracted from an audio file (WAV format) which are used in combination with the affective value of song lyrics and other data from Internet to map a song onto a psychological-based emotion space. First the audio content of the song is analyzed and intensity, timbre and rhythm features are extracted, then lyrics file is analyzed using language processing. Then the mathematical values of all these data are mapped onto 2-Dimensional emotional space (Thayer's Energy Stress Model), each quadrant representing a set of moods. Finally play-lists are generated for these four quadrants which are afterwards arranged according to the data extracted from Internet. Such a system can have the following uses: • Context-Aware Play-list Generation • Retrieving Lost Music • Music Classification • Music Recommendation.

1.1 Mood Models To classify a song based on mood, it is really necessary to choose a model to map mathematical values onto an emotional space so that we can separate each song from another according to mood. There are many emotional models[10][11][12][13] in existence and traditionally they all use adjective descriptor based approach such as 'gloomy', 'depressing', 'energetic' etc. But the problem with such an approach is the wide range of adjective which keeps on changing in different mood models thus there is no standard model in this type of approach. Furthermore, the more adjectives there are there the more would be the complexity and probability of error in mood detection. This ambiguity makes it difficult to detect the true mood, hence a simpler and yet a profound model was needed and work by Thayer[9][13] is best for this purpose.


12

2.1 Feature extraction

Figure. 1 Thayer's Model of mood

Unlike other adjective based models which collectively form a mood pattern, this dimensional approach adopts the theory that mood is entailed from two factors:  Stress (happy/anxious)  Energy (calm/ energetic) And divides music mood into four clusters: Contentment, Depression, Exuberance and Anxious/Frantic as shown in Fig.1.

2. Mood detection using music In this section we will discuss the algorithm, design and complete approach of mood detection using audio content analysis. Detection of mood from music needs a very basic step of retrieving information from music which can be done by extracting features such as intensity, timbre and rhythm which have their sub features. For extracting features, first all the music files are down-sampled to 16000Hz, 16 bits, and mono-channel. Then a set of 10 features are extracted from music for every 10 seconds of the music file. Once all the features are extracted for a song on every 10 second basis then all features are normalized using Max-Min normalization so that each song has the uniform scale. (1) Thus each song will have values of every feature in between 0 to 100, for every 10 seconds of the song. After normalizing sub features, the overall parent feature value is calculated using weighted average of sub features. At last these mathematical values are scaled on a scale of -10 to 10 making it easy to map the features on Thayer's mood model.

Feature extraction is necessary for knowing the type of music, It was indicated that mode, intensity, timbre and rhythm are of great significance in arousing different music moods (Hevner, 1935; Radocy, and Boyle, 1988; Krumhansl, 2002).We have used only three features viz intensity, timbre and rhythm to map onto Thayer's mood model, where intensity is associated with energy while timbre and rhythm are in combination associated with stress. Intensity: This feature has two sub-features: signal's RMS and low-energy. RMS of a signal is a more prominent feature out of the two. If intensity is higher than it can be a song with excitement or a frantic song. Low-energy of a signal is the average of the values of the peak having values lower than a threshold which is generally very low. Thus low-energy will have values opposite to RMS i.e., for a song having high RMS, value of low-energy would be low. Overall intensity is calculated using weighted average of the two sub-features: Intensity = (RMS) × 0.8 + (100-lowenergy) × 0.2

(2)

Low-energy is subtracted from 100 because using Eq. (1) we have scaled low-energy on a scale of 0-100. This overall Intensity is extracted for each 10 second of the song file. Timbre: Timbre is the quality of the music. Existing results show that the timbre of sound is determined primarily by the spectral information [16]. We have used four features for extracting timbre: • • • •

Zero-cross (Z) Centroid (C) Roll off (R) Entropy (E)

Higher values of the feature mean lower stress on Thayer's 2-D emotional space. Thus we can say that a higher centroid is generally associated with positive emotion. Centroid and roll off are major source of mapping emotional intensity out of these four features. Overall timbre is calculated using weighted average of these four features: Timbre = Z × 0.2 + C × 0.4+ R × 0.3 + E × 0.1

(3)

Rhythm: It is said that three features of rhythm are more important for music strength, tempo and regularity [1]. Thus, more tempo and regularity in the song means positive emotion while low rhythmic values will mean negative emotion. The following features are extracted for calculating rhythm: • Average Beat-spectrum (B) • Average Tempo (T) • Average Regularity (Ar)

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org • Average Positive Autocorrelation (Pa) • Average Negative Autocorrelation (Na) Out of the five features most important features are tempo, beat-spectrum and regularity. Overall rhythm is calculated using weighted average of these five features: Rhythm = B×0.25+T × 0.45+Ar×0.2 +(Pa + Na)×0.05 (4) 2.2 Scaling and Averaging

After calculating features for every 10 seconds of the song file we have generated arrays of these three features. Now scaling is done to map the values of the features on a scale of (-10, 10). Scaling is important for mapping the values retrieved on Thayer's mood model. Before scaling we have passed each value of feature into the following equation: (5) By passing through this equation we have increased each value by a factor of the maximum value making the maximum value to be 100%. After scaling, an average is taken of all the scaled values for each feature, the result would be the final mathematical values of the all three features of the song.

2.3 Mood Mapping This will be the final and important step of mood detection. As we have discussed earlier that the mathematical values are mapped onto 4 quadrants:

Figure. 2 The four quadrants of the Thayer’s mood model Table 1: Valance - Arousal relation with Thayer's Model

Quadrant (Figure 2)

Valance (Emotion)

Arousal (Intensity)

Possible Moods

1

Positive

Positive

Very happy, Exciting

2

Positive

Negative

Soothing, Pleasure

3

Negative

Negative

Sad, Depressing

4

Negative

Positive

Frantic, Tense

1 and 2

Positive

zero

Serene, Happy

2 and 3

Zero

Negative

Dreamy, Sad

3 and 4

Negative

Zero

Sorrow, Disturbing

1 and 4

Zero

Positive

Exciting, Disturbing

1, 2, 3 & 4

Zero

Zero

Nothing can be said

13

When the features are scaled on the scale of (-10,10) then the feature having value in between (-1,1) will be marked as 'Zero'. The feature been marked as 'Zero' can be on either side of the quadrant. For calculating Valence out of timbre and rhythm, it has been stated that Timbre features are more important than the Rhythmic features for differentiating between contentment and depression while vice-versa is true for exuberance and anxious differentiation [1]. Thus to calculate valence (emotion) of the song we have used the following equation: Valence [2 – 3] = Timbre × 0.7 + Rhythm × 0.3

(6)

Valence [1 – 4] = Timbre × 0.3 + Rhythm × 0.7

(7)

2.4 Results Using this process we have completed the first step of mood detection of song file, we have tested the algorithm on more than 70 Indian Songs and reached an accuracy of 70.7%. The confusion matrix is as follows: Table 2: Mood Detection confusion matrix

Quadrant

1

2

3

4

1

75.6

6.3

0

18.1

2

6.3

65.7

24.1

3.9

3

0

20.6

70.1

9.3

4

14.2

0

8.4

77.4

As we aimed to tag a song file which not only contains music but also contains singer's voice which adds the confusion and ambiguity resulting lesser accuracy. Also the Indian music is too versatile, for example ‘Sufi’ songs bring contentment but are tagged as energetic song as they have high rhythm and timbre as well as intensity. Mood of a song is also subjective to the listeners; hence the accuracy varies from listener to listener. Thus we can say that the results obtained are satisfactory.


3. Mood detection using Lyrics In this section, complete algorithm and approach we have used for mood detection using lyrics has been discussed. Our aim is to refine the results of audio content analysis; we examined previous works which used the same approach [14] [15], and found that lyrics are a weak source of mood information. As the lyrics do not follow any particular grammar rule in favor of making it more rhyming, this weak source can considerably improve the result coming from music analysis. Meaning extraction from lyrics is a tedious task thus we used Bag-of-Words technique combined with Bi-gram model. Both the techniques are not based on semantic analysis of the text; these techniques are dependent on adjectives and combination of two adjectives. Unlike the mood mapping technique used earlier [14] we have observed that lyrics have more associativity with valence than arousal. When we analyzed 'lrc' files which have timestamp of each lyric phrase sung and plotted a graph with respect to time with the arousal and valence, we got the following results:

14

Figure 4 clearly depict the disassociation of results of lyrics analysis with the intensity feature extracted. This was true for most of the songs we analyzed. This motivated us to use 1-D model for lyrics analysis rather than a 2-D model. In this 1-Dimensional model a lyrics are classified in two categories: • Positive Emotional • Negative Emotional The other reason to use this approach was the fact that classification on the basis of intensity has much more accuracy than on the basis of valence. In Table 2 the confusion matrix has lower confusion in 1-2, 1-3, 2-1, 2-4, 3-1, 3-4, 4-2, 4-3. Hence we can say that if we can reduce the confusion between 1-4, 4-1, 2-3 and 3-2 it will improve the accuracy of the results. Thus we do not have much need of a two dimensional model for refining the results, and mapping lyrics emotion onto a 1-D model is less complex than onto the 2-D.

3.1 Bag-of-Words Analysis Bag-of-Words or BOW is a collection of words where each word is assigned tags from a dictionary. A word can have different tags, each tag is predefined as positive or negative according to mood. We have used the following tags: Table 3: Different tags used for BOW technique

Positive Emotion

Negative Emotion

Happy words

Sad words

Sexual Words

Anger Words

Insightful words

Death Words

Achievement words Figure. 3 Rhythm and lyrics analysis graph for the song Roobaroo( Rang de basanti) which depicts the associativity of this song’s lyrics with rhythm.

Other tags which we have decided on the basis of previous tags were: • Social words • Feel • Family • Health • Affect words • Friendship • Body • Cause • Inclusion • Exclusion We first calculated all positive and all negative words using simple word count, and then the left out tags were tagged on the basis of the values attained from the positive and negative word counts.

Figure. 4 This plot clearly depicts the associativity of lyrics analysis with valence. (The song is chor bazari from the movie “Love aaj kal”)


15

The left out tags are used according to the following rules:

3.3 Weighted average and mood mapping

if(total positive > 1.5 X total negative) { total positive = total positive + Social + Feel + Family + Friend + Affect + Cause + Inclusion + Exclusion + Body + Health; } else if(total negative > 1.5 X total positive) { total negative = total negative + Social + Feel + Family + Friend + Affect + Cause + Inclusion + Exclusion + Body + Health; }

A weighted average is taken of the values attained from both techniques. As it is said that BOW is an equally important technique, so the weighted average is taken as:

Thus the left out tags will be added only in the cases when positive and negative tags are 1.5 times the other. This will help in classifying the lyrics in three major categories High, Medium and Low Positive Emotion. The final emotion would be in terms of percentage of positive and negative emotions. If positive emotion in lyrics is more, clearly it means that song contains more words tagged as positive on emotional scale. The final result will be on a scale of 0 to 100.

3.2 Bi-gram Analysis Bi-gram analysis can be said to work as a pseudo-semantic analysis where no actual meaning is extracted but still a meaningful emotional value is generated. We used the same tags as of BOW but analyzed the combination of the two words. We marked each combination with a certain value lying between the range (-10, 10) depending on the mood they will elicit when combined together. There are more than 100 combinations possible for analysis; a few amongst them are as follow: Table 4: Bi-gram showing different emotional values

Tag1

Tag2

Emotional value

Example

Happy

Family

6

Smiling Brothers

Exclusion

Family

-5

Without Brothers

Body

Sexual

4

Hand Massaged

Negate

Happy

-6

Not Glad

Cause

Happy

7

Made Glad

After marking each bi-gram with a mathematical value we again calculated positive and negative percentages just as we did in BOW technique. We could have used trigram or n-gram for better results, but as the number of combinations increase the complexity increases. Bi-gram itself had more than 100 combinations, thus mood detection on the basis of the above mentioned two factors is optimal in terms of complexity.

Emotion = BOW × 0.4 + Bigram × 0.6

(8)

If finally the value of positive emotion is above 60 % then we will say the song has positive valence, and if the final emotional value is below 40 % we will say the song havs negative valence. We have not used the extent of positivity or negativity to refine the result; if the song was lyrically positive we added 1 to the valence and subtracted 1 if the song was negative. Thus on a 1-D emotional model we mapped any song file as positive emotional, negative emotional and if the value lies between 40-60 % then results could not be refined on the basis of lyrics.

3.4 Results Using this process we have refined the results significantly. The confusion matrix of the result from lyrics analysis is as follows: Table 5: Lyrics based mood detection confusion matrix

Emotion

Positive

Negative

NULL

Positive

62.3

7.3

30.4

Negative

9

64.4

26.6

NULL

12.2

8.5

79.3

As the NULL result will not affect the results of the audio content analysis, and the confusion in positive and negative is lesser so we can assume that Lyrics analysis method will refine the result without harming it much. We will discuss the final confusion matrix and accuracies in Results (5) section.

4. Implementation and Architecture Implementation is one of the important tasks to support research; we accomplished this task by using best tools available for each problem. We used Matlab for extracting music features using MIRToolBox, Perl for language processing of lyrics, and finally used the both results, generated clusters (play list of a particular mood) and sorted them according data acquired from Internet in C#. We have used XML file to store the data acquired from audio and lyrics processing and used them from C#. The basic idea behind the use of XML file is that it can store live data; using XML we can create a system where listeners can directly download these preprocessed files from Internet and use a music player to sort their music

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org according to mood with a mood equalizer. The XML file will have all the data needed to support the mood detection with some meta-data and manually tagged genre of song file.

16

process because our technique was not based on semantic analysis but the bag of words and word combinations, and adjectives and tags generally remains the same in Hindi as they were in English. After mood detection we arranged the clusters using the data we acquired from You-tube and IMDB. Suppose a song has 98 likes and 2 dislikes and has IMDB rating of the album 7.9 out of 10 while another in the same cluster has You-tube's 93 likes and 7 dislikes and IMDB rating for the album to be 8.5 out of 10. So we simply multiplied “likeness factor’s percentage” with IMDB rating. Table 7: Calculating likeness factor of two songs

Song

Total rating

Song 1

98% X 7.9 = 7.742

Song 2

93 % X 8.5 = 7.905

Thus Song 2 will have more likeness factor hence is arranged above Song 1 while play list generation.

Figure. 5 Complete architecture of the implementation

After mood detection is complete we have used You-tube's like dislike factor and IMDB rating of the album for sorting the clusters.

5. Results and Discussion After the whole processing we will have 4 clusters (playlists) as a result arranged in order of probability of likeness. The final confusion matrix of the final result is as follows: Table 6: Final Mood Detection confusion matrix

Quadrant

1

2

3

4

1

83.2

6.3

0

10.5

2

6.3

79.4

10.4

3.9

3

0

13.4

77.4

9.2

4

10.5

0

8.4

81.1

This clearly indicates the reduction in confusion and ambiguity. The overall accuracy after lyrics analysis refining, is increased approximately to 79.2 %. It must be noted that for attaining more efficiency we must have songs of more than 1 minute duration which will give it at least 6 different feature sets for better normalization. Corrupted of pirated music files having high noise ratio will also face the problem as noise will add up to the RMS values and disturbing timbre and rhythmic values. The data set we have used for the experiment was Indian music which was the basis of the whole research. Indian music has lyrics in 'Hindi' language so we converted each 'lrc' file into English, because English corpora are easily available. But the results will not be degraded using this

6. Conclusion and future work In this article we discussed simpler but detailed algorithms to find mood from a song file. The architecture is not completely new but the way of implementation is novel. The results we obtained from experiment and implementation were more than satisfactory. The basic motive behind the experiment was to create a system where music listeners can browse music according to mood of the music. In future one can try to improve the results using better music feature extraction and incorporating better lyric analysis methods, music features are really important for the type of music. Thus research can be done to find the best features for a particular type of music. Internet digital libraries are continuously tagged by millions of users for example www.last.fm can be used to find genre and finding features according to the genre. This could be a hierarchical model where at each stage results are generated at each node and classified. Acknowledgments We would like to acknowledge all the people who participated in the survey to tag the music in a particular quadrant so that we can validate our result. We would specially like to thank Prabodh Prakash and Tanvi Shukla for continuously motivating and helping to make the work possible. Finally we would like to thank Department of Computer science of Jaypee Institute of Information Technology for providing us the platform and opportunity to work in this area. References

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org [1] Dan Liu,Lie Lu and Hong Jiang Zhang, Automatic Mood Detection from Acoustic Music Data, The Johns Hopkins University , ISMIR, 2003. [2] Yu-Hao Chen, Jin-Hau Kuo, Wei-Ta Chu, and Ja-Ling Wu, Movie Emotional Event Detection based on Music Mood and Video Tempo, IEEE,2006. [3] Tao Li and Mitsunori Ogihara, Detecting Emotion in Music, The Johns Hopkins University, ISMIR, 2003. [4] Owen Craigie Meyers, A Mood-Based Music Classification and Exploration System, MS thesis, MIT, 2007. [5] Sofia Gustafson-Capková, Emotions in Speech: Tagset and Acoustic Correlates, term paper, Stockholm University, 2001. [6] Sabine Letellier-Zarshenas & Dominique Duhaut Thi-Hai-Ha Dang, Comparison of recent architectures of emotions,IEEE, 2008. [7] John A. Schinka , Wayne F. Velicer and Irving B. Weiner, handbook of psychology, [8] Rajinder kumar math, A.V. Sutagundar, S.S. Manvi, Analysis of Automatic Music Genre Classification System , International Journal of Computer Science and Information Security, Vol 1, No. 1, 2009. [9] R. E. Thayer, The origin of everyday moods: managing energy, tension, and stress, Oxford University Press, New York, 1996. [10] Campbell, J. P. (1997). Speaker recognition: tutorial.Proceeding of the IEEE, 85 (9), 1437-1462.

a

[11] Erling, W., et al (1996). Content-based classification,search, and retrieval of audio. IEEE Trans. Multimedia, 3,27-36. [12] Hevner, K. (1935). Expression in music: a discussion ofexperimental studies and theories. Psychological Review,42, 186-204. [13] Thayer, R. E. The biopsychology of mood and arousal. Oxford University Press, 1989 [14] Jens Grivolla, Cyril Laurier, Perfecto Herrera. Multimodal Music Mood Classification using Audio and Lyrics, IEEE, 2008 [15] Xiao Hu, J. Stephen Downien Ehmann.LYRIC TEXT MINING IN CLASSIFICATION, ISMIR, 2009

and Andreas F. MUSIC MOOD

[16] Zhang, T. & Kuo, J. Hierarchical system for content-based audio classification and retrieval. Proceeding of SPIE's Conference on Multimedia Storage and Archiving Systems III, 3527,1998, (pp. 398-409).

17

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 ISSN (Online): 1694‐0784 ISSN (Print): 1694‐0814

18

An Improved k-Nearest Neighbor Classification Using Genetic Algorithm N. Suguna1, and Dr. K. Thanushkodi2 1

Professor in Computer Science and Engg, Akshaya College of Engineering and Technology, Coimbatore, Tamil Nadu, India.

2

Director, Akshaya College of Engineering and Technology, Coimbatore, Tamil Nadu, India.

Abstract k-Nearest Neighbor (KNN) is one of the most popular algorithms for pattern recognition. Many researchers have found that the KNN algorithm accomplishes very good performance in their experiments on different data sets. The traditional KNN text classification algorithm has three limitations: (i) calculation complexity due to the usage of all the training samples for classification, (ii) the performance is solely dependent on the training set, and (iii) there is no weight difference between samples. To overcome these limitations, an improved version of KNN is proposed in this paper. Genetic Algorithm (GA) is combined with KNN to improve its classification performance. Instead of considering all the training samples and taking k-neighbors, the GA is employed to take k-neighbors straightaway and then calculate the distance to classify the test samples. Before classification, initially the reduced feature set is received from a novel method based on Rough set theory hybrid with Bee Colony Optimization (BCO) as we have discussed in our earlier work. The performance is compared with the traditional KNN, CART and SVM classifiers.

Keywords: k-Nearest Neig hbor, Gen etic Algo rithm, S upport Vector Machine, Rough Set.

1. Introduction Nearest neighbor search is one of the most popular learning and classification techniques introduced by Fix and Hodges [1], which has been proved to be a simple and powerful recognition algorithm. Cover and Hart [2] showed that the decision rule performs well considering that no explicit knowledge of the data is available. A simple generalization of this method is called K-NN rule, in which a new pattern is classified into the class with the most members present among the K nearest neighbors, can be used to obtain good estimates of the Bayes error and its probability of error asymptotically approaches the Bayes error [3]. The traditional KNN text classification has three limitations [4]: 1. High calculation complexity: To find out the k nearest neighbor samples, all the similarities between the

training samples must be calculated. When the number of training samples is less, the KNN classifier is no longer optimal, but if the training set contains a huge number of samples, the KNN classifier needs more time to calculate the similarities. This problem can be solved in 3 ways: reducing the dimensions of the feature space; using smaller data sets; using improved algorithm which can accelerate to [5]; 2. Dependency on the training set: The classifier is generated only with the training samples and it does not use any additional data. This makes the algorithm to depend on the training set excessively; it needs recalculation even if there is a small change on training set; 3. No weight difference between samples: All the training samples are treated equally; there is no difference between the samples with small number of data and huge number of data. So it doesn’t match the actual phenomenon where the samples have uneven distribution commonly. A wide variety of methods have been proposed to deal with these problems [6-9]. Another problem is that the classification algorithms will be confused with more number of features. Therefore, feature subset selection is implicitly or explicitly conducted for learning systems [10], [11]. There are two steps in neighborhood classifiers. First an optimal feature subspace is selected, which has a similar discriminating power as the original data, but the number of features is greatly reduced. Then the neighborhood classifier is applied. In this paper, we have used a novel method based on Rough set theory hybrid with Bee Colony Optimization (BCO) to select the optimal feature set as discussed in our previous work [12]. Then the proposed GKNN classifier is analyzed with this reduced feature set. In this paper, Genetic Algorithm (GA) is combined with kNearest Neighbor (KNN) algorithm called as Genetic KNN (GKNN), to overcome the limitations of traditional KNN. In traditional KNN algorithm, initially the distance between all the test and training samples are calculated and the k-

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org neighbors with greater distances are taken for classification. In our proposed method, by using GA, k-number of samples are chosen for each iteration and the classification accuracy is calculated as fitness. The highest accuracy is recorded each time. Thus, it is not required to calculate the similarities between all samples, and there is no need to consider the weight of the category. This paper is structured as follows: the following section presents the traditional KNN algorithm. Section 3 explains the proposed GKNN classifier. The comparative experiments and results are discussed in Section 4 and the work is concluded in Section 5.

2. KNN Classification Algorithm In pattern recognition field, KNN is one of the most important non-parameter algorithms [13] and it is a supervised learning algorithm. The classification rules are generated by the training samples themselves without any additional data. The KNN classification algorithm predicts the test sample’s category according to the K training samples which are the nearest neighbors to the test sample, and judge it to that category which has the largest category probability. The process of KNN algorithm to classify sample X is [14]:  Suppose there are j training categories C1,C2,…,Cj and the sum of the training samples is N after feature reduction, they become m-dimension feature vector.  Make sample X to be the same feature vector of the form (X1, X2,…, Xm), as all training samples.  Calculate the similarities between all training samples and X. Taking the ith sample di (di1,di2,…,dim) as an example, the similarity SIM(X, di) is as following: m

SIM ( X , d i ) 



X j 1

j

.d ij

2

 m   m    X j  .   d ij       j 1   j 1 

2

Choose k samples which are larger from N similarities of SIM(X, di), (i=1, 2,…, N), and treat them as a KNN collection of X. Then, calculate the probability of X belong to each category respectively with the following formula.

P( X , C j ) 

 SIM ( X , d ) . y(d , C i

d

i

j

)

Where y(di, Cj) is a category attribute function, which satisfied

1, d i C j 0, d i C j

y(d,, Cj) =  

Judge sample X to be the category which has the largest P(X, Cj).

3. Improved KNN Classification Based On Genetic Algorithm

19

Genetic algorithm (GA) [15], [16] is a randomized search and optimization technique guided by the principles of evolution and natural genetics, having a large amount of implicit parallelism. GAs perform search in complex, large and multimodal landscapes, and provide near-optimal solutions for objective or fitness function of an optimization problem. In GA, the parameters of the search space are encoded in the form of strings (called chromosomes). A collection of such strings is called a population. Initially, a random population is created, which represents different points in the search space. An objective and fitness function is associated with each string that represents the degree of goodness of the string. Based on the principle of survival of the fittest, a few of the strings are selected and each is assigned a number of copies that go into the mating pool. Biologically inspired operators like cross-over and mutation are applied on these strings to yield a new generation of strings. The process of selection, crossover and mutation continues for a fixed number of generations or till a termination condition is satisfied. An excellent survey of GA along with the programming structure used can be found in [16]. GA have applications in fields as diverse as VLSI design, image processing, neural networks, machine learning, job shop scheduling, etc. String Representation - Here the chromosomes are encoded with real numbers; the number of genes in each chromosome represents the samples in the training set. Each gene will have 5 digits for vector index and k number of genes. For example, if k=5, a sample chromosome may look as follows: 00100 10010 00256 01875 00098 Here, the 00098 represents, the 98th instance and the second gene say that the 1875 instance in the training sample. Once the initial population is generated now we are ready to apply genetic operators. With these k neighbors, the distance between each sample in the testing set is calculated and the accuracy is stored as the fitness values of this chromosome. Reproduction (selection) - The selection process selects chromosomes from the mating pool directed by the survival of the fittest concept of natural genetic systems. In the proportional selection strategy adopted in this article, a chromosome is assigned a number of copies, which is proportional to its fitness in the population, that go into the mating pool for further genetic operations. Roulette wheel selection is one common technique that implements the proportional selection strategy. Crossover - Crossover is a probabilistic process that exchanges information between two parent chromosomes for generating two child chromosomes. In this paper, single point crossover with a fixed crossover probability of pc is used. For chromosomes of length l, a random integer, called the crossover point, is generated in the range [1, l-1]. The portions of the chromosomes lying to the right of the crossover point are exchanged to produce two offspring. Mutation - Each chromosome undergoes mutation with a fixed probability pm. For binary representation of chromosomes, a bit position (or gene) is mutated by simply

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org flipping its value. Since we are considering real numbers in this paper, a random position is chosen in the chromosome and replace by a random number between 0-9. After the genetic operators are applied, the local maximum fitness value is calculated and compared with global maximum. If the local maximum is greater than the global maximum then the global maximum is assigned with the local maximum, and the next iteration is continued with the new population. The cluster points will be repositioned corresponding to the chromosome having global maximum. Otherwise, the next iteration is continued with the same old population. This process is repeated for N number of iterations. From the following section, it is shown that our refinement algorithm improves the cluster quality. The algorithm is given as. 1. 2.

