Modified LDA classifier in multi resolution wavelet ...

Arab Republic of Egypt Ministry of Defense Military Technical College

4th International Scientific Conference of The Military Technical College 27-29 May 2008

‫اﻟﻤﺆﺗﻤﺮ اﻟﺪوﻟﻰ اﻟﻌﻠﻤﻰ اﻟﺮاﺑﻊ ﻟﻠﻜﻠﻴﺔ اﻟﻔﻨﻴﺔ اﻟﻌﺴﻜﺮﻳﺔ‬ ٢٠٠٨ ‫ ﻣﺎﻳﻮ‬٢٩-٢٧

Proceedings of The 6th International Conference on Electrical Engineering ICEENG-6

‫ﻣﺠــــــﻠﺪ‬

‫اﻟﻤﺆﺗﻤﺮ اﻟﺪوﻟﻲ اﻟﺴﺎدس ﻟﻠﻬﻨﺪﺳﺔ اﻟﻜﻬﺮﺑﻴﺔ‬ Cairo 2008

PREFACE The Military Technical Collage is pleased to organize the 6th international conference on electrical engineering, which is sponsored by the ministry of defense in the period 27 - 29 May 2008. The main objective of this conference is to bring together scientists, researchers and engineers of Egyptian Armed Forces and their colleagues in academic and industrial institutions. This occasion provides a chance to exchange information in the following designated fields of interest: 1.

Communication Systems

2.

Computer Engineering

3.

Biomedical Engineering

4.

Electrical Power Engineering

5.

Circuits, Signals and Systems

6.

Electromagnetic Fields and Waves

7.

Electronic Measurements and Instrumentations

8.

Remote Sensing and Avionics

9.

Opto-Electronics

10.

Guidance and Control

11.

Radar Systems

From (274) abstract submitted to the conference, a number of (259) abstract were accepted and the rest of abstracts were rejected. The (259) full manuscript submitted to the conference secretary, were reviewed. Just (189) papers have been accepted for suitable publication in the conference and (70) were rejected. These selected papers will be presented during the conference interval, from 27 to 29 May 2008 in different (37) scientific sessions. In addition to scientific sessions, the conference activities include different (11) workshop, invited lecture, invited paper, planned talk and seminar. The list of workshops, invited lectures invited papers, planned talks and seminars are: 1.

Simulations of Electric Machine Drives

2.

Phasor Measurement Unit Applications

3.

Energy Efficiency Improvement i

4.

Load Balancing in Multi-Phase Circuits

5.

Bioelectromagnetic Effects

6.

Statistical Learning Machines

7.

Renewable Energy and its Impacts

8.

Information and Comm. Technology

9.

Image Sensors Design and Technology

10.

Lightning Electromagnetics

11.

Supercapacitors

An alphabetic index of the authors of the papers is provided at the end of this booklet, which will serve to identify the session in which the paper will be presented. Finally the conference high committee hopes that the conference will achieve its planned mission and would like to acknowledge all contributors, members of the scientific committee and chairmen of the conference session.

Hint for authors: Please remember that you have only 10 min. to present the paper and 5 min. for discussion.

ii

Contributors Egyptian Universities, Institutes and Authorities: -

Ain Shams University Al-Azhar University Alexandria University Arab Academy for Science, Technology & Maritime Transport, Alexandria Arab Academy for Science, Technology & Maritime Transport, Cairo Assiut University Atomic Energy Authority Benha University Cairo University Egyptian Armed Forces Egyptian Electricity Transmission Company Electronics Research Institute Elminia University Energy Efficiency Improvement and Greenhouse Gas Reduction Project Federal Electric Egypt French University in Egypt German University in Cairo Helwan University Higher Technological Institute Higher Technological Institute (HTI), 10th of Ramadan City Menoufia University Mentor Graphics Military Technical College Ministry of Electricity and Energy Misr International University Misr Petroleum Company Modern Academy for Engineering and Technology Modern Academy in Maadi National Authority for Remote Sensing and Space Science Nilesat - Egyptian company Silicon Valley Academy South Valley University Suez Canal University Tanta University The Arab Contractor Osman Ahmed Osman & Co. The Higher Institute of Engineering, C.S.C., 6th of October City Zagazig University

iii

Arab and Foreign Universities and Institutes: -

Alegria, UER Electrotechnique, EMP, Algiers Algeria, Higher School of Technical Teaching and Engineering, Oran Algeria, Microwaves Laboratory, Polytechnic Military School, Bordj El Bahri Algerie, LASA, University Badji Mokhtar, Annaba Bangladesh, Bangladesh University of Engineering and Technology (BUET), Dhaka Belgium, European commission, DG Information Society Brazil, Federal Technical University of Parana, Ponta Grossa Brazil, Federal University of Santa Catarina, Florianopolis Canada, BTI Photonic Systems Inc., Ottawa, Ontario Canada, University College of the North, Thompson, MB Canada, University of Windsor, Windsor, Ontario China, Changchun University of Science and Technology, Changchun China, National Taiwan University, Taipei, Taiwan Crete, Technical University of Crete Crete, Technological Educational Institute of Crete Czech Republic, University of Defence, Faculty of Military Technologies, Kounicova, Brno Estonia, Tallinn University of Technology, Tallinn France, Institut Supérieur de l’Aéronautique et de l’Espace France, Vitry High Technology Institute, Paris 12 University India, LDC Institute of Technical Studies, Allahabad, India India, M. N. N. I. T., Allahabad India, N. I. T. Hamirpur (H. P.) Indonesia, Engineering Faculty, Mataram University Iran, Amirkabir University of Technology, Tehran Iran, Iran University of Science & Technology, Tehran Iran, Isfahan University of Technology, Isfahan Iran, Islamic Azad University, Abhar Branch, Abhar Iran, K. N. Toosi University, Tehran Iran, Sabzevar Azad University, Sabzevar Iran, Semnan University, Semnan Iran, Tehran University, Tehran Iran, University of Zanjan, Zanjan Iraq, North Iraqi Petroleum Company Iraq, University of Mosul, Mosul Italy, Etnoteam Spa, Milan Japan, Graduate School of Science and Technology, Kumamoto University Kumamoto-Shi Jordan, Mutah University, Jordan Jordan, Philadelphia University iv

