Please use this identifier to cite or link to this item:
Title: Dimensionality Reduction for Face Recognition Soft Computing Approaches
Researcher: Dinesh Kumar
Guide(s): Shakti Kumar and C.S. Rai
University: Guru Gobind Singh Indraprastha University
Completed Date: 2009
Abstract: Face Recognition, one of the three most popular biometrics (face, fingerprint and voice), always involves high dimensional data. As a consequence of increasing requirements due to its applications in the government and commercial sectors, the face recognition task has exhibited increased complexity and large storage capacity requirements. In order to reduce the complexity, it becomes essential to keep the dimensionality of the data as small as possible so as to make the system efficient as far as classification accuracy is concerned. The feature extraction/selection is done from the data to retain most of the information that represents the original data. The soft computing procedures make it possible to obtain the optimal features to be retained for future processing. Representation such as quotEigenfacesquot is based on Principal Component Analysis (PCA) that deals with second order statistics of the input. In this research work, a comparison of PCA has been done that includes Eigen decomposition (ED), Singular value decomposition (SVD) and Neural network based PCA for dimensionality reduction and face recognition. PCA does not take into account the higher order statistical dependencies. Independent Component Analysis (ICA), a generalization of PCA, addresses the high-order dependencies in the input. An ICA based technique that uses different source distribution models is proposed for face recognition. A comparison of the proposed method is carried out with the one that uses Infomax ICA algorithm with logistic function as the nonlinearity. The former yields results better than the latter one. A simplified view of information theory based different approaches to PCA and ICA is also presented. Besides PCA and ICA, Self organizing map (SOM) is another algorithm used for self organization or unsupervised learning that discovers the significant features in the input data without a teacher and has been successfully used for dimensionality reduction in face recognition. In this thesis, two SOM methodologies, i.e., local ...
Appears in Departments:University School of Engineering and Technology

Files in This Item:
File Description SizeFormat 
01 tilte certificate ack abst publication.pdfAttached File1.71 MBAdobe PDFView/Open
02 contents.pdf2.21 MBAdobe PDFView/Open
03 chapter 1.pdf4.3 MBAdobe PDFView/Open
04 chapter 2.pdf9.34 MBAdobe PDFView/Open
05 chapter 3.pdf6.17 MBAdobe PDFView/Open
06 chapter 4.0.pdf2.58 MBAdobe PDFView/Open
07 chapter 4.1.pdf1.51 MBAdobe PDFView/Open
08 chapter 5.pdf3.76 MBAdobe PDFView/Open
09 chapter 6.pdf5.94 MBAdobe PDFView/Open
10 chapter 7.pdf1.61 MBAdobe PDFView/Open
11 reference.pdf5.41 MBAdobe PDFView/Open

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