3.

4.

Choose k number of samples from the training set to generate initial population (p1). Calculate the distance between training samples in each chromosome and testing samples, as fitness value. Choose the chromosome with highest fitness value store it as global maximum (Gmax). a. For i = 1 to L do i. Perform reproduction ii. Apply the crossover operator. iii. Perform mutation and get the new population. (p2) iv. Calculate the local maximum (Lmax). v. If Gmax < Lmax then a. Gmax = Lmax; b. p1 = p2; b. Repeat Output – the chromosome which obtains Gmax has the optimum K-neighbors and the corresponding labels are the classification results.

4. Experiments & Results

Dataset Name Dermatology Cleveland Heart HIV Lung Cancer Wisconsin

Table 1 Datasets Used for Reduct No. of Features in Total No. of Total No. of the Reduced Set Instances Features (BeeRSAR [12]) 366 34 7 6 300 13 500 32 699

the performance with various k values on three different distance measures (1-norm, 2-norm and n-norm). Table 2 depicts the performance accuracy of our proposed classifier compared with traditional KNN, CART and SVM classifiers [17]. From the results it is shown that our proposed method outperforms the others with greater accuracy.

5. Conclusion The KNN classifier is one of the most popular neighborhood classifier in pattern recognition. However, it has limitations such as: great calculation complexity, fully dependent on training set, and no weight difference between each class. To combat this, a novel method to improve the classification performance of KNN using Genetic Algorithm (GA) is proposed in this paper. Initially the reduced feature set is constructed from the samples using Rough set based Bee Colony Optimization (BeeRSAR). Next, our proposed GKNN classifier is applied for classification. The basic idea here is that, instead of calculating the similarities between all the training and test samples and then choosing k-neighbors for classification, by using GA, only k-neighbors are chosen at each iteration,, the similarities are calculated, the test samples are classified with these neighbors and the accuracy is calculated. This process is repeated for L number of times to reach greater accuracy; hence the calculation complexity of KNN is reduced and there is no need to consider the weight of the samples. The performance of GKNN classifier is tested with five different medical data collected from UCI data repository, and compared with traditional KNN, CART and SVM. The experiments and results show that our proposed method not only reduces the complexity of the KNN, also it improves the classification accuracy. References [1]

[2]

The performance of the reduct approaches discussed in this paper has been tested with 5 different medical datasets, downloaded from UCI machine learning data repository. Table 1 shows the details about the datasets and the reduced feature set used in this paper.

21 56 09

8 4 4

[3] [4]

[5]

[6] [7]

[8] With the reduced feature set, the GKNN classifier is applied. Ten-fold cross validation method is performed for analyzing

20

E. Fix, and J. Hodges, “Discriminatory analysis. Nonparametric discrimination: Consistency properties”. Technical Re port 4, US AF S chool of Av iation Medicine, Randolph Field, Texas, 1951. T.M. Cover, and P.E. Hart, “Nearest neighbor pattern classification”, IEEE T ransactions on In formation Theory, 13, pp. 21–27, 1967. R.O. Duda, and P.E. Hart, Pattern cl assification an d scene analysis, New York: Wiley, 1973. W. Yu, and W. Zhengguo, “A fast kNN algorithm for text categorization”, Proceedings of the Si xth International Conference on Machine Learning and Cybernetics, Hong Kong, pp. 3436-3441, 2007. W. Yi, B. Shi, and W. Zhang’ou, “A Fast KNN Algorithm Applied to Web Text Categorization”, Journal of T he Chin a Society f or Scien tific and Technical Information, 26(1), pp. 60-64, 2007. K.G. Anil, “On optimum choice of k in nearest neighbor classification”, Computational Sta tistics a nd D ata Analysis, 50, pp. 3113–3123, 2006. E. Kushilevitz, R. Ostrovsky, and Y. Rabani, “Efficient search for approximate nearest neighbor in high dimensional spaces”. SIAM J ournal on Comp uting, 30, pp. 457–474, 2000. M. Lindenbaum, S. Markovitch, and D. Rusakov, “Selective sampling for nearest neighbor classifiers”, Machine Learning, 54, pp. 125–152, 2004.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org [9] [10]

[11] [12]

[13]

[14] [15] [16] [17]

C. Zhou, Y. Yan, and Q. Chen, “Improving nearest neighbor classification with cam weighted distance”. Pattern Recognition, 39, pp. 635–645, 2006. D.P. Muni, and N.R.D. Pal, “Genetic programming for simultaneous feature selection and classifier design”, IEEE Transactions on Systems Man and Cybernetics Part B – Cybernetics, 36, pp. 106–117, 2006. J. Neumann, C. Schnorr, and G. Steidl, “Combined SVMbased feature selection and classification”, Machine Learning, 61, pp. 129–150, 2005. N. Suguna, and K. Thanushkodi, “A Novel Rough Set Reduct Algorithm Based on Bee Colony Optimization”, International Journal of Granular Computing, Rough Sets and Intelligent Systems, (Communicated) 2010. Belur V. Dasarathy, “Nearest Neighbor (NN) Norms: NN Pattern Classification Techniques”, Mc Graw-Hill Computer Science Series, IEEE Computer Society Press, Las Alamitos, California, pp. 217-224, 1991. Y. Lihua, D. Qi, and G. Yanjun, “Study on KNN Text Categorization Algorithm”, Micro Computer Information, 21, pp. 269-271, 2006. D.E. Goldberg, Genetic A lgorithms in Search, Optimization a nd M achine Learning, Addison-Wesley, New York, 1989. L. Davis (Ed.), Handbook of Ge netic Algo rithms, Van Nostrand Reinhold, New York, 1991. Q. Hu, D. Yu and Z. Xie, Neighborhood Class ifiers, Expert Systems with Applications, 2007.

21

Author Biographies N.Suguna received her B.E degree in Computer Science and Engineering from Madurai Kamaraj University in 1999 and M.E. degree in Computer Science and Engineering from Bangalore University in 2003. She has more than a decade of teaching experience in various Engineering colleges in Tamil Nadu and Karnataka. She is currently with Akshaya College of Engineering and Technology, Coimbatore, Tamilnadu, India. Her research interests include Data Mining, Soft Computing and Object Oriented Systems. Dr.K.Thanushkodi has more than 35 years of teaching experience in various Government & Private Engineering Colleges. He has published 45 papers in International journals and conferences. He is currently guiding 15 research scholars in the area of Power System Engineering, Power Electronics and Computer Networks. He has been the Principal in-charge and Dean in Government College of Engineering Bargur. He has served as senate member in Periyar University, Salem. He has served as member of the research board, Anna University, Chennai. He Served as Member Academic Council, Anna University, Chennai. He is serving as Member, Board of Studies in Electrical Engineering, Anna University, Chennai. He is serving as Member, Board of Studies in Electrical and Electronics & Electronics and Communication Engineering, Amritha Viswa Vidya Peetham, Deemed University, Coimbatore. He is serving as Governing Council Member SACS MAVMM Engineering College, Madurai. He served as Professor and Head of E&I, EEE, CSE & IT Departments at Government College of Technology, Coimbatore. He is serving as Syndicate Member of Anna University, Coimbatore. Currently, he is the Director of Akshaya College of Engineering and Technology, Coimbatore.

Table 2. Performance Analysis of the Classifiers Classifier

K Value

KNN

5

10

15

20

GKNN

5

10

15

20

SVM CART

Distance Measure 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm 1-norm 2-norm n-norm

Dermatology 75.75 ± 0.02 82.98 ± 0.05 65.77 ± 0.18 66.03 ± 0.83 73.80 ± 0.70 63.79 ± 0.76 73.97 ± 0.94 68.24 ± 0.84 75.96 ± 0.68 70.75 ± 0.26 64.08 ± 0.40 63.80 ± 0.45 93.79 ± 0.15 94.24 ± 0.69 94.80 ± 0.67 94.51 ± 0.94 93.54 ± 0.31 94.88 ± 0.28 93.97 ± 0.19 93.57 ± 0.66 93.86 ± 0.12 93.65 ± 0.53 94.60 ± 0.26 93.80 ± 0.65 89.12 85.76

Cleveland Heart 74.27 ± 0.06 66.62 ± 0.71 73.46 ± 0.89 69.65 ± 0.89 67.60 ± 0.30 67.42 ± 0.24 73.86 ± 0.45 71.68 ± 0.92 71.77 ± 0.14 69.96 ± 0.80 69.00 ± 0.63 72.64 ± 0.69 87.18 ± 0.92 92.63 ± 0.35 85.64 ± 0.47 92.81 ± 0.45 96.67 ± 0.40 90.47 ± 0.19 93.42 ± 0.75 81.21 ± 0.50 85.56 ± 0.63 90.45 ± 0.61 87.87 ± 0.19 88.96 ± 0.85 85.23 83.23

HIV 67.59 ± 0.92 68.31 ± 0.71 66.85 ± 0.63 73.13 ± 0.39 73.53 ± 0.94 74.64 ± 0.76 74.69 ± 0.09 72.99 ± 0.08 66.83 ± 0.69 68.73 ± 0.33 72.71 ± 0.40 69.22 ± 0.50 81.33 ± 0.61 90.71 ± 0.07 91.86 ± 0.90 80.53 ± 0.65 82.85 ± 0.35 92.98 ± 0.17 82.99 ± 1.00 91.96 ± 1.00 91.64 ± 0.93 82.64 ± 0.85 87.64 ± 0.58 83.89 ± 0.58 82.33 80.45

Lung Cancer 68.86 ± 0.23 68.07 ± 0.66 68.64 ± 0.10 68.52 ± 0.96 70.07 ± 0.88 70.05 ± 0.49 68.27 ± 0.14 68.50 ± 0.95 69.37 ± 0.71 67.85 ± 0.27 69.98 ± 0.62 69.57 ± 0.87 86.24 ± 0.56 90.17 ± 0.69 88.76 ± 0.29 88.91 ± 0.08 84.86 ± 0.37 82.99 ± 0.87 83.23 ± 0.36 82.88 ± 0.27 86.69 ± 0.81 84.69 ± 0.58 87.51 ± 0.49 86.78 ± 0.35 87.01 82.29

Wisconsin Breast Cancer 76.42 ± 0.31 69.48 ± 0.66 76.07 ± 0.67 75.60 ± 0.84 68.94 ± 0.69 78.09 ± 0.48 77.36 ± 0.37 80.44 ± 0.93 68.14 ± 0.42 72.10 ± 0.85 67.07 ± 0.40 71.69 ± 0.63 97.56 ± 0.01 90.89 ± 0.98 87.96 ± 0.70 97.92 ± 0.51 92.63 ± 0.84 92.62 ± 0.81 89.73 ± 0.36 86.15 ± 0.41 95.70 ± 0.53 86.39 ± 0.47 94.28 ± 0.78 91.38 ± 0.84 86.77 85.34


Empirical Evaluation of Suitable Segmentation Algorithms for IR Images 1

2

Dr. (Mrs.) G.Padmavathi1, Dr.(Mrs.) P.Subashini2 and Mrs.A.Sumi3 Professor and Head, Department of Computer Science, Avinashilingam University for Women Coimbatore, Tamilnadu 641043, India

Associate Professor, Department of Computer Science, Avinashilingam University for Women Coimbatore, Tamilnadu 641043, India 3

Research Staff, Department of Computer Science, Avinashilingam University for Women Coimbatore, Tamilnadu 641043, India

Abstract Image segmentation is the first stage of processing in many practical computer vision systems. Development of segmentation algorithms has attracted considerable research interest, relatively little work has been published on the subject of their evaluation. Hence this paper enumerates and reviews mainly the image segmentation algorithms namely Otsu, Fuzzy C means, Global Active Contour / Snake model and Watershed. These suitable segmentation methods are implemented for IR images and are evaluated based on the parameters. The parameters are Variation Of Information (VOI), Global Consistency Error (GCE) and Probabilistic Rand Index (PRI). The objective of the paper is to identify the best segmentation algorithm that is suitable for IR images. From the experimentation and evaluation it is observed that the Global Active Contour/Snake model works better compared to other methods for IR images.

