1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Enhancement of face recognition using adaboost
Authors:	Faghani Mohsen (Md Jan Nordin Prof. )
Supervisor:	Md Jan Nordin, Prof.
Keywords:	Adaboost Face recognition Biometric identification
Issue Date:	2011
Description:	AdaBoost algorithm was presented by Freund and Schapire in 1995 for the first time. One of practical problems of Boosting with filtering methods needs training samples. This problem can be addressed by AdaBoost. In this algorithm there is the possibility of using training information. The aim of this algorithm is to find final classifier that has low error rate with respect to data distribution to training samples. This algorithm is different from other boosting algorithms. AdaBoost is set as adaptable according to errors of weak classifier in weak training models. Performance boundaries of AdaBoost are related to performance of weak training model. In this algorithm simple samples from training set from previous weak classifier that was classified correctly gain less weights and difficult classifier samples from training set that was classified incorrectly gain more weights. So AdaBoost algorithm concentrates to samples with difficult classification. In this work , using the results of classifier composition is one of the methods of increasing efficiency of face recognition systems that many researchers paid attention to it in recent years. However AdaBoost algorithm is as one of the efficient boosting algorithm that has been used as to decrease the dimensions of characteristic space extracted from face recognition systems. In our work for face recognition, Yale and ORL database are used for simulation. Each of them has specification that is useful for results analysis. In this work we used extracted specifications according to PCA and LDA conversion. For training and testing this system, 400 image form ORL database and 165 images from Yale database was used. AdaBoost as compound classifier could improve results with respect to singular classifier. The final classifier presented is not sensitive against different state of face and light changes. The method presented in ORL database has correct recognition of Adaboost & Bayesian & PCA methods 96.4% and improved the results of Bayesian method with PCA specification. Also the method presented in Yale database has correct recognition of Adaboost & Bayesian &PCA 94.3% that has improvement performance with respect to method with PCA specification. Finally by using proposed method with LDA specification in ORL database, the recognition percent is of Adaboost& Bayesian & LDA methods 95.3% in Yale database has correct recognition of Adaboost &Bayesian & LDA methods 93.6% that has better performance with respect to Bayesian method with LDA specification.,Master/Sarjana
Pages:	75
Call Number:	TA1650.F438 2011 3 tesis
Publisher:	UKM, Bangi
URI:	https://ptsldigital.ukm.my/jspui/handle/123456789/476203
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.