-

Jordan, University of Jordan, Amman Korea, School of Electrical Engineering and INMC, Seoul National University Korea, Sogang University, Seoul Latvia, Riga Technical University, Riga Libyan Armed Forces Malaysia, International Islamic University Malaysia (IIUM), Kuala Lumpur Malaysia, Multimedia University, Cyberjaya Malaysia, Sarawak Electricity Supply Corporation (SESCO), Sarawak Malaysia, Syarikat SESCO Berhad, Sarawak Malaysia, Telekom Research and Development Malaysia, Universiti Kebangsaan Malaysia, Selangor Darul Ehsan Malaysia, Universiti Malaysia Sarawak (UNIMAS), Kota Samarahan, Sarawak Malaysia, Universiti Sains Malaysia, Pulau Pinang Malaysia, Universiti Tenaga Nasional Malaysia, University Malaysia Perlis (UNIMAP) Malaysia, University of Science Malaysia (USM) Malaysia, University Putra Malaysia (UPM) Malaysia, University Technology Petronas, Bandar Seri Iskandar, Tronoh, Perak Pakistan, FAST, National University of Computer and Emerging Sciences, Karachi Pakistan, Government College of Science, Wahdat Road, Lahore Pakistan, UET, Lahore Pakistan, University of the Punjab, Lahore Philippines, Mapua Institute of Technology, Muralla, Intramuros, Manila Philippines, Mariano State University Qatar, Texas A & M University at Qatar, Doha Romania, “Gh. Asachi” Technical University of Iasi, Iasi Romania, Transelectrica Company, Bucharest Russia, Institute of Electrical Engineering, Tomsk Polytechnic University, Tomsk Russia, St. Petersburg Electrotechnical University Saudi Arabia, Arab Open University Saudi Arabia, King Abdualaziz City for Science and Technology, Riyadh Saudi Arabia, King Abdulaziz University, Jeddah Saudi Arabia, King Saud University, Riyadh Saudi Arabia, Riyadh Group of Cables, Riyadh Saudi Arabia, Saudi Electricity Company, Riyadh South Africa, University of Pretoria, Pretoria Sudanese Armed Forces Switzerland, ALaRI, Universita Della Svizzera Italiana, Lugano Switzerland, EMC SA, Bellinzona v

-

Syrian Armed Forces Tajikistan, Physical Technical Institute of Academy of Sciences, Dushanbe Turkey, Dokuz Eylül University, zmir Turkey, Erciyes University, Kayseri Turkey, stanbul Technical University, Maslak, stanbul Turkey, K kkale University, K kkale Turkey, Selcuk University, çel, Konya, Turkey UK, Loughborough University, Loughborough UK, University of Nottingham, Nottingham Ukraine, V. N. Karazin Kharkov National University, Kharkov USA, Florida International University, Miami, FL USA, St. Cloud State University, South St.Cloud, MN USA, State university of NY at Buffalo, Buffalo, NY USA, University at Buffalo, Amherst, NY USA, University of Florida, Gainesville, Florida USA, Virginia Polytechnic Institute and State University, VA

vi

Conference High Committee Maj. Gen. A. El-bardaoini

Chairman

Maj. Gen. I. Abd-El-Ghafar

Vice-Chairman

Maj. Gen. (R) M. Salam

Gen. Secretary

Brig. Gen. M. A. H. Eleiwa

Reporter

Col. Waheed Sabry

Reporter

Conference Steering Committee Brig. Gen. M. A. Eleiwa Col. Waheed Sabry

vii

Special Acknowledgment The conference steering committee would like to express deepest gratitude to our Professor, Dr. Fahmy Metwally Bendary, Faculty of Engineering (Shoubra), Benha University, for his continuous support, unlimited help and invaluable suggestions throughout the preparation work of this conference.

viii

Reviewers Professors and Ass. Professors from M.T.C., national Universities and Institutes, and International Institutes and Universities: A. A. Hegazy A. Feliachi A. Hamdy Abd-El-Hamid A. M. Asaad A. Reda Dawood A. Safwat El-Kabany A. Uddin A. Y. Abd-El-Aziz A. Y. Amin Darwish A. Y. El-Rafaay Abdel Razzak Nosseir Abd-El-Aziz A. Mitkees Abd-El-Fattah A. Hegazy Abd-El-Hadi Ammar Abd-El-Hamid A. Gaafar Abd-El-Lattif El-Zein Abd-El-Mhaimen Soliman Abd-El-Monaaem Mousa Abd-El-Sameaa B. Kotb Achim Autenrieth Adel Mohamadeen Ahdab El-Morshedt Ahmed Bahgat Ahmed E. Abd-Allah Ahmed E. El-Mahdi Ahmed El-Kashlan Ahmed El-Othmany Ahmed Geniedy Akram I. El-Metwally Alaa Assissi Alaa Fahmy Ali Kamel Al-Kharashi Ali M. F. El-Mashad Ali Oshieba Ali Rashed Amal F. Abd-El-Gawad Andreas Kirstädter Ashraf M. Abd-El-Aziz

Attef Ghoniem Ben Bella E. Tawfiq Bing Lu C. D. Dimitrakopoulos D. B. Mitzi E. Rocon Effat A. M. Mousa El-Said Othman Abd-El-Aziz El-Sayed M.Azzoz Esmat Abd-El-Fattah Ezz-El-Din Zakzouk F. Jurado Fahim A. Khalifa Fahmy Bendary Farouk El-Kady Farouk Ismail Fatehy Mahrous Fatehy Nasr Fathy Abd-El-Kader Fayek Farid Farag Allah Gamal El-Sheikh Gamal Mabrouk George D. Vendelin Gwenaëlle Toulminet H. El-Din Mostafa Atteia H. El-Motaafy H. Kombor H. M. Mashaly H. Mahmoud Hamed Galal Hassan H.EL-Tamaly I. Abd-El-Moneam I. El-Henawy I. El-Shaaer I. Helal I. Yaseen Ibrahim M. Salem Ibrahim Megahed ix