Keywords: IR Image, Segmentation, Otsu, Global Active Contour/Snake, Fuzzy C Means, Watershed.

1. Introduction Infrared heat wave image is different from the visible light images. It reflects the distribution of the object surface temperature and latent characteristics of material form. The infrared heat radiation due to the imperfections of the system will bring a variety of noise in the imaging process. The noise of complex distribution of infrared images have made the signal to noise ratio lower than visible light images. In addition, there are still nonuniformity and low-resolution features in infrared images, which result in a higher demand to infrared image segmentation.

Segmentation is an essential pre-processing step for many image analysis applications. From the segmentation results, it is possible to identify regions of interest and objects in the scene, which is very beneficial to the subsequent image analysis or annotation. The aim is to partition the image into a finite number of semantically important regions. In this paper four types of segmentation methods, Watershed [5], Otsu [7], Fuzzy C means [1] and Global active Contour\snake model [10] are used and compared using evaluation parameters. The parameters are Probabilistic Rand Index (PRI)[12] counts the fraction of pairs of pixels whose labellings are consistent between the computed segmentation and the ground truth, averaging across multiple ground truth segmentations to account for scale variations in human perception. Global Consistency Error (GCE)[15] measures the extent to which one segmentation can be viewed as a refinement of the other. Segmentations that are related in this manner are considered to be consistent, since they could represent the same natural image segmented at different scales. Variation Of Information (VOI)[11] defines the distance between two segmentations as the average conditional entropy of one segmentation given the other, and thus roughly measures the amount of randomness in one segmentation, which cannot be explained by the other. In an effort to compare the performance of current segmentation algorithms to human perceptual grouping as well as understand the cognitive processes that govern grouping of visual elements in images, much work has gone into hand-labeled segmentations of IR images. The above segmentation algorithms mainly are applied for Infrared, and for some GPR and X-ray images. In the next section, the theoretical foundation is given for infrared image segmentation algorithms. Section 3 gives the theoretical explanation of four parameters used

22

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org to evaluate the segmentation methods. Section 4 gives the experimental results obtained by using some benchmark pictures of IR, GPR, and X-Ray. Finally, conclusions and discussion are given.

2. Segmentation Algorithms A variety of segmentation algorithms are available in the literature. Out of which, four distinct algorithms are presented with details. They are as follows: i. Watershed Segmentation, ii. Global Active Contour / Snake Model, iii. Fuzzy C Means (FCM) and iv. Otsu.

23

meeting in order to avoid the merging of catchment basins. The watershed lines are defined by the catchment basins divided by the dam at the highest level where the water can reach the landscape. As a result, watershed lines can separate individual catchment basins in the landscape. The idea is described in Figure 1, which describes the flooding or rain falling process of watershed algorithm (Hsiesh, 2006). The process of rain falling is described in Figure 2.

2.1 Watershed Segmentation Watershed segmentation is a morphological based method of image segmentation. The gradient magnitude of an image is considered as a topographic surface for the watershed transformation. Watershed lines can be found by different ways. The complete division of the image through watershed transformation relies mostly on a good estimation of image gradients. The result of the watershed transform is degraded by the background noise and it produces the over-segmentation. Moreover, under segmentation is produced by low-contrast edges that generate small magnitude gradients, causing distinct regions to be erroneously merged. Watershed transformation is a morphological based tool for image segmentation. In grey scale mathematical morphology, the watershed transformation for image segmentation is originally proposed by Digabel and Lantuejoul (1977) and later improved by Li et. al. (2003). The watershed transform can be classified as a regionbased segmentation approach [6].

Figure 2: Illustrations of flooding (rain-falling) process of watershed transform.

2.2 Global Active Contour / Snake Model The active contour/snake model is one of the most successful variation models in image segmentation. It consists of evolving a contour in images towards the boundaries of objects. This new formulation is said geometrically intrinsic because the proposed snake energy is invariant with respect to the curve parameterization. The model is defined by the following minimization problem:



min c EGAC (C )  

The idea of watershed can be viewed as a landscape immersed in a lake catchment basins filled with water starting at each local minimum. Dams must be built where the water coming from different catchment basins may be



g ( I 0 (C ( s )) ) ds ,

(1)

where ds is the Euclidean element of length and L(C)

L(C)

ds is the length of the curve C defined by L(C)  0 Hence, the energy functional above in equation(1) is actually a new length obtained by weighting the Euclidean element of length ds by the function g which contains information concerning the boundaries of objects . The function g is an edge indicator function that vanishes at object boundaries such as g

Figure 1: Illustration of immersion process of watershed transforms. (CB is for catchment basins)

L (C )

0

  I0  

1 1    I0

2 ,

(2)

where I0 is the original image and β is an arbitrary positive constant. The calculus of variations provides us the Euler–Lagrange equation of the functional EGAC and the gradient descent method gives us the flow that minimizes as fast as possible EGAC [9].


 02 (T )

2.3 Fuzzy C Means (FCM) The FCM method applied to image segmentation is a procedure of the label following an unsupervised fuzzy clustering. It suits for the uncertain and ambiguous characteristic in images. However the FCM exploits the homogeneity of data only in the feature space and does not adapt to their local characteristics. The FCM algorithm is an iterative algorithm that finds clusters in data and uses the concept of fuzzy membership instead of assigning a pixel to a single cluster. Each pixel will have different membership values on each cluster. The Fuzzy C-Means [2] attempts to find clusters in the data by minimizing an objective function shown in the equation (3) below:

J 

N

C

  i 1

j 1

 imj x i  c

2 j

(3)

hence J is the objective function. After one iteration of the algorithm the value of J is smaller than before. It means the algorithm is converging or getting closer to a good separation of pixels into clusters. N is the number of pixels in the image. C is the number of clusters used in the algorithm, and which must be decided before execution. μ is the membership table -- a table of NxC entries that contains the membership values of each data point and each cluster. m is a fuzziness factor (a value larger than 1). xi is the ith pixel in N. cj is jth cluster in C. |xi - cj | is the Euclidean distance between xi and cj.

2.4 Otsu method Otsu method is based on the optimal thresholding for image segmentation. The minimization of the criterion function is the major focus. The criterion for Otsu [8] is the minimization of the within-group variance of the two groups of pixels separated by the threshold. The function of the Otsu method is as follows:

 2 within(T )  nB (T ) B2 (T )  n0 (T ) 02 (T )

(4)

where T 1

nB (T )   p(i ) i 0

N 1

n0 (T )   p(i ) i T

S2 (T)

= the variance of the pixel in the background (T)

and [0 ,N-1 ] is the range of intensity levels.

3. Parameters Used for Evaluation Unsupervised image segmentation is an important component in many image understanding algorithms and practical vision systems. However, the evaluation of segmentation algorithms thus far have been largely subjective, leaving a system designer to judge the effectiveness of a technique based only on intuition and results in the form of a few example segmented images. This is largely due to image segmentation being an illdefined problem and there is no unique ground-truth segmentation of an image against which the output of an algorithm may be compared. Segmentation algorithms taken are generally applicable to all images, and different algorithms are not equally suitable for a particular application. Here needs a way of comparing them, so that the better ones can be selected. Evaluation results vary significantly between different evaluators, because each evaluator may have distinct standards for measuring the quality of the segmentation. Any evaluation metric desired should take into account the following effects:  Over-segmentation. A region of the reference is represented by two or more regions in the examined segmentation.  Under-segmentation. Two or more regions of the reference are represented by a single region in the examined segmentation.  Inaccurate boundary localization. Ground truth is usually produced by humans that segment at different granularities.  Different number of segments. One needs to compare two segmentations when they have different numbers of segments. So, this paper presents three different parameters that are used to evaluate the experimented segmentation methods. The Parameters Used for Evaluation are as follows: i. Global Consistency Error (GCE) ii. The Probabilistic Rand Index (PRI) iii. Variation Of Information (VOI)

3.1 Global Consistency Error It is a Region-based Segmentation Consistency, which measures to quantify the consistency between image segmentations of differing granularities. It is used to compare the results of algorithms to a database of manually segmented images. Let S and S’ be two

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org segmentations as before. For a given point xi (pixel), consider the classes (segments) that contain xi in S and S0. These sets are denoted in the form of pixels by C (S, xi) and C (S0, xi) respectively. Following [5], the local refinement error (LRE) is then defined at point xi as: LRE ( S , S ', xi ) 

| C ( S , xi ) \ C ( S ', xi ) | | C ( S , xi ) |

(5)

Global Consistency Error (GCE) forces all local refinements to be in the same direction and is defined as: GCE ( S , S ') 

1 min  LRE ( S , S ', xi ),  LRE ( S ', S , xi ) N

(6) It measures the extent to which one segmentation can be viewed as a refinement of the other. Segmentations that are related in this manner are considered to be consistent, since they could represent the same natural image segmented at different scales.

3.2 The Probabilistic Rand Index (PRI) Rand Index is the function that converts the problem of comparing two partitions with possibly differing number of classes into a problem of computing pair wise label relationships. PRI counts the fraction of pairs of pixels whose labelling are consistent between the computed segmentation and the ground truth, averaging across multiple ground truth segmentations to account for scale variation in human perception. It is a measure that combines the desirable statistical properties of the Rand index with the ability to accommodate refinements appropriately. Since the latter property is relevant primarily when quantifying consistency of image segmentation results. Consider a set of manually segmented (ground truth) images {S1, S2, . . . , SK} corresponding to an image X = {x1, x2, . . . xi, . . . , xN}, where a subscript indexes one of N pixels. Stest is the segmentation of a test image, and then PR Stest , Sk  

1 I liStest  l Sj test pij  I liStest  l Sj test 1 pij     N   i, j    i j 2









1 K I lik  lkj  K k 1 PRI is defined as

pij  P li  l j  

(7)

25

This measure takes values in [0, 1] – 0 when S and {S1, S2, . . . , SK} have no similarities and 1 when all segmentations are identical (i.e. when S consists of a single cluster and each segmentation in {S1, S2, . . . , SK} consists only of clusters containing single points, or Vice versa).

3.3 The Variation of Information (VOI) It measures the sum of information loss and information gain between the two clustering, and thus it roughly measures the extent to which one clustering can explain the other. The VOI metric is nonnegative, with lower values indicating greater similarity. It is based on relationship between a point and its cluster. It uses mutual information metric and entropy to approximate the distance between two clustering across the lattice of possible clustering. More precisely, it measures the amount of information that is lost or gained in changing from one clustering to another (and, thus, can be viewed as representing the amount of randomness in one segmentation which cannot be explained by the other). The variation of information is a measure of the distance between two clustering (partitions of elements). A clustering with clusters X1,X2, , , , ,Xk is represented by a random variable X with X={1 ….K} such that pi | X i | / n iЄX and n  i X i the variation of information between two clustering X and Y so represented is defined to be: VI ( X , Y ) : H ( X )  H (Y )  2 I ( X ; Y )

(8)

where H(X) is entropy of X and I(X,Y) is mutual information between X and Y. VI(X,Y) measures how much the cluster assignment for an item in clustering X reduces the uncertainty about the item's cluster in clustering Y.

4. Experiments and Results The segmentation methods discussed above are applied to a set of bench mark images. Annexure-I shows the segmentation results of four methods applied on twenty-five dataset images. These three metrics discussed are calculated for the four mentioned segmentation methods .Figure 1 show the application of sample IR image segmentations .

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org Figure 1: Application of segmentation methods over an IR landmine image

Figure 3: Evaluation graph using Global Consistency

Table 1: Sample performance measures for the segmentation results of four methods on IR landmine images presented in Figure 1.

Variation Of Information

17.2

Probabilistic Rand Index

1.1 1.2 1.3 1.4

0.9928 0.9953 0.9933 0.9945

Global consistency Error 0.9935 0.9971 0.9943 0.9959

16.8 16.6 16.496736

Variation of information

16.4

Probabilistic randIndex

0.9955 0.994948 0.995 0.994508

15.8 15.6 Gacm Variation Of Information


0.993376

0.9935 0.993

5. Conclusions Since segmentation is the step important for object recognition, it is necessary to find out the best algorithms suitable for IR images. In this paper, four different segmention algorithms are experimented for set of IR images and some of X-ray and GPR images. Performance evaluation of segmented images showed that under Global active contour / snake model exhibit better performances for above said images. respectively. The experiment are conducted using matlab tool.

