Author: Ma, Jiahuan
Title: Face recognition based on manifold learning
Degree: M.Sc.
Year: 2010
Subject: Hong Kong Polytechnic University -- Dissertations
Human face recognition (Computer science)
Face perception
Principal components analysis
Multivariate analysis -- Data processing.
Linear models (Statistics) -- Data processing
Department: Department of Electronic and Information Engineering
Pages: iii, 94, v leaves : ill. ; 31 cm.
Language: English
Abstract: Face recognition, due to its intuitive traits, imperceptibility and user-friendly characteristics, is a hot topic in the field of pattern recognition. Face-recognition devices are now used in society. such as face-recognition security doors, face-recognition attendance systems, and so on. Manifold learning is a popular approach for face recognition, which finds a potential low-dimensional manifold embedded in a high-dimensional space, if it exists, to represent face images. There are many manifold learning methods. In this thesis, we focus on the Principal Components Analysis (PCA) method, the Linear Discriminant Analysis (LDA) method, and the Locality Preserving Projections (LPP) method. The PCA algorithm is designed to best represent the original data; thus it is not optimal for classification, but it is a good method for dimensionality reduction. The LDA algorithm tries to find projections that maximize the between-class distance and minimize the within-class distance at the same time. For LDA, the biggest challenge is the small-sample-size problem. Theoretically speaking, the LDA algorithm, which belongs to a supervised learning method, performs better than the PCA algorithm, which belongs to an unsupervised learning method in face recognition. The LPP algorithm, which is an approximation of a nonlinear manifold learning method, retains the inherent local manifold structure of samples while extracting the most representative features to reduce the dimensionality. The traditional linear algorithms, such as PCA and LDA. do not have this ability. Therefore, with the aspect of retaining the local characteristics, the LPP algorithm has obvious advantages. However, because it also suffers from the same small-sample-size problem as the LDA method does, the pre-processing for dimensionality reduction is needed before applying the LPP algorithm. To achieve a better performance in face recognition, we also employ Gabor features to represent face images. These features can be extracted using Gabor filters, which function like a bank of band-pass filters for feature extraction. A major advantage of the Gabor feature is that the feature is robust to illumination variation. In the thesis, we have investigated the performances of different popular face-recognition algorithms under different conditions. We have also proposed a method that combines the Locality Preserving Projections algorithm and the Fisher Linear Discriminant algorithm at ranking level to improve the face-recognition performance. This can utilize the advantages of both methods and maximize the use of the information obtained from the testing images, as well as from the training samples.
Rights: All rights reserved
Access: restricted access

Files in This Item:
File Description SizeFormat 
b23568100.pdfFor All Users (off-campus access for PolyU Staff & Students only)29.58 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: