Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/14803
Title: Sample selection and variance discriminant analysis for sample-based face detection
Authors: Yu, Wei
Keywords: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics
Issue Date: 2009
Source: Yu, W. (2009). Sample selection and variance discriminant analysis for sample-based face detection. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: In recent years, face detection has been a very active area of research. These technologies can be applied to the domains of computer vision, pattern recognition, and machine learning. Among the many existing categories of face detection algorithms, the sample-based method is one of the most widely-used approaches. The essence of the sample-based method is to solve a two-class classification problem of face versus non-face. Many classification algorithms such as the Naive Bayesian, Neural Network and Support Vector Machines (SVM) have been used for this purpose. This thesis showcases a research study into face detection technologies. This document is in two main parts. Firstly, in the sample preparation section, new passive sample selection and active sample generation algorithms are proposed to assist existing sample-based algorithms in solving the problem of face detection. Secondly, in the classification section, a new Bayesian-based classification method is proposed for face detection. Sample-based algorithms have generally resulted in the best reported face detection performance. Sample-based methods in the thesis mean the methods that extract features or select features based on the machine learning from samples. A face detection algorithm that depends on sample-based approaches must consider various issues. The primary issues include how to determine a suitable algorithm to construct the classifier, selecting representative samples and balancing training samples. One relevant approach to optimize face detection performance involves improving efficiency by selecting and adding useful samples into the training set without collecting new samples.
URI: https://hdl.handle.net/10356/14803
DOI: 10.32657/10356/14803
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
YuWeiThesis.pdf2.76 MBAdobe PDFThumbnail
View/Open

Google ScholarTM

Check

Altmetric


Plumx

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.