References [1]

Gacm

watershed

fuzzy

otsu

0.994508 0.994216 0.993376 0.994948

[2]

Figure 2: Evaluation graph using Probabilistic Rand Index

[3] Global Consistency Error

0.997 0.99644 0.9965

0.996284

[4]

0.996 0.995416

Global Consistency Error

0.99502 0.995

0.9945

[5] 0.994 Global Consistency Error

Otsu

16.49674 16.95031 16.19265 16.9856

Figure 4: Evaluation graph using Variation Of Information

0.9925

0.9955

Watershe Fuzzy C d

0.994216

0.994


Variation Of Information

16

15.8120 17.4327 16.0488 16.6339

Following figures 2,3,4 give comparative performance measure of four segmentation algorithms using the three evaluation parameters. From this evaluation, it is found that Global Active Contour / Snake Model segmentation is well suited for the IR images. The PRI value should be higher for an image and VOI, GCE values must be lower for an image [14].

16.192652

16.2

Table 1 shows the parameter values of different segmentation of single image in figure 1.

0.9945

16.985604

16.950308

17

Images& Model

26

Gacm

Watershed

Fuzzy C

Otsu-5

0.995416

0.99644

0.99502

0.996284

S. R. Kannan, "Segmentation of MRI Using New Unsupervised Fuzzy C-Means Algorithm" ICGST-GVIP Journal, Vol. 5, Issue 2, Jan. 2005 Jin WU, Juan LI Jian LIU, Jinwen TIAN Infrared “Image Segmentation via Fast Fuzzy C-Means with Spatial Information” Proceedings of the 2004 IEEE International Conference on Robotics and Biomimetics, page 742-745, August 22 - 26, 2004. Du Gen-yuana,b,c, Miao Fanga,c, Tian Sheng-lib, Liu Yeb , “A modified fuzzy C-means algorithm in remote sensing image segmentation” International Conference on Environmental Science and Information Application Technology IEEE,page 447-450 , Issue 10,Nov.2009. Xin-Bo ZHANG, Li JIANG “An Image Segmentation algorithm Based on Fuzzy C-Means Clustering” of Information and Electronic Engineering,ZheJiang Gongshang University Hangzhou, zhejiang,china, International Conference on Digital Image Processing,2009 IEEE,PAGE 22-26, Issue 10,Nov,2009. Aseel Ajlouni And Alaa Sheta ,“Landmine Detection With IR Sensors Using Karhunen Loeve Transformation And Watershed Segmentation”, Information Technology Department, Ai-Balqa Applied University Salt, 5th International Multi-Conference on Systems, Signals and Devices,IEEE-2008.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org [6]

[7]

[8] [9]

[10]

[11]

[12]

[13]

[14]

[15]

Malik Sikandar Hayat Khiyal, Aihab Khan, and Amna Bibi, “Modified Watershed Algorithm for segmentation of 2D images” Issues in Informing Science and Information Technology,page 879-886, Volume 6, 2009. Jun Zhang and Jinglu Hu ,“Image Segmentation Based on 2D Otsu Method with Histogram Analysis” International Conference on Computer Science and Software Engineering,IEEE,page 105-108, Issue 10,Nov,2009. DongjuLiu, JianYu, “Otsu method and K-means”, Ninth International Conference on Hybrid Intelligent Systems.issue 10-11-2009,page 344-349 ,IEEE,2009 Xavier Bresson· Selim Esedo¯glu · Pierre · Jean-Philippe Thiran · Stanley Osher “Fast Global Minimization of the Active Contour/Snake Model” Published online: 14 July 2007 Springer Science + Business Media, page 151-167, 2007. Xavier Bresson and Jean-Philippe Thiran “Image Segmentation Model Using Active Contour And Image Decomposition “Signal Processing Institute (ITS), Swiss Federal Institute of Technology of Lausanne (EPFL) CH1015 Lausanne, Switzerland Xavier.Bresson,.2005 Kingsford, Carl (2009). "Information Theory Notes" (PDF). ttp://www.cs.umd.edu/class/spring2009/cmsc858l/InfoThe oryHints.pdf. Retrieved 22 Sept. 2009. Ranjith Unnikrishnan , C. Pantofaru Martial Hebert “A Measure for Objective Evaluation of Image Segmentation Algorithms” The Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213 , 2005 IEEE M.Sujarithaa and S. Annaduraib, “Bayesian Colour Image Segmentation using Pixon and Adaptive Spatial Finite Mixture Model” ICGST-GVIP Journal, Volume 9, page 45-52,Issue 5, September 2009, ISSN: 1687-398X. M.sujaritha ,S.Annadurai “Bayesian Colour Image Segmentation using Pixon and Adaptive Spatial Finite Mixture Model”,ICGST-GVIP Journal, Volume 9, Issue 5, September 2009, ISSN:1687-398X. Fernando C. Monteiro1, and Aurlio C. Campilho1, “Performance Evaluation of Image Segmentation” Springer-Verlag Berlin Heidelberg 2006, ICIAR 2006, LNCS 4141, page 248–259.

27

Dr. Padmavathi Ganapathi is the Professor and Head of Department of Computer Science, Avinashilingam University for Women, Coimbatore. She has 21 years of teaching experience and one year Industrial experience. Her areas of interest include Network security and Cryptography and real time communication. She has more than 50 publications at national and International level. She is a life member of many professional organizations like CSI, ISTE, AACE, WSEAS, ISCA, and UWA. Dr. Subashini is the Associate professor in Department of Computer Science, Avinashilingam Deemed University for Women, Coimbatore. She has 16 years of teaching experience. Her areas of interest include Object oriented technology, Data mining, Image processing, Pattern recognition. She has 55 publications at national and International level. Ms.A.Sumi received MCA Degree in Kongu Arts and Science college, Erode in 2002 and M.phill Degree from Bharathiyar University, Coimbatore in 2007 respectively. She has 2 years of teaching experience and currently working as a research staff in Department of Computer Science in Avinashilingam Deemed University for women. Her research interests are Image processing, scripting and Networking.


Annexure BENCH MARK IMAGES TAKEN FOR STUDY AND THE RESULTS OF FOUR ALGORITHMS

28


29

30


Numerical Analysis of the DQPSK Modulation Formats Implementation With 40 Gbits/s Hadjira Badaoui1, Yann Frignac2, Petros Ramantanis2, Badr Eddine Benkelfat2 and Mohammed Feham1 1

Laboratoire STIC, Département de Génie Electrique Faculté de Technologie, Université Abou-Bekr Belkaïd -Tlemcen BP 230, Pôle Chetouane, 13000 Tlemcen- Algeria 2

Institut TELECOM & Management SudParis, 9, rue Charles Fourier - 91011 Evry Cedex - France

Abstract

performances of such a transmission system of which we are obliged to test it by a digital simulation.

This paper treats the performances of an optical telecommunication system functioning with the 40 Gbit/s flow by a numerical analysis of the physical effects limiting the light guided propagation. The format studied here is a DQPSK modulation format (Differential Quaternary Keying Phase-Shift). It allows a transmission of information on four different levels of phase of the optical signals. Taking into account the complexity of the communication optical systems and joint action of many propagation physical effects (linear or non-linear), the optimization of the systems functioning with DQPSK modulation formats must be apprehended beforehand by digital simulations to direct the choices of the future designs. It is a study on a transmission with a single channel on only one optical fiber section. In particular, an analysis must be led concerning the impact on the transmission quality of the non-linear Kerr effects combined with the chromatic dispersion inherent of the propagation on optical fiber. The most effective technique to thwart the harmful impact of these effects is the management of chromatic dispersion. The obtained results will be presented and discussed. Keywords: Optical fiber, optical transmission systems, 40 Gbit/s flow, in phase modulation formats, OCEAN, numerical communications.

In the case of a transmission system of single channel, the step dz is taken equal to (1/100) the Kerr length.

1. Introduction

3. Optical Fiber Digital Transmission System

The use of optical fiber completely revolutionized the telecommunications world. One arrives from now on, at increasingly powerful transmission systems reaching several Tbit/s on several thousands of kilometers. Current research in this field aims at improving the systems performances multiplexed in wavelength Nx40 Gbit/s to even make them evolve to Nx100 Gbit/s systems [1]. Our objective relates on studies of the physical aspect of telecommunications optics of only one channel and to only one fiber section. This allowed us to estimate the

The principal objective of a numerical system is to ensure the transmission without information error in the numerical form of a transmitter towards a receiver distant from a certain distance. We present on the Fig. 1 [3-5], the optical transmission systems diagram followed to carry out the digital simulations of the transmissions on only one fiber SSMF, while using the simulation program OCEAN [6]. The noted transmitter Tx generate a single channel, centered over the 1550 nm wavelength, modulated to 40 Gbit/s and modeled by a quaternary sequence of 1024 symbols. The modulation format used is DQPSK.

2. Simulations Tool As regards the guided propagation simulation of an optical signal, we use OCEAN software « Optical Communication Emulator for Alcatel Network». It is based on the iterative Split-Step Fourier Method SSFM. The SSFM method principle consists of simulating the propagation of a visible signal on successive fiber portions where the dispersions effects and the nonlinear effects could be regarded as independent one of the other [2]. The SSFM method is an optimization method. The optical fiber is cut out in infinitesimal sections length dz. The length of these sections, defined according to the fiber Kerr, being. Where: 1 (1) L  Kerr

Psignal  


Laser source 1550 nm Receiver

31

The Fig. 3 illustrate an ideal case of a temporal evolution of a DQPSK signal with 40 Gbit/s for a number of symbols equal to 40 and each symbol lasts a time T equal to 46.49 ps.

Single optic

Tx

Pre‐compensation

fiber section

Post‐compensation

Rx

Fig. 1 General diagram of the digital simulations on only one optical fiber section

7

Phase Module

6 5.4977 5

4. Differential Quaternary Phase-Shift Keying modulation format The differential modulation in phase DQPSK for (Differential Quaternary Keying Phase-Shift) is a modulation format which comprises four different levels of phase. Moreover, the signal intensity coded in DQPSK remains constant in time, except on the level of the transitions from phase where reductions in intensity are observed in certain transmitter assemblies. Each differential of phase codes on a group of two bits, sometimes called dibit or symbol, to choose among four: « 11 », « 10 », « 00 » ou « 01 ». Fig. 2 recapitulates this attribution of debits according to the differential of phase and clarifies it by a representation on the trigonometrical circle.

4 3.9269 3 2.3561 2

1 0.7853 0 40

Time-symbol

(a) 90

1

120

0.8

60

0.6

150

30

0.4 0.2

90° (01)

(00) 0

180° (11)

180

0

330

210

(10) 270°

Fig. 2. Representation on the trigonometrical circle of attribution

300

240 270

(b) Fig. 3 (a) Temporal evolution of DQPSK signal with 40 Gbit/s (b) Diagram in constellation of DQPSK signal with 40 Gbit/s

4. Simulation results 4.1. Linear physical effects results Initially, we present the simulation results without transmission. For a numerical signal with 40 Gbit/s and optically modulated by the means of a modulation on four levels, the optical signal will transmit only 20 Gsymbols/s. Its modulation rate will be then 20 Gbaud, and its information frequency of 20 GHz.

It is noticed that this signal is coded on four levels of phase and its amplitude is quasi constant. The return to zero, are carried out on the level of each transition from different phase is 0.7853 [rd] towards 3.9269 [rd] or 2.3561 [rd] towards 5.4977 [rd]. In the same way, one can trace this signal in the complexes plan (Fig. 3 (b)). In a more general way, one can locate each symbol in this diagram by its amplitude and its phase. This symbol takes four different states are « 00 », « 01 », « 10 », and « 11 » and each states can be associated a PRQS sequence

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org symbol [7-8]. The simulated amplitude is standardized with a mathematical value of 1. In the second part of simulations, we will introduce only the chromatic dispersion effect to our preceding optical signal. In the continuation of this paper, if nothing is specified, when the chromatic dispersion term is employed, it implicitly refers to cumulated dispersion. The Fig. 4 illustrates the variation of the module of a temporal DQPSK signal evolution with 40 Gbit/s under the chromatic dispersion effect.

32

One gathers this amplitude and phase variation in the introduction of the constellation illustrated by the Fig. 5. Indeed, it makes it possible to locate certain effects by their characteristic signature. This last show a chromatic dispersion equalizes with -32 ps/nm

1.0

0.8

0.6

0.4

0.2 0 0

10

20

30

40

Time-symbol

(a)

(a) 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 0

10

20

30

40

Time-symbol

(b) Fig. 4 (a) Module of a temporal DQPSK signal evolution without dispersion (b) Module of a temporal DQPSK signal evolution with dispersion

One represents this temporal DQPSK signal evolution by the fact that a light wave modulated with a certain non null spectral width, has spectral components which are not propagated at the same speed. This generates a spectral dephasing which results in a temporal deformation from the optical signal.