Ibtsam M. Said J. R. Saenz Jiang Hu K. M. Shebl Kamel yaseen Kh. A. Shehata Khairy El-Barbary Luigi Alberti M. A. Eliewa M. A. Izzularab M. A. Nasr Askoura M. Abd-El-Aal M. Abd-El-Hakkam M. Abou-El-Saad M. C. Newton M. El-Arini M. El-Hadidy M. El-Said Nasr M. El-Sayed Gad-Allah M. El-Sengaby M. H. Abd-El-Azeem M. I. Yousef M. M. Awad M. M. El-Gazzar M. M. El-Metwally M. M. Mansour M. Mamdouh Abd-El-Aziz M. Medhat Mokhtar M. Zaky Abd-El-Mageed M. Zayan Magdy El-Kafafy Mazen Abdel salam Mohamed A.Tantawy Mohamed Abd-El-Latif Badr Mohamed Abd-El-Rehim Bad Mohamed El-Sherbiny Mohamed Mahrous Mohamed Z. El-Sadek Mohie Mandour Mohsen Tawfiq Mustafa Muheilan

N. Bianchi Nabil A. El-Deib Nabil Gergis Nagy Sorial Nahed Hegy Naser Abd-El-Raheem Norm Jouppi Omar Abd-El-Haleem Osama Mahgoub P. A.Warburton P. Pampaloni R. A. Refaaie R.E. Miles S. Bolognani S. Cetinkunt S. Firth S. Hosny El-Banna S. M. El-Debikey S. Mabrouk Sharaf S. Narasimha S.-K. Sul Said A. Gawish Salah G. Ramadan Samah Sonbol Sayed A. Hassan Sherif M. Wasfy Sohier Sakr Soliman El-debeky Steven Tanimoto Tarek Sharaf Thomas J. Kleespies Wafaae Boghdadt Wagdy M. Mansour Waheed Sabry Weixiang Shen Y. Mandour Yasser G. Abd-El-Razek Yasser Galal Yousri El-Nahas Zaher Abd-El-Moaatee Zhe Chuan Feng

x

Authors A. A. El-Hennawy A. A. Mousa A. A. Al-Arainy A. A. El-Mahallay A. A. Heggo A. A. Ishak A. A. Madkour A. A. Mahfouz A. A. Mitkees A. A. Mohammed A. A. Mousa A. Abd-El-Nazir Ahmed A. Abou Elazm A. Afshar A. Alijamat A. AlQawasmi A. E. Emam A. El Shahat A. El-Din Sayed Hafez A. El-Mahdy A. F. Abdel-Gawad A. H. Asseesy A. Hafez Zaki A. Hosny A. K. Othman A. Kidher Mahmood A. M. Abdeen A. M. Allam A. M. Deghedy A. M. El-Hadidy A. M. Fouda A. M. Rashed A. M. Salama A. Mahran A. Mohammad A. N. Abdel-Latief A. Ple ca A. Tolga Bozdana A. Y. Abdelaziz Abass, Y. M.

Abbas Nasrabadi Abd Elkawy M. S. Abd-Elatief M. El-Zein Abd-El-Azez Mitkes Abdel-Aziz T. Shalabi Abdelelah K. M. Abd-El-Maged Allam Abid Yahya Abolfazl Halvaei Niasar Abolfazl Vahedi Adel Ali Abou El Ela Adel.Rizk Adrian Baraboi Ahmad Cheldavi Ahmad. K. Atieh Ahmed Abd El-Wahab Ahmed Abo-Elhadeed Ahmed E. Abdalla Ahmed ElDeib Ahmed Ibrahim Salem Ahmed M. Bakhraiba Ahmed M. Elaiw Ahmed M. Othman Ahmed M. Youssef Ajnadeen Khalil Alaa A. Elrahim Alaa Eldin Rohiem Alaa Eldin. S. Abdelaziz Alaa Rohaiem Aladin Kamel Al-Emam S. Ragab Al-Homosy, G. M. Ali A. Orouji Ali Doustmohammadi Ali El Moghazy Ali Ibraheem Alie El-Din Mady Al-Kholy, S. A. Alvin K. L. Li Aly A. Fahmy

237

Alyani Ismail Amen Nassar Amged S. El-Wakeel Amin Mohamed Nassar Amr M. Ahmad Andrea Tonini Andrew J. Arana Ashraf M. Aziz Ashraf Diaa Elbayoumy Ashraf Mohamed Ali Ashraf Uddin Asmaa Abdel Tawab Atalla I. Hashad Ayman El Gezawy Ayman G. Sobih Ayman H. M. Kassem Aysen Demiroren B. Atrouz Bassem Ibrahim Sheta Bilal Hassanein Borhanuddin Mohd Ali C. J. Chai Carlos Rodríguez Casal Catalin Pancu Cheong-Hwan Kim D. F. El-Hossary D. V. Mayboroda D. Vinnikov Dae-Seung Ban Davide Finardi Derya Ahmet Kocaba Doaa M. Atia E. A. El-Diwany E. Elshimy Elfatih J. Hamdi El-Said El-Bagoury Elsayed Khadrogy Eraky Atta Ercan Yaldiz Essam A. Eldiwani Ezz Eldin Farouk Ezzat A. Mansour

F. A. Khalifa F. Askari F. E. Abd Al Kader F. El-Bendary F. Mohd-Yasin F. Y. Chua Fadhila Mohammad Faiz Yousif Mohmmed Fares Hassan Taha Farid Ghani Faten H. Fahmy Florence Choong G. A. Al-Sheikh G. Kahela G. M. Abdel-Hamid Gamal A. Elshiekh Gamal M. El Bayoumi Gamal Sarhan Gerard Bourdon Ghada Farouk Gouda Ismail Salama H. A. Attia H. A. Hassan H. B. Hussain H. Farouk H. H. El-Banna H. H. El-Tamaly H. K. Temraz H. M. El Maghraby H. M. El Shewy H. M. Hassan H. M. Salem H. S. Thye H. Y. Abd-halim Hadia El-Hennawy Haitham El-Sayed Akah Hakan Saracoglu Haldun Sarnel Hamada Attia Hamid Lesani Hani Ragaie Hany A. Mansour