(b) Fig. 5 (a) Constellation transitions without dispertion of a DQPSK signal with 40 Gbits/s (b) Constellation transitions with dispertion of a DQPSK signal with 40 Gbits/s

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org The physical interpretation of such a deformation of the constellation is shown in Fig. 5 by using a reasoning "time-frequency". Let us consider the spectral analysis of the sampled signal during its evolution between two transitions. The spectrum of this temporally truncated signal will have energy in the low frequencies, also truncated temporally but around a transition will have energy in higher frequencies. The phase variation due to the chromatic dispersion will be thus more significant for the transitions "rich" in high frequencies than for the symbols themselves what represents the phase rotation into spiral of the arms of transition from the constellation and stability in phase of the states. The visual anomaly on the constellation of the Fig. 5 for the state located at a phase of 315° comes owing to the fact that for this constellation only 40 symbols are represented on the 1024 symbols of the complete signal. The temporal variations of the optical signal characteristics consecutive to its modulation result in a certain width of its spectrum, in the spectral domain. i  U ( z ,  )  U ( z ,0). exp  2 2 z  2 

(2)

Whith :  2 is the dispersion the group speed given by the equation (3). 2  

2 Dcum 2  c

(3)

U(z,0) is the signal spectrum without chromatic dispersion Fig. 6 shows the evolution of a modulated optical DQPSK signal versus frequency with 40 Gbit/s. 6000

5000

4000

3000

2000

0

6000

5000

4000

3000

2000

1000

0 1.211 1.212 1.213 1.214 1.215 1.216 1.217 1.218 1.219 1.22

1.211 1.212 1.213 1.214 1.215 1.216 1.217 1.218 1.219 1.22 1.221 x 1015

(a)

1.221

x 1015

(b) Fig. 6 (a) DQPSK signal evolution versus frequency (b) Dispersed DQPSK signal evolution versus frequency

4.2. Non-linear physical effects results In this section we present the non-linear physical effects results, in particular, the influence of the power on the propagation at the flow of 40 Gbit/s with DQPSK modulation (simulations with only one channel). We previously saw that, more the channel power increases, more the non-linear effects appear. In particular, for transmissions with single channel without noise only the automodulation of phase (SPM, Self Phase Modulation) intervenes. To characterize the non-linear effects accumulation, one can define two sizes which are as follows: Integrated power Pint , equal to : Pint  Nsections . Pin (4) Where : Pin is the input average power of the signal of a sigle channel, Nsections is the sections number of an optical transmission system. Non-linear phase measured in radian [rd] : when the propagation takes place on only one fiber section, the dephasing formula corresponding to the non-linear phase can be simplified in:

 nl   .Pin .Leff

1000

33

(5)

Where :  is the fiber non-linear coefficient [ w 1 km 1 ] that one finds in the equation of Schrödinger non-linear in fiber used, Pin is the channel injected power into, Leff is the fiber effective length.

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org To illustrate the presence of Kerr effect, best is to use the constellation representation of an optical signal modulated on four levels of phase. The Fig. 7 show two constellation examples without the transitions of a DQPSK modulation format (Differential Quaternary Keying Phase-Shift).

1 45

135

34

4.3. Influence of the Propagation Effects on the Quality Factor We present here the simulation results and, in particular, the variation of the quality factor according to the maximum pre-compensation values. Let us take for example the case of transmission of a pseudo-random quaternary sequence [9-10] on a distance of 100 km of fiber with same chromatic dispersion as the Teralight fiber (Alcatel) (Fig. 8 (a)) and that of SMF (Fig. 8 (b)) for which the input power in the section is fixed at 4 dBm. One observes that there is a negative optimal precompensation value and, all the more strong in absolute value that the fiber dispersion is high. 1 PRQS

225

315

Quality factor

0.98

SPM

0.96

0.94

0.92

0.9

0.88 -600

 NL

-500

-400

-300

-200

-100

0

Maximum pre‐compenstion [ps/(nm)] (a) 0.99 PRQS

(a)

(b)

Fig. 7. Effect of the SPM on a signal modulated in phase for: (a) Dcum=4 ps/nm, precompmax=-75 ps/nm, and Pin=0 dBm (b) Dcum=4 ps/nm, precompmax=-75 ps/nm, and Pin=12 dBm

Values of phase initially taken of (

 3 5 7 ,

,

,

Quality factor

0.98

0.97

0.96

0.95

0.94

0.93

) will

4 4 4 4 evolve of a value equal to the non-linear phase  NL . When the injected input power in the optical transmission system, each of the four symbols is seen affected of an amplitude and phase fluctuation. Moreover the non-linear phase varies with the power. This effect degrades the information transport quality which will be able to prove penalizing for the optical signal propagation.

0.92 -600

-500

-400

-300

-200

-100

0

Maximum pre‐compenstion [ps/(nm)]

(b) Fig. 8 The influence of the pre-compensation on the quality factor: (a) Dcum=8 ps/nm, precompmax=-150 ps/nm, and Pin=4 dBm (b) Dcum=16 ps/nm, precompmax=-200 ps/nm, and Pin=4 dBm


4.4. Comparison Between two Types of Sequences Partially Different in Term of Quality Factor To evaluate the variation of the transmission quality according to the values of the pre-compensation, we will schematize two cases corresponding to the two types of sequences. 1 0.995

Pseudo-random quaternary sequence Quaternary sequence

Quality factor

0.99 0.985 0.98 0.975 0.97

ps/nm. As in session 4.3, one takes two examples of transmission on 100 km of Teralight fiber (Fig. 8 (a)) and of fiber having like characteristic a dispersion equal to 32 ps/(nm.km) to 1550 nm (Fig. 8 (b)). While fixing the input power equal to 0 dBm (1 mW on a linear scale), the maximum pre-compensation is always negative, larger in absolute value if dispersion is stronger. In our example, for dispersion equal to 32 ps/nm the maximum precompensation is of 250 ps/nm in absolute value which is larger only in the case of a Teralight fiber (Fig. 8 (a)). In the first case, that which is showed in Fig. 9 (a), notices that the quality factor equal to 0.995, is as good as that of Fig. 9 (b). In the continuation, one focuses in the same reasoning but this time, one is interested in the results allowing to give a vision on the influence of the power on the quality factor.

0.965

1

0.96 0.955 -500

-400

-300

-200

-100

0


(a) 0.99 0.989


0.99

Quality factor

0.95 -600


0.988

0.98

0.97 0.96 0.95 0.94 -600

-500

-400

-300

-200

-100

0

0.987


0.986 0.985

(a)

0.984

0.983

0

0.982

-10

0.981 -600

-500

-400

-300

-200

-100

0

Maximum pre‐compenstion [ps/(nm)] (b) Fig. 9 Comparison between two types of sequences in term of quality factor : (a) Dcum=8 ps/nm, precompmax=-150 ps/nm, and Pin=0 dBm (b) Dcum=32 ps/nm, precompmax=-250 ps/nm, and Pin=0 dBm

Quality factor

Quality factor

35

-20 -30 -40 -50 -60

-80 -600

Fig. 9 illustrates that there is not a great variation in term of quality factor between a pseudo-random quaternary sequence PRQS and another sequence which does not present all the properties that of a PRQS sequence but this variation is more significant when the local dispersion of fiber is strong and the pre-compensation lower than -200


-70 -500

-400

-300

-200

-100

0


(b) Fig. 10 Comparison between two types of sequences in term of quality factor (a) Dcum=4 ps/nm, precompmax=-75 ps/nm, and Pin=0 dBm (b) Dcum=4 ps/nm, precompmax=-50 ps/nm, and Pin=20 dBm

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org Fig. 10 (a) shows that the quality factor of a fiber known as (Broad Effective Area Fiber) is equal to 0.9952 when the power is 1 mW. In the worse case i.e. for a power equalizes to 20 dBm, one notice that the values of the quality factor take negative values, note that it has a value of -0.3369 for a maximum pre-compensation equalizes with -50 ps/nm.

4.4. Treatment on the Maximum Pre-compensation The chromatic dispersion management utilizes unfortunately a great number of parameters and the joint influence of dispersion and the non-linear effects make its optimization vast and complex. Fig. 11 shows the influence of chromatic dispersion on the maximum precompensation for various values of power going from 0 dBm to 20 dBm. Notices indeed that for a fiber having a cumulated dispersion to 4 ps/nm equalizes, the maximum pre-compensation is of -75 ps/nm and remains identical for various values of power except in the case of a power of 20 dBm. Maximum precompensation [ps/(nm)]

0 -25

P=0 dBm P=4 dBm P=8 dBm P=12 dBm P=16 dBm P=20 dBm

-50 -75 -100 -125 -150 -175 -200 -225 -250

0

4

8

12

16

20

24

28

32

35

Local dispertion [ps/(nm.Km)] Fig. 11 Influence local dispersion of fiber on the maximum precompensation

5. Conclusions In this paper, we carried out a simulation of transmission at 40 Gbit/s by channel, which proves now imminent for the terrestrial systems. The numerical analysis of the propagation effects taught us that the introduction of the nonlinear effects of Kerr type particularly by the SPM implies a rotation of the phase proportional to the power of the signal, and that chromatic dispersion results in a fluctuation of phase to each transition. During all the carried out studies, the chromatic dispersion management presents the heart of the found results. This technique enabled us to control not only the non-linear dispersion but also effects. To ensure an optimal transmission within a transmission system, it is well necessary to adjust several

36

parameters such as chromatic dispersion, the input power and the compensation along the system. Acknowledgments This work was made possible through collaboration between the institute & TELECOM Management Sud Paris (old INT, Institut National des Telecommunications) and the Faculty of Technology, Department of Electrical engineering of the University of Tlemcen. The objective of this research is included in the national project initiated by France of the teams Research Alcatel-Lucent. References [1] Y. Frignac, J. C. Antona, S. Bigo and J. P. Hamaide, « Numerical optimization of pre- and in-line dispersion compensation in dispersionmanaged systems at 40Gbit/s », OFC'02, ThFF5, Anaheim, California, March 17-22. [2] L. K. Wickham et al., “Bit pattern length dependence of intrachannel nonlinearities in pseudolinear transmission”, IEEE Photon. Tech. Letters, Vol. 16 n° 6, june 2004. [3] G. Keiser, “Optical fibre communication system,” second edition, John Wiley & Sons, Inc., New York, 1997. [4] G.P. Agrawal, “Fibre-Optic Communication Systems,” 2nd edition, John Wiley & Sons, Inc., New York, 1997. [5] G. Agrawal, Non Linear Fiber Optics, Academic Press, 1995. [6] OCEAN « Optical Communication Emulator for Alcatel Network». [7] B. Spinnler et al.., “Performance assessment of DQPSK using pseudo-random quaternary sequences.”, in proc. ECOC 2007, Berlin, Germany. [8] P.RAMANTANIS, H.BADAOUI and Y.FRIGNAC, “Quaternary sequences comparison for the modeling of optical DQPSK dispersion managed transmission systems”, The Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference OFC/NFOEC 2009, March 22–26, 2009, California, U.S.A. [9] J. C. Antona et al., “Revisiting Binary Sequence Length Requirements for the Accurate Emulation of Highly Dispersive Transmission Systems”, ECOC’08, Brussels, Belgium, 22-25 sept. [10] F. J. MacWilliams and N.J.A. Sloane., “Pseudorandom sequences and arrays.”, Proc. of IEEE Vol. 64, n°12, pp.1715–1729, 1976.


37

N-Dimensional Self Organizing Petrinets for Urban Traffic Modeling Manuj Darbari*, Vivek Kumar Singh*, Rishi Asthana* and Savitur Prakash # *Babu Banarasi Das National Institute of Technology and Management Lucknow A-649, Indira Nagar Lucknow-226016, Uttar Pradesh, India

#Doordarshan Kendra (T V Station), Lucknow, India 268 Uphar ELDECO-1, Post- Bhadrukh, Lucknow-226002, Uttar Pradesh, India

Abstract This paper highlights the Urban Traffic Simulation using causal petrinets. The focus of this paper is to generate a producer consumer network of networks and have grid simulation of Petrinets. The paper suggests a grid petrinet model with self organizing producerconsumer nets. Keywords: Self organizing Network, Causal Producer – Consumer Petrinets

1. Introduction 1.1 Problem Statement Traffic Management System addresses the objective of reducing congestion, vehicle delay time, fuel consumption and pollution. The most common technique to regulate and manage Urban Traffic is grouped into two classes: 1. Fixed-Time System. 2. Traffic Responsive System. The first group has fixed on-off time periods for traffic flow. The second group employs actuated signal timing plans and performs an on-line optimization and synchronization of traffic signals. The real sense sensors/ detectors located on traffic intersection, which feed information on, the actual system state to the real time controller. To achieve traffic control using these strategies Traffic Network has to be appropriately modeled for simulation purpose.

2. Literature survey The dynamics of Urban Traffic systems depends on the complex interactions of the timing of various discrete events, such as arrivals or departures of vehicles at the intersections and beginning or completion of the various phases in the signal timing plans of the traffic lights as stated by Tzes, Kiran and Mc Shane [8]. An example of colored PNs model of Traffic Light was first proposed by Jensen [6]. Later on Tzes, Gallego and Farges Henry [4] have modified the model of Jensen which shares the idea of adjusting signal controlling according to the distinct tokens deposited in PN controller. The latest paper by List and Cetin [5] discuses the use of PNs in modeling traffic signal controls and performs a structural analysis of the control PN Model by P- invariants, mainly focusing safety rules. Similarly, Di Febbraro [1, 2, 3] et al presented a model in a timed PN framework, where token are vehicles and places are part s of lanes and intersections. We will be using all these models with modifications with respect to the behavioral aspects of Indian Traffic conditions. The focus will be on spatio-temporal relation of Traffic movement.