238

Hany H. Ahmad Hasan O. Farahneh Hassan H. EL-Tamaly Hassan H. Monfared Hassan Moghbeli Hazem H. Ali Helio Voltolini Hocine Kimouche Hossam A. Elshiekh Hossam M. Hendy Hulusi Karaca I. B. Said I. El Shair I. G. Pasek Suta Wijaya I. Galkin I. Ismail I. M. El-Desoky I. Murtaza I. Qazi Ibrahim M. Mansour Ibrahim N. Abu-Isbeih Ibrahim Yassin Mahmoud Imad Merzouk Imbaby I. Mahmoud Irshad Khokhar Isa Yossef Ismail A. Ghaffar Ismail Razi J. Laugis Jan Leuchter Javad Haddadnia Jinyong Lee Juhan Laugis K. Ab. Hamid K. Abbaszadeh K. Abdel-Aty K. H. Gharib K. H. Moustafa K. P. Leslie Chai K. W. Chew Kafetzis G. Kamel Aliouane

Keiichi Uchimura Kemal Polat Kh. I. Saleh Khairy A. Elbarbary Khaled Ali Shehata Khaled Daqrouq Khaled F. Elsayed Khalid A. S. Al-Khateeb Khasan S. Karimov Kheira Abdelmoudjib Kolokotsa D. Ku Maziana KuMamat L. C. Kho Leslie K. P. Chai Levent Seyfi M. A. Ali M. A. Awadallah M. A. Badr M. A. El-Hadidy M. A. Ghazy M. A. Mostafa M. A. Shedid M. A. Zayan M. Abd Elghany M. Abdel-Gawad M. Abdel-Salam M. Abo Rizka M. Abou El-Ata M. AbouelSaad M. Ahmadi M. Ali Soliman M. Allahiari M. Ardebili M. Asghar Saqib M. B. I. Reaz M. Bitchikh M. E. Kaliberda M. E. Mandour M. Egorov M. El Bahy M. H. Abd El-Azeem M. H. Abdel-Aall

239

M. H. Abd-El-Wahab M. H. El-Mahlawy M. H. Saleh M. I. El-Singaby M. I. Ibrahimy M. I. Qureshi M. I. Youssef M. K. Amirhosseini M. Karem M. Khalaj-Amirhosseini M. M. Abu-Elnaga M. M. El Kholy M. M. Korjani M. M. Mansour M. M. Samy M. Mahroof-Tahir M. Medhat Mokhtar M. N. Saati M. Nuri Seyman M. S. Alshol M. S. Barakat M. S. El Samahy M. S. Islam M. Safiuddin M. Safy M. Saleem M. Sharifzadeh M. Shokair M. Soliman M. Y. El Nahas M. Y. Kamel M. Zakaria M. Zaki Magdy A. Elbar Mahamod Ismail Mahdi M. El-Arini Mahmoud A. El-Sawi Mahmoud Magdy Mahmoud S. Hamed Mahmud E. Gadallah Majdi Salem Mardina Abdullah

Maricel Adam Mas R. bin Abd Rahim Masoud Aliakbargolkar Mat Kamil Awang May Salama Maya M. Emarah Mazen Abdel-Salam Mehdi Foroozanfar Mehmet Bayrak Metin Çiçek Michael C. Pacis Moataz M. Salah Moawad I. M. Dessouky Mohamed A H Eleiwa Mohamed Abdelhady Mohamed E. Nasr Mohamed Elbar Mohamed K. Saad Mohamed Ramadan Mohamed A. ElShafie Mohammed A. Kotb Mohd. Fadzil Ain Mohi Ahmad Morteza Fathipour Morteza Ghazisaedy Muhammad Asif Muhammad H. Sayyad Muhammad Saleem Muhammad Yaseen Mukhtar Ali Mustafa Muheilan Mutabar Shah Mutamed Khatib N. Abdel-Gawad N. H. Malik N. Y. Abed Nasreddine Debbache Necmi Ta nar Nor Kamaria Noordin Norbahiah Misran Norsuzila Ya’acob Nursyarizal Mohd Nor

240

O. A. Al-Nather O. A. Mohammed O. M. Shahril O. S. Ebrahim O. S. Kachin uzhan Y lmaz Omar H. Abdalla Ömer Eyercio lu Osama A. Mahgoub Osama Fathy Hegazy Otakar Kurka Othman O. Khalifa Othman Sidek Patelis P. Perumal Nallagownden Pierre Magnan R. A. Elez R. Amirfattahi R. H. Akanda R. K. Singh R. M. Mostafa R. Strzelecki Rabah W. Aldhaheri Rafael G. Maramba Ragab A. El-Sehiemy Ramazan Akkaya Ramiah Jegatheesan Renato Carlson Rimon Elias Ronald B. Corpuz Roshdy A. AbdelRassoul S. A. Gawish S. A. Gholamian S. A. Pogarsky S. Ahmed S. Alirezaee S. Anam S. F. Mekhamer S. Gazy S. I. Kachin S. K. Kee S. K. Y. Nikravesh

S. Liu S. M. J. Razavi S. M. Tabatabaie S. M. Wasfy S. N. Tiwari S. P. Majumder S. Pispiris S. S. Ngu S. Sadri S. Shams El-dein S. Z. Abbas Saad Sharaf Sabira Khatun Sabri Altunkaya Salah El-Agooz Saleh M. Al-Saleem Salih Güne Salwa H. El Ramly Samaneh Sharbati Samer J. AL-Abed Samir Mahmoud Waly Samy S. A. Ghoniemy Sassan Azadi Seongsu Yang Sherif Welsen Shaker Soojin Kim Stavrakakis G. S. Suzan M. Shoukry Syed A. Kashif Syed Idris Syed Hassan T. Amin T. E. Taha T. El Arif T. Lehtla T. Nazmy T. S. Abdel-Salam Taj Mohammad Baloch Tamer Farouk Badran Tarek Abdolkader Tarek H. Elsayed Tarq Kalf Jassim Umair Soori

241

Ümit Y lmaz V. A. Rakov Valery Vodovozov W. L. Pang W. M. Mansour W. Sabry W. Swelam Wael Fikry Wael M. Abdullah Wael Mohamed Yousf Walaa Wahba Ibrahim Waleed A. Yousef Wedad M. Refaey X. Xia Y. G. Hegazy Y. Z. Elhalwagy Yahia M. El-Sayed Yahya Z. Mohasseb Yalç n I k Yamina Menasria Yavuz Senol Yi Qu Yong-Hwan Lee Younglok Kim Z. Liu Zaghlol S. El-Razaz Zahra Moravej Zakir Husain Zhe Chuan Feng Zhencheng Hu Zubair Ahmad

242

Scientific Sessions

6th International Conference on Electrical Engineering ICEENG 2008

Military Technical College Kobry El-Kobbah, Cairo, Egypt

Date: Wednesday, 28 May, 2008 Time: 11:00 – 12:45

Session: IP Room: I

Image Processing I Chairman: - Prof. M. Shaarawy Ibrahim Helwan University, Helwan, Egypt - Prof. Sassan Azadi Semnan University, Semnan, Iran - Prof. Haldun Sarnel Dokuz Eylül University, Izmir, Turkey EE022 Real time tracking in 3D space by image processing EE083 Accurate affine image registration using radial basis neural networks EE096 Modified LDA classifier in multi resolution wavelet domain for multipose face recognition EE160 Multi-agent model for face recognition using multi-features and multi-classifiers EE202 A new target-tracking approach in FLIR imagery using monocular thermal image sequence

Session Coordinator: Col. Gouda Ismail M. Salama

Khalid A. S. Al-Khateeb Mat Kamil Awang Othman O. Khalifa Haldun Sarnel Yavuz Senol I. G. Pasek Suta Wijaya Keiichi Uchimura Zhencheng Hu Aly A. Fahmy Gouda I. Salama Alaa A. Elrahim Magdy A. Elbar Gouda Ismail Salama Osama Fathy Hegazy Wael Mohamed Yousf

Military Technical College 16

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 1

6th International Conference on Electrical Engineering ICEENG 2008

Military Technical College Kobry El-Kobbah, Cairo, Egypt

Modified LDA classifier in multi resolution wavelet domain for multi-pose face recognition By I Gede Pasek Suta Wijaya*,**

Keiichi Uchimura*

Zhencheng Hu*

Abstract: This paper presents multipose faces recognition. The proposed scheme is based on holistic information of face image and small modification of classical LDA (modified LDA) classifier. The holistic information called as facial features is obtained by multiresolution wavelet analysis. The modified LDA (MLDA) classifier that works based on multivariate analysis classifies the facial features to a person’s class. The objectives of the proposed method are to create a compact and meaningful facial features without removing significant face image information, to build a simple classification technique which can well classify face images to a person’s class, to make the M-LDA-based training system to solve the retraining problem of the PCA and LDA based recognition system, to reduce the high memory space requirement of classical LDA and PCA, and to compare the effectiveness of proposed method to established LDA based recognition systems such as RLDA, DLDA, and SLDA. The result shows that the proposed method gives good enough performance i.e. high enough success rate, short time processing, and small enough EER compare to establish LDA. In addition, the wavelet transforms is an efficient way for reducing the dimensional size of original image. Keywords: facial features, matching, wavelet, LDA, eigenface

* Computer Science and Electrical Engineering Graduate School of Science and Technology, Kumamoto University Kumamoto-Shi, Japan ** Electrical Engineering Dept., Engineering Faculty, Mataram University Indonesia

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 2

1. Introduction: Human face image recognition is an active research area in image processing applications because there are many potential applications, which cover human computer interactions, forensics, surveillance, and security systems. Recently, the demand of biometrics system as security system has been increasing significantly to substitute password and PIN security system. The main weakness of the password and PIN system is little bit easy cracked them by guessing because 81% of users use the same password, PIN, and 30% write them down or store them in a file. Therefore, face is one of the biometrics systems that can be implemented as security system for substituting the password and PIN system. This paper proposes an alternative face recognition system, which is based on holistic or global information of face image and MLDA. The holistic information of face image called as facial features is obtained by multiresolution wavelet analysis of entire image without geometrical normalization and localization. The function of the MLDA is to classify the facial features to a person’s class. The main aims of this method are to solve large computational costs, high memory space requirements and retraining problems LDA based face recognition. 2. Previous Work: The previous works related to our approach are face recognition based on holistic or global approach as described in Refs. [1,2,3,4,5,9]. Ref. [4] describes face recognition based on wavelet packet tree analysis for frontal view of human faces under roughly constant illumination. The facial features were built by implementing wavelet- packet tree analysis of bounding box face and then calculating the mean and variance of sixteen matrixes wavelet coefficients. That approach does not work for non-frontal and small variation faces view and needs constant illumination to make the face-bounding box. Ref. [5] describes face recognition based on combination of DCT analysis and face localization technique for finding the global information of face image, but it requires eyes coordinate, which have to input manually, to perform geometrical normalization. The global face information was created by keeping small part of big magnitude values of DCT coefficients. The mostly related approach to our system is face recognition based on the LDA and its variations as described in Ref. [2,3,6]. Ref. [2] proposed a combination of D-LDA and F-LDA to cover the weakness of classical LDA. It only solves the poor discriminatory and singularity problem. Ref. [3] implemented DCT to reduce data dimensional and only small part of DCT coefficients is analyzed by LDA. Ref. [6] implemented the