3. The model The Model is divided into two basic parts: The forward propagation petrinet model which focuses on simple cause effect framework.The second part focuses on continuous Petrinet Learning mechanism.

3.1 Forward Propagation of Petri nets


We will discuss each of these frameworks with references to Indian Traffic Movement Conditions. This model represents the flow of Traffic based on causal runs where: 'e' represents the places. 't' represents the Transitions. 'q' represents the flow of token For a simple situation of a roadblock we can have a cause and effect relationship shown in figure 1.

38

Consider this producer - consumer as one single unit denoted by ‘X’. This main - road itself is a part of the entire road grid of the city. Assuming that there is some pattern of traffic movement, which is present at various interval of time. Applying Unidirectional Petrinet Model with tokens moving in Feed Forward mode as shown in the figure 4.

t

Fig. 1: Simple Petrinet Model

Event e1 can cause event e2 with a transition‘t’. Considering a situation when multiple road jams at the same instances and the traffic by-lanes exists points connected to a single main road.

Fig. 4: Dynamic Producer Consumer with Buffer Interlocked with Control Center

The Algorithm for Dynamic Producer Consumer is given as:

Fig. 2: Road Network

Since these by lanes have an outflow at the main road the flow of traffic is drastically reduced as the vehicle outflow from various by-lanes starts pouring into the main road due to congestion in other parts. Simple producerconsumer nets can represent the workflow of the entire traffic dynamics. (Figure 3).

Step 1: Initialize each of the Producers- Consumers situation (x). Set the pattern learning rate as ''. Step 2: Set the control center such that: Xi = Si is achieved. Step 3: Let the Token release rate is defined as 1/N; where N is defined as the number of producer consumer initial states. Step 4: The release of Token are updated as: x: (producer- old) = xi (producer - new state) + r (pre- post) Where 'w' is the bulk arrival rate of Tokens. Step 5: Stop when system has transferred all the tokens and traffic reaches a balancing state.

3.2 Grid Model:

Fig 3. Causal Run of Producer - Consumer of Traffic Flow

Now consider a situation when we consider the producer consumer network to be trained and the combination of Control Center and Producer - Consumer as one single unit denoted as 'Y' for the first layer, 'Z' for the second and 'N' for the third layer. It can be shown by the help of two dimensional grid of

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 ISSN (Online): 1694-0784 ISSN (Print): 1694-0814 39

producer - consumer link with buffer interlocked with Virtual Counter which keeps the track of diversion of Token based on the congestion in the road network (i.e. consumer) and supply of new lanes ( i.e. producers) and diversion of traffic through VMS. The counter (M) keeps track of tokens Number stamped by the layer through which is it is being generated and finally arriving at particular layers.

Where 'n' is the layer - n to which token moves. Step 4: For each of the movement of token freeze the producer - consumer network which has become stable and these are no more to be initiated. Step 5: After all the forward steps has been performed. Log the virtual counter to current state. Step 6: Move in the reverse direction this time consumer driving the producer for necessary action calming. Step 7: This Forward and Backward movement of Tokens takes place till the time entire network is Self- Organized and further flow of token is not required within a particular time frame. Step 8: Test and store the stopping condition. The key idea to have a self organizing causal grid net is to find the heavy bottleneck situation in the entire city traffic network and then flow of token starts layer - bylayer to achieve the final stability. The grid works in two modes: (i) The Initial formation of back- forward flow of tokens. (ii) Convergence to a stable state.

4. Conclusion and future scope

Fig. 5: Grid Network of Producer- Consumer Structure

The purpose of showing the arrow in both the direction has special significance, during the first instance the token moves in forward learning mode and this state is now stored in Virtual Counter. During the next cycle there is a change in sequence will behave in vice-versa mode the road network which is now having heavy traffic movement will eventually settle down with lesser frequency of traffic with time. This requires a signal from the consumer end to the producer end to check the new status. This process of adjustment and training of producerconsumer from their nearest neighbors goes on till the entire system becomes stable.

3.3 Training Algorithm: Step 1: Set the Virtual Counter to value; set the stability session the grid has to perform. Step 2: While stopping condition is false; perform step 3-8. Step 3: For each layer the compute the complexity of link of the token movement is given as: x (n) =  (Layer 1 - Layer n) 2

The paper provides a real time dynamic simulation of Urban Traffic System using the concept of Causal Producer Consumer Theory. It extends the producer-consumer theory to N- dimensional representation where the focus to provide a buffer mechanisms and virtual counter to keeps track of the entire token flow across the grid. The future extension of our work will be simulation of our network using Petrinet Simulator to judge its efficiency with training of each Producer- consumer unit as a single entity. We will try to keep the token time stamped and then analyze how soon the network self organizes itself.

References 1. Di Febbraro, A., & Sacco, N. (2004). On modeling urban transportation networks via hybrid Petri nets. Control Engineering Practice, 12(10), 1225-1239. 2. Di Febbraro, A., & Sacco, N. (2004). An urban traffic control structure based on hybrid Petri nets. IEEE Transactions on Intelligent Transportation Systems, 5(4), 24-237. 3. Di Febbraro, A, Giglio, D, & Sacco, N. (2002). On applying Petri nets to determine optimal offsets for coordinated traffic light timings. Proceedings of the 5th IEEE International Conference on Intelligent Transportation Systems , Singapore (pp. 87-706). 4. Gallego, J.-L., Farges, J.-L., & Henry, J.-J. (1996). Design by Petri nets of an intersection signal controller. Transportation Research Part C, 4(4), 231-248.


5. List, G. F., & Cetin, M. (2004). Modeling traffic signal control using Petri nets. IEEE Transactions on Intelligent Transportation Systems, 5(3), 177-187. 6. Jensen, K. (1992). Colored Petri nets: basic concepts, analysis methods and practical use, Vol. 1. New York: Springer. 7. List, G. F., & Cetin, M. (2004). Modeling traffic signal control using Petri nets. IEEE Transactions on Intelligent Transportation Systems, 5(3), 177-187. 8. Tzes, A., Kim, S., & McShane, W. R. (1996). Applications of Petri networks to transportation network modeling. IEEE Transactions on Vehicular Technology, 45(2), 391-400.

Dr. Manuj Darbari is currently working as an associate professor in the Dept. of Information technology at B.B.D.N.I.T.M (Babu Banarasi Das National Institute of Technology And Management), Lucknow. He holds a PhD from Birla Institute of Technology Mesra Ranchi, India and having a teaching experience of more than eleven years. Prior to his current assignment he has taught for one year In M.N.R.E.C Allahabad as lectur‐ er and ten years in B.B.D.N.I.T.M. Lucknow in different positions. He has published fifteen papers in referred international and national journals. He is selected for marquis who’s who in science and engineering 2003‐2007. His teaching areas are information science, ERP, software engineering, workflow management. Vivek Kumar Singh is currently working as an assistant professor in the Dept. of Information technology at B.B.D.N.I.T.M (Babu Banarasi Das National Institute of Technology And Management), Lucknow. He holds a teaching experience of more then seven years. He has published ten papers in referred international and national journals. His teaching areas are Automata Theory, Algorithms, software engineering, MIS. Rishi Asthana is currently working as an associate professor in the Dept. of Electrical and Electronics Engineering at B.B.D.N.I.T.M (Babu Banarasi Das National Institute of Technology And Management), Lucknow. He holds a PhD from Petroleum University, Dehradun,India and having a teaching experience of more than fifteen years. Prior to his current as‐ signment he has taught for five years In Pauri Garhwal as lecturer and ten years in B.B.D.N.I.T.M. Lucknow in different positions. He has pub‐ lished fifteen papers in referred international and national journals. His teaching areas are Electrical Power system, system modeling, soft com‐ puting. Savitur Prakash is presently working in the Dept. of Engineering, ‘All India Radio and Doordarshan’ under the ministry of Information and Broadcasting, Government of India and presently posted at Doordarshan Kendra (Television station) Lucknow. He is having fourteen years of services and started his carrier from All India Radio Aizawl (Mizoram), India in 1996 and since then served various places of India, and his cur‐ rent research area is soft computing, wireless communication, Network management, supply chain management, Total Quality Management and Network security. He has finished his M.Tech in Computer Engi‐ neering and holds a Two year Post graduate diploma in Business Admin‐ istration with specialization in Operations management from Symbiosis Centre Pune, India.

40


41

Domain Driven Data Mining – Application to Business Adeyemi Adejuwon and Amir Mosavi University of Debrecen, Faculty of Informatics, Egyetem ter1, Debrecen 4032, Hungary

Abstract Conventional data mining applications face serious difficulties in solving complex real-life business decision making problems when practically deployed. This work in order to improve the operations in a collection of business domains aims to suggest solutions by reviewing and studying the latest methodological, technical, practical progresses and some cases studies of data mining via domain driven data mining (DDDM). The presented paper tries to answer this question: “what can domain driven data mining do for real-life business applications?” Moreover this work attempts to provide information and abilities to fill the existing gap between academic researches and real-world business problems.

the consumer [3]. Through a balanced mix of economic theories and IT, businesses can effectively devise appropriate strategies. The effective use of IT to achieve business intelligence via data mining techniques enable businesses to more quickly and accurately analyze operations in areas such as customer relationship management, personnel management, and finance. In the following sections, we examine business intelligence, data mining and domain driven data mining and how a combination of these can be used to solve reallife business decision making problems. We thereafter review case studies in which domain knowledge has significantly impacted on the results obtained from the application of data mining to real-life business problems. 1.1 Business Intelligence

Keywords: business, business intelligence, Domain Driven Data Mining, Data Mining.

1. Introduction In recent times, the application of information technology (IT) to yield better performance in the business domain is all pervasive. However, past research suggests that not all investments in IT made by businesses result in improved performance. Rather, a high conversion effectiveness of the IT investments to measurable business objectives is necessary before a positive impact can be achieved [1]. Businesses that can efficiently transform data into useful information can use them to make quicker and more effective decisions and thus form better actionable business strategies which will give them a competitive edge. The development of actionable business strategies is however not an easy task due to domain knowledge constraints and expectations of key decision makers of the business domain [2]. When arrived at, the correct business strategy addresses such issues as personnel selection, accurately identifying target markets, consumer preferences and effectively managing the process by which goods and services are produced and delivered to

One of the primary factors that influence the performance of businesses is the ability to make effective and timely decisions in a consistent manner. Businesses now have access to large amounts of business related data but are unable to utilize them. Business intelligence aims to bridge this gap by providing businesses with tools and methodologies with which to harness the potential in already available business related data, thereby facilitating more effective and timely decisions [4]. The ability of a business to correctly transform and utilize information and knowledge in a timely manner is called “business intelligence”. Business intelligence methodologies are varied and complex and have a wide area of application. The major advantage deriving from the adoption of a business intelligence system is found in the increased effectiveness of the decision-making process. The complexity of business intelligence real-world problems can be categorized into:

• • •

Human roles and intelligence, Domain knowledge and intelligence, Network and web intelligence,

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org • • •

Organizational and social intelligence, In-depth data intelligence, Metasynthesis of the above intelligences.

42

can also navigate through business related data to answer a multitude of business questions in a timely manner. In the next section, the concept of domain driven data mining is briefly introduced.

As it is mentioned in [5] there is inadequate literature in these regard which leaves lots of room for further research in above categories.

1.3 Domain Driven Data Mining

In the business environment, the main objectives of business intelligence are to make effective and timely decisions, and to reduce uncertainty. These objectives are realized through the application of data mining as a tool of business intelligence by providing the means to transform data into useful and actionable knowledge [6].

An imbalance exists between the number of data mining algorithms published and the very few that are actually useful in a business setting. Hence conventional data mining has failed to deliver adequate results in decision making for real life business. A need therefore arises for a better framework within which better results can be obtained from existing data mining methodologies, techniques, tools and applications [11].

1.2 Data Mining Data mining or knowledge discovery has emerged as one of the most active areas in information and communication technologies (ICT). Data mining is an iterative process involving a combination of techniques from several disciplines. When applied to large data sets, data mining yields interesting knowledge, patterns, or high-level information which can be viewed from different angles. The discovered knowledge can be applied to decision making, process control and information management. This has pushed data mining into the forefront of recent developments in ICT [7]. The versatility of data mining motivates its research and development in academia and its applications in the business community [8]. To further increase this versatility, latest developments in data mining presented as publications in recent journals and conferences should be integrated into business applications in order to get better results as has been set as goals of most recent conferences such as SIGKDD [9]. Our goal would be finding out such developments and techniques to improve the efficiency of business intelligence. The developments and applications of actionable knowledge discovery (AKD), a new paradigm shift in data mining, to real-world businesses and applications are based on Domain Driven Data Mining. Studies and research in this regard will make a huge difference in Business intelligence [10]. The final goal is to have data mining well integrated into the decision-making process for real life businesses by generating more accurate, timely and relevant information. With a more timely and streamlined flow of more accurate, business related information, decision makers across the pyramid structure of businesses have a better idea of what is happening in the world in which they operate. Not only do they more quickly receive reports that are more understandable, but

This led to the emergence of domain driven data mining which primarily aims to deliver better decision making solutions for businesses by presenting tools for actionable knowledge that can be passed on to business people for direct decision-making and action-taking. Domain driven data mining aims to introduce a new paradigm shift; from data-centered hidden pattern mining to domain driven actionable knowledge discovery and delivery [12].

2. Review Data mining techniques are effective at generating useful statistics and finding patterns in large volumes of data, but as Pohle [13] mentioned, not as effective at interpreting these results, which is crucial for turning them into interesting, understandable and actionable knowledge. A so called knowledge acquisition bottleneck is caused by experts who have gathered a lot of experience over long periods of time in a particular domain but are unable to use this to effectively solve problems in a timely manner. A major benefit of using data mining techniques is that it bypasses the knowledge acquisition bottleneck [14]. The application of data mining techniques alone is not sufficient in solving real-world business problems. Dybowski et al. (2003) [15] investigated how data mining techniques and domain knowledge can be combined to construct more useful, efficient and effective decision support systems. Fayyad et al. [16] also suggested that the use of domain knowledge is important in all stages of the knowledge discovery process. Domain driven data mining goes beyond the conventional data mining methods. It involves the application of relevant intelligence surrounding the business i.e., human

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org intelligence, domain intelligence, network intelligence and organizational/social intelligence, and the combination of such relevant intelligence into a complete humancomputer-cooperated problem-solving system [17]. The significance of human intelligence in data mining was investigated by S. Sharma and K. Osei-Bryson [18]. The researchers indentified twelve data mining processes which require human intelligence. They posited that data mining requires human intelligence in order to generate valid and meaningful results. As a direct application of this, a recent research carried out by Atish et al. (2008) [19] established that there is interaction between the classification method of data mining and domain knowledge. They concluded that the incorporation of domain knowledge has a higher influence on performance for some data mining methods than for others. Chien et al. (2006) [20] collaborated with domain experts to develop specific recruitment and human resource management strategies using data mining techniques. Their results were successfully applied in a real-world business. Zhao et al. (2009) examined the effects of feature construction [21] guided by domain knowledge, on classification performance. The results of their study showed that feature construction, guided by domain knowledge, significantly improves classifier performance. Based on the surveys presented in [17] many more studies exist showing the importance of domain knowledge in data mining. This solidifies the argument for the inclusion of domain knowledge in data mining techniques to increase their relevance, efficiency and efficacy in realworld business decision making.

3. Case Studies In this section, three case studies are reviewed. These case studies show how domain driven data mining can be applied to the business domain, more specifically in the sectors of risk management in insurance, churn prediction and personnel selection. 3.1 Risk Management in Insurance Data mining aims to derive valuable business knowledge from patterns in database. In the majority of cases there is theoretical and domain dependent knowledge available. This study, carried out by Daniels and Dissel (2002) [22], investigates risk management in the Insurance business. In

43

this case study, it is observed that the effectiveness of data mining systems can be substantially improved as compared to data mining systems based on blind search only, by including knowledge about the model to be constructed and knowledge of experienced domain experts. This approach has two advantages. First, the otherwise blind search in databases is now guided by expert experience leading to substantially more accurate results. Second, since in general experts find it difficult to combine decision rules into a single risk score, the framework discussed offers the possibility to combine and fine-tuned expert knowledge using real cases. Figure 1 below shows the flow of claims and how domain experts is added as an extra check after the data mining system has filtered out suspicious claims.

Incoming Claims

Fraud Detection Module

Suspicious Claims

Fraud Expert

Others

Routine Handling

Figure 1: Flow of claims

3.2 Churn Prediction Several data mining models and algorithms exist that carry out churn prediction [23]. In this case study [24], the researchers showed how to make data mining models developed for churn prediction more understandable and compliant by combining it with relevant domain knowledge. More specifically, they showed how the analysis of coefficient signs in logistic regression and the monotonicity analysis of DTs can be used to check whether the knowledge contained in data mining models is in accordance with domain knowledge, and how to correct any discrepancies found. The idea is to help companies discover which customers are more valuable and also to help them identify the main elements in their data that can contribute positively or negatively to the relationship with the customer, and through that, define strategies that would benefit both company and customer alike. 3.3 Personnel Selection High-tech companies rely on human capital to maintain competitive advantages. This study developed a data mining framework to extract useful rules from the relationships between personnel profile data and their work behaviors. Furthermore, the researchers developed useful strategies with domain experts in the case company and most of the suggestions have been implemented. With

IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 4, No 2, July 2010 www.IJCSI.org an effective personnel selection process, organizations can find the suitable talents at the first attempt to improve retention rate and generate better performance [20].

4. Conclusion This study examined how data mining via domain driven data mining can be applied to businesses in order to yield more useful results. Three case studies were reviewed which show the effectiveness and efficacy of this method in the business domain. In each case study reviewed, domain knowledge was applied in addition to the data mining techniques and this yielded a significant improvement in the results obtained. Areas of future study could be the expansion of the scope to consider other fields such as agricultural, engineering and medical applications. Acknowledgments The supports of the Dr. Laszlo Kozma, Director of the International Studies Centre of Arts, Humanities and Sciences and Ms.Denissza Blanar head office of International Relations at University of Debrecen are strongly acknowledged. References [1] P. Weill, “The Relationship between Investment in Information Technology and Firm Performance", Information Systems Research, Vol 3, No 4, 1992, pp 307-333. [2] L. Cao and C. Zhang, Knowledge Actionability: Satisfying Technical and Business Interestingness, International Journal of Business Intelligence and Data Mining, Vol. 2 No. 4, 2007, pp. 496-514. [3] L. Cao, P. S. Yu, C. Zhang and Y. Zhao, Domain Driven Data Mining, New York: Springer Publishers, 2010. [4] Moss L., Atre S. (2003). Business Intelligence Roadmap: The Complete Project Lifecycle for Decision-Support Applications. [5] Carlo Vercellis, Business Intelligence: Data Mining and Optimization for Decision Making, Indianapolis: John Wiley Publishers, 2009. [6] S. Kudyba and R. Hoptroff, Data Mining and Business Intelligence: A Guide to Productivity, London: Idea Group Publishing, 2001. [7] J. Han and M. Kamber, Data Mining: Concepts and Techniques, 2nd edition, London: Morgan Kaufmann, 2006. [8] H. Varian, Intermediate Microeconomics Fourth Edition, New York: W. W. Norton & Company, 1996. [9] Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining 2009, Paris, France, June 28 - July 01, 2009. [10] L. Cao, and C. Zhang, The Evolution of KDD: Towards Domain-Driven Data Mining, International Journal of Pattern Recognition and Artificial Intelligence, Vol. 21, No. 4, 2007, pp. 677-692.

44

[11] Ankerst M. Report on the SIGKDD-2002 Panel the Perfect Pata Mining Tool: Interactive or Automated? ACM SIGKDD Explorations Newsletter, 4(2):110111, 2002. [12] Cao L and et al. Domain-driven data mining: a practical methodology, Int. J. of Data Warehousing and Mining, 2(4): 4965, 2006. [13] C. Pohle, Integrating and updating domain knowledge with data mining, In: M.H. Scholl, T. Grust (Eds.), Proceedings of the VLDB 2003 PhD Workshop (electronic edn.), Berlin, Germany, 2003 [14] R.R. Hoffman, The problem of extracting the knowledge of experts from the perspective of experimental psychology, AI Magazine, Vol. 8, 1987, pp. 53–67. [15] R. Dybowski, K.B. Laskey, J.W. Myers and S. Parsons, Introduction to the special issue on the fusion of domain knowledge with data for decision support, Journal of Machine Learning Research Vol 4, 2003, pp. 293–294. [16] U. Fayyad, G. Piatetsky-Shapiro and P. Smyth, From data mining to knowledge discovery in databases, AI Magazine Vol. 17 No. 3, 1996, pp. 37–54. [17] L. Cao, P. S. Yu, C. Zhang and Y. Zhao (eds), Data Mining for Business Applications, New York: Springer Publishers, 2009. [18] S. Sharma and K. Osei-Bryson, Role of Human Intelligence in Domain Driven Data Mining In: Data Mining for Business Applications, New York: Springer Science+Business Media, 2009. [19] A. P. Sinha and H. Zhao, Incorporating domain knowledge into data mining classifiers: An application in indirect lending, Decision Support Systems vol.46, 2008, pp.287–299. [20] C. Chien and L. Chen, Data mining to improve personnel selection and enhance human capital: A case study in hightechnology industry, Expert Systems with Applications, Vol. 34, 2008, pp. 280–290. [21] H. Zhao, A. P. Sinha and W. Ge, Effects of feature construction on classification performance: An empirical study in bank failure prediction. Expert Systems with Applications Vol. 36, 2009, pp. 2633–2644 [22] H. Daniels and H. Dissel, Risk Management based on Expert Rules and Data Mining: A Case Study in Insurance. In: Proceedings of the 10th European Conference on Information Systems (ECIS), 2002, Gdansk. [23] H. Liu and H. Motoda, Less is more In Feature Extraction, Construction and Selection: a data mining perspective, Norwell: Kluwer Academic Publishers, 1998. [24] E. Lima, C. Mues and B. Baesens, Domain knowledge integration in data mining using decision tables: case studies in churn prediction, Journal of the Operational Research Society Vol. 60, 2009, pp. 1096–1106.

Adeyemi Adejuwon Adeyemi Adejuwon is doing his masters degree in computer science at University of Debrecen, Faculty of Informatics. He is interested in data mining studies and its applications to business. Amir Mosavi is a PhD candidate and teacher assistant at university of Debrecen Faculty of Informatics. He is working in engineering MCDM and MOO as well as data mining. He has more than ten published journals and conference proceedings so far in these fields.

IJCSI CALL FOR PAPERS JANUARY 2011 ISSUE Volume 8, Issue 1 The topics suggested by this issue can be discussed in term of concepts, surveys, state of the art, research, standards, implementations, running experiments, applications, and industrial case studies. Authors are invited to submit complete unpublished papers, which are not under review in any other conference or journal in the following, but not limited to, topic areas. See authors guide for manuscript preparation and submission guidelines. Accepted papers will be published online and indexed by Google Scholar, Cornell’s University Library, DBLP, ScientificCommons, CiteSeerX, Bielefeld Academic Search Engine (BASE), SCIRUS, EBSCO, ProQuest and more. Deadline: 05th December 2010 Notification: 10th January 2011 Revision: 20th January 2011 Online Publication: 31st January 2011                

Evolutionary computation Industrial systems Evolutionary computation Autonomic and autonomous systems Bio-technologies Knowledge data systems Mobile and distance education Intelligent techniques, logics, and systems Knowledge processing Information technologies Internet and web technologies Digital information processing Cognitive science and knowledge agent-based systems Mobility and multimedia systems Systems performance Networking and telecommunications

              

Software development and deployment Knowledge virtualization Systems and networks on the chip Context-aware systems Networking technologies Security in network, systems, and applications Knowledge for global defense Information Systems [IS] IPv6 Today - Technology and deployment Modeling Optimization Complexity Natural Language Processing Speech Synthesis Data Mining

For more topics, please see http://www.ijcsi.org/call-for-papers.php All submitted papers will be judged based on their quality by the technical committee and reviewers. Papers that describe on-going research and experimentation are encouraged. All paper submissions will be handled electronically and detailed instructions on submission procedure are available on IJCSI website (www.IJCSI.org). For more information, please visit the journal website (www.IJCSI.org)

© IJCSI PUBLICATION 2010 www.IJCSI.org

IJCSI

The International Journal of Computer Science Issues (IJCSI) is a well‐established and notable venue for publishing high quality research papers as recognized by various universities and international professional bodies. IJCSI is a refereed open access international journal for publishing scientific papers in all areas of computer science research. The purpose of establishing IJCSI is to provide assistance in the development of science, fast operative publication and storage of materials and results of scientific researches and representation of the scientific conception of the society. It also provides a venue for researchers, students and professionals to submit ongoing research and developments in these areas. Authors are encouraged to contribute to the journal by submitting articles that illustrate new research results, projects, surveying works and industrial experiences that describe significant advances in field of computer science. Indexing of IJCSI 1. Google Scholar 2. Bielefeld Academic Search Engine (BASE) 3. CiteSeerX 4. SCIRUS 5. Docstoc 6. Scribd 7. Cornell's University Library 8. SciRate 9. ScientificCommons 10. DBLP 11. EBSCO 12. ProQuest

© IJCSI PUBLICATION www.IJCSI.org