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 3

wavelet transforms to reduce the dimension of face image, employ a regulation scheme for the within-scatter matrix, and use optimization procedure. It was reported that the Daubechies (Db-6) was implemented to filter image to resolution 29 x 23. However, those methods have limitations: large computational cost, high memory spaces requirement, and retraining problems. In our method, we implement multiresolution wavelets analysis for reducing the original data dimensional and MLDA classifier for classifying the face class without geometrical normalization and bounding box processing. It is difficult to compare our results with previous works because their time consuming were rarely reported and the tests were carried out with different databases. Therefore, our approach results will be compared to establish LDA, which has been tested with data from four face databases. 3. LDA based face recognition: 3.1 Classical LDA The main purpose of LDA analysis is to find a linear transformation such that feature clusters are most separable after the transformation. It can be achieved by the betweenclass scatter matrix Sb and the within-class scatter matrix Sw analysis, as explained in Ref. [3]. The class separation is measured by the ratio of determinant of the Sb matrix to the Sw matrix using the equation below. E = arg max E

E TSb E

(1)

E TSw E

Where E =[e1,e2,e3,…,em] is set of eigen-vectors corresponding to m largest eigenvalues λi which satisfy the equation: S b e i = λ i S W e i , i = 1, 2, 3, … , m. The eigenvectors and eigen-values are obtained by computing the inverse of Sb and then solving −1

the eigen problem of S W S b matrix. Finally, the projection of the linear discriminant function is given by:

(

Yi (C) = E T C i − m i

)

(2)

The intrinsic problem of above algorithm is the singularity problem of scatter matrix due to the high data dimensional and small number of training samples called as small size problem (SSS). Some methods have been proposed to solve that problem such as DLDA, RLDA, sand PCA+LDA as described in Refs. [2,7,8]. However, those methods still require large computational costs and the retraining problems.

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 4

3.2. M-LDA for Face Recognition In order to solve retraining problem, we propose the M-LDA approach. The M-LDA is based on assumption that the matrix scatter has small dimension and the covariance of the training images is multivariate normal distribution. The proposed algorithm is described as below.

[ ]

Let define a big matrix, Q = [X i ]i =1 , containing n classes with each class X i = x i , j n

m j=1

consisting m column vector of facial features, where m is number of member class Xi. Next, the mean of each class is easily determined and then placed it into mean matrix as Μ=[µ1,µ2,µ3,…,µc], where µ i = E (X i ) = (1 / m i )∑ mj=i1 x i , j . Finally, we can determine the global covariance using the with-in class (Sw) equation. S w = C g = ∑ ∑ (x j − µ i )(x j − µ i ) c

T

(3)

i = 1 x j∈ C i

If Cg is multivariate normal distribution, we can classify of each facial features to person’s classes using the equation below.

Fc = max[g 1 (x), g 2 (x), g 3 (x),..., g m (x) ]

(4)

Where gi(x) is given by:

g i ( x ) = µ i C −1 x T − 0.5µ i C −1µ i

T

(5)

The equation (5) is derived from maximum a posteriori (MAP) discriminant, as describe below:

g i ( x ) = P (ϖ i | x ) = g i (x) =

1 (2πn / 2 ) Ci

1/ 2

P( x | ϖ i ) P (ϖ i ) P( x )

exp{− 1/ 2(x − µi ) T C−i 1 (x − µi )}P(ϖi )

(6) 1 P(x)

(7)

where P(x) is total probability of x. By eliminating the constant term and taking the natural log, it becomes.

g i (x) = − 12 (x −

i

) T C −i 1(x −

i

) − 12 log ( C i ) + log (P(ϖ i ) )

(8)

As mentioned previously that all classes have identical covariance and the same prior probability, the equation (8) can be simplified as below:

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

g i ( x ) = − 12 xC −1 x T + µ i C −1 x T − 12 µ i C −1µ i

EE096 - 5

T

(9)

By keeping just the terms dependent on µi and C, the equation (5) is obtained. This algorithm has some advantages for classifying the face image class: 1. It is simple because it does not require the eigen-values and eigen-vectors analysis. 2. It can solve the retraining problem as illustrated: firstly, when a new class added to the system, the M-LDA calculates the mean and the covariance of its class; secondly, the newest mean is placed into the matrix M; and finally, the previous covariance is updated by adding it with the newest class covariance. 3. The computation complexity is less than the PCA and LDA computation complexity because of not requiring eigen analysis. The weakness of M-LDA is singularity problem due to the high data dimensional and small number of training samples. To overcome the singularity problem, we implement frequency analysis to reduce the data dimensional as explained in the next section. 4. Facial features extraction: In this research, we develop a holistic approach for facial features extraction based on multiresolution DWT analysis in the entire image without geometrical normalization and bounding process. The multiresolution wavelet analysis is performed by implementing repeatedly classical DWT called as filter bank decomposition, as shown in Fig. 1. The A, H, V, and D are calculated by equation (10).

[ H = [g * [ h * f ] V = [h * [ g * f ] D = [g * [ g * f ]

A = h * [ h * f ]x ↓ 2 x

↓2

x

↓2

x

↓2

] ] ] ]

y

↓2

y

↓2

y

↓2

y

↓2

(10)

lev. j I

Aj

lev.j+1

lev. n

...

Aj+1

…

An-1

An

Hn-1

Hn

Hj

Hj+1

Vj

Vj+1

Vn-1

Vn

Dj

Dj+1

Dn+1

Dn

Figure (1): filter-bank wavelet decomposition[12].

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 6

where, * denote convolution , ↓2 represent down sampling for x and y direction, g and h are high and low pass filter respectively. In order to make simple and fast decomposition process, we apply two different Daubachies wavelets basis, namely Db4 and Db1. First, Db4 basis decomposes face images until level 2 and it just return the approximation coefficients. Second, the Db1 basis decomposes the Db4’s approximation coefficient until maximum level. This decomposition returns the wavelet coefficients as shown in Fig. 2. From these coefficients, the compact and meaningful facial features called as holistic information are created by three steps: firstly, convert the frequency domain coefficients to vector using row ordering technique; secondly, sort the vector descending using quick sort algorithm, finally truncate a small number of vector elements (i.e., less then 100 elements). Those processes are performed on both training and query (probe) face images. However, in the training process, those are performed one time. Consider the compact facial features, if they are reconstructed into the face images, the reconstructed face images will be different. However, we can still understand that they are the face images, as shown in the Fig. 3(a). Meanwhile, if the compact facial features are removed the reconstructed face images are exactly different and we do not know that they are face images at all, as shown in the Fig. 3(b). This illustration proves that the most information of image exists in low frequency components. The compact facial features can be used as a basis of multipose face recognition because they consist of dominant of frequency components of the face image. It means that the facial features of any face pose variations in a single face are identical. It can be proved by calculating the correlation coefficients of both compact facial features and original image of any face pose variations in a single face, as shown in Fig. 4. It shows that the correlations of compact facial features are almost the same for all face poses. However, the correlations of the original have large different values for all face poses.

(a)

(b)

(c)

Figure (2): the output of multiresolution wavelet analysis: (a) original image, (b) the first step decomposition coefficients, (c) the second step decomposition coefficients[12].

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

(a)

EE096 - 7

(b)

Figure (3): (a) the reconstructed images of the compact facial features that the size ranges from 16 elements (top-left) until 225 elements (bottom-right), (b) the reconstructed images when the compact facial features are removed[12]. 1 0.95 Original face 0.9

Facial feature

0.85 0.8 0.75 Correlation coefficients 0.7

1

2

3

4

5 6 7 Face poses

8

9

10

11

Figure (4): the correlation coefficients comparison between the original image and the compact facial of eleven face poses of India face database[12]. 5. Face recognition algorithm: The proposed face recognition algorithm can be illustrated briefly as Fig. (5). There are three main process in this algorithm, namely preprocessing process, training process, and recognition process. The preprocessing unit consists of color space transformation, equalization, and multiresolution wavelet analysis. In this research, NTSC (YIQ) color space is implemented for to convert the color image (RGB) to gray component (Y). Actually, there are two main function of preprocessing process: firstly, to decrease the effect of the non-uniform lighting condition on image face, which is performed using standard equalization, secondly, to create compact facial features, using multiresolution wavelet transforms as described in section 4. In the training process, the compact facial features set is analyzed by MLDA for finding mean of each class and the global covariance and then save them in database as a meaningful data for face classification.

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

Target Faces Collection

Query Face

Pre-Processing

Pre-Processing

EE096 - 8

Multi-Res. Multi-Res. Wavelet Analysis Facial Feature Wavelet Analysis Extraction M-LDA-based Sub- M-LDA Classification Mean and Cov. Matrixes

Decision Rule

Training Process

Recognition Process Face Likeness

Figure (5): brief face recognition algorithm. Finally, we calculate the similarity between the input facial features and the training facial features set using MLDA based classifier. In this case, the minimum score is concluded as the best likeness. 6. Exsperiment and Result: The experiments were carried out using data from four face databases: ITS-Lab. Kumamoto University database, EE-UNRAM database, India database [10], and ORL database [11]. Each database has special characteristics. The ITS-Lab database consists of 48 people and each person has 10 pose orientations as shown in Fig. 6. The face images were taken by Konica Minolta camera series VIVID 900 under varying lighting condition. The India database consists 61 people (22 women and 39 men), each person has eleven pose orientations: looking front, looking left, looking right, looking up, looking up towards left, looking up towards right, and looking down. Indian database also included the emotions: neutral, smile, laughter, sad/disgust. The EE-UNRAM database consists of 40 people and each person has 8 pose orientations: looking front, looking left about 300, looking right about 300, looking up, looking down, and wearing accessory such as glasses. The ORL database was taken at different times, under varying the lighting conditions, facial with different expressions (open/closed eyes, smiling/not smiling) and facial details (glasses/no glasses). All of the images were

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

Q1

Q2

Q3

Q4

Q5

Q6

Q7

Q8 (a)

Q9

Q10

EE096 - 9

Figure (6): Example of face poses of ITS-Lab. face database[12]. taken against a dark homogeneous background. The faces of the subjects are in an upright, frontal position (with tolerance for some side movement). The ORL database is a grayscale face database that consists of 40 people, mainly male. In order to know the performance of the proposed method, some experiments were done using data from previously mentioned databases. All of experiments were performed in image size of 128 x 128 pixel, the facial vector size of 49 elements, and 5 training face per class except the UNRAM database was 4 training face per class. They were chosen based on the information in Ref. [9]. The first experiment investigated the accuracy of the proposed method in tested databases and compared with the establish LDA such as DLDA, RLDA, and SLDA. The experimental result can be shown in Fig. (7). The result shows that the proposed method gives little bit better accuracy for all databases because the facial features contain not only the low frequency components but also small number of high frequency components, as explained in section 4. Regarding 100

MLDA

100.0

DLDA

99.0

Accuracy (%)

Cumulative Match Score

SLDA 98

SLDA

96

94

92

90 SLDA

88

98.0 97.0 96.0 95.0

MLDA

94.0

DLDA

93.0

RLDA

92.0

SLDA

SLDA DLDA

86 ITS

ORL

MLDA INDIA

UNRA M

91.0 1

2

3

4

5

6

7

8

9

10

Rank

Figure (7): the recognition accuracy (a) and CMS (b) comparison of LDA based face recognition.

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 10

to the Cumulative Match Score, the proposed method show better performance than established LDA in the same facial features size. In this case, the MLDA was show better accuracy and CMS because the LDA discriminate the facial features like the Mahalanobis distance classifier. However, the MLDA has problems in terms of singularity. In this research, it is solved by reducing the original data size using multiresolution wavelet transforms (MWT). By MWT the original data can be compacted to 49 elements of 16384 elements, it means the original data is compressed by about 99.66%. The second experiment was performed to investigate both training time and querying time and compare with the established LDA’s time consumption. The results shows that the proposed method requires shorter training and querying times than LDA’s time process, as show on Fig 9(a). Regarding to the retraining problem the MLDA method require very short time when new class is added to the system as shows in Fig. 9(b). This result can be achieved because the MLDA does not require eigen value analysis and the covariance analysis do not depend on the global mean of the training set. 0.065

35 30

Querying Time (Sec.)

Querying Time (Sec.)

0.06

0.055

0.05

0.045

MLDA

DLDA

RLDA

SLDA

0.04

25 20 15 v

MLDA

DLDA

RLDA

SLDA

10 5 0

40

50

60

70

80

90

100 110 120 130 140 150 160

Number of Class (Gradualy Training)

40

50

60

70

80

90

100 110 120 130

140 150

160

Number of Class (Gradualy Training)

(a) (b) Figure (9): the time consumption comparison of LDA based method when the training was performs gradually: (a) querying time, (b) training time. There are two aspects, which can be used to justify a good recognition system: first, how well the system can match image from the same people; second, how well the system distinguish images from different people [9]. Therefore, the last test was performs in order to know the performance of the proposed method in term of ROC analysis. The main parameters that want to know are equal error rate (ERR), false acceptance rate (FAR), and false rejection rate (FAR). The detail explanations of the ROC can be found in Ref. [9]. The system which performs perfect classification is denoted by 100% true positive rate and 0% false positive rate or the value of ERR is small or close to zero. The last experiment was performed on all databases, 49 elements facial features size, and 5 faces training per class. For the tested databases, the ROC

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 11

0.25 MLDA

RLDA

SLDA

DLDA

False Rejection Rate

0.2

0.15

0.1

c

0.05

0 0

0.05

0.1

0.15

0.2

0.25

False Acceptance Rate

Figure (10): ROC analysis of LDA based face recognition. curve show that the EER of the proposed method is almost the same with EER of the established LDA, however our method gives better success rate, requires less time consumption, and can solved the retraining problem. All of the experimental results show that the proposed method has good performance, robust to tested face databases, and need short training and querying times. This performance can be achieved because facial features have good low-frequency resolution representation. In this case, the low frequency component is good enough for face image representation because most information of signal can be found in low frequency component, as shown in Fig. (3). It can be described that if an image is transformed to frequency domain and the high frequency components are removed, the reconstruction image will loss a little significant information. This phenomenon was successfully used for signal compression. Moreover, wavelet decomposition property gives advantage for features extraction, such as it has good capability to separate information signal to low frequency components and their coefficients are uncorrelated with other frequency indices. The multiresolution wavelet decomposition is an efficient way of reducing the original data dimensional. In this paper, we show that the original data size can be reduced about 99.66% of original size (i.e., 49 elements of 16384 elements), while the success rate is high enough. Also note that facial features vector is built by fast wavelet transforms and the M-LDA based classification just need mean of each face class and global covariance to classify face image to a person’s class. Computational complexity of wavelet decomposition is linear with the number (N) of computed coefficients (O(N)), where N number of data. Therefore, our method needs short training and querying times

Proceedings of the 6th ICEENG Conference, 27-29 May, 2008

EE096 - 12

6. Conclusions: The proposed method is alternative face recognition because it gives good enough performance i.e. high enough success rate, short time processing, and small enough EER. In addition, the multiresolution wavelet transforms is an efficient way for reducing the dimensional size of original image and it requires a short decomposition time. Moreover, the proposed method can cover the retraining problem as explained in section 3.3 and is proved in the second experiment (Fig. 9(b)). However, this method has to test in large databases size for knowing its accuracy, CMS, and EER. Moreover, this process needs some improvements, such as considering other color space component and implementing the hybrid facial features (i.e DCT and DWT) to make the powerful facial features. Next, the research will focus on face image preprocessing analysis and finding optimum classification method in order to increase the accuracy. References: [1] Chellappa, R., Wilson, C., and Shirohey, S., Human and Machine recognition of faces: A survey, In Proc. IEEE, vol. 83, no. 5, P. 705-740, 1995. [2] Lu, J., Plataniotis K.N., and Venetsanopoulus A.N., “Face Recognition Using LDABased Algorithmn”, IEEE Transactions on Neural Networks, vol. 14, no. 1, P. 195200, 2003. [3] Chen, W., Er, Meng J., and Wu S., PCA and LDA in DCT Domain, ELSEVIER Pattern Recognition Letter, 26, P. 2474-2482, 2005 [4] Garcia C., Zikos G., and Tziritas G., Wavelet Packet Analysis for Face Recognition, Image and Vision Computing, 18, P. 289-297, 2000 [5] Hafed, Ziad M., and Levine, Martin D., Face Recognition Using the Discrete Cosine Trasnforms, International Journal of Computer Vision, 43(3), P. 167-188, 2001 [6] Dai, Dao-Qing and Yuen P.C., Wavelet Based Discriminant Analysis for Face Recognition, ELSEVIER Applied Mathematics and Computation, 175, P. 307-318, 2006 [7] H. Yu, J. Yang, A direct LDA algorithm for high-dimensional data—with application to face recognition, Pattern Recognition 34, P. 2067–2070., 2001 [8] Gao H, Davis James W., Why direct LDA is not equivalent to LDA, ELSEVIER Pattern Recognition Letter, 39, P. 1002-1006, 2005 [9] Wijaya, I G.P., Uchimura, K., and Hu, Z., Face Recognition Based on Holistic Information and Minimun Mahalanobis Classifier, Proceedings of the ACCV 2007 Workshop Subspace 2007 Tokyo, P. 53-60, November 2007 [10] http://vis-ww.cs.umass.edu/~vidit/IndianFaceDatabase [11] http://www.cl.cam.ac.uk/research/dtg/attarchive/facedatabase.html [12] Wijaya, I G.P., Uchimura, K., and Hu, Z., Multipose Face Recognition Based on Frequency nalaysis and Modified LDA, The Journal of the Institute of Image Electronics Engineers of Japan, vol. 37, no. 3, P. 231-243, 2008