Font Size: a A A

A Study Of Human Face Recognition

Posted on:2001-07-24Degree:DoctorType:Dissertation
Country:ChinaCandidate:D L ZhouFull Text:PDF
GTID:1118360182972396Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Biometrics is a kind of science of using individual personal characteristics to verify identity. It represents the most secure way to identify individuals because it provides a novel approach to recognize the identity, or verify the claimed identity through a unique, highly reliable and robust physical characteristic or personal trait. Face recognition is one of the most outstanding techniques of biometrics. This dissertation is concerned with Automated Face Recognition. The main contributions of this dissertation can be summarized as follows: The approach of edge detection based on fuzzy sets is improved to be an automatic one. Two new membership functions are proposed and different enhancement operators are used for edge detection. The algorithm of edge detection based on fuzzy sets proposed by Pal and King is improved. This method is also extended to the edge detection of the multi-thresholds image. Color edge detection and color edge image redisplay based on K-L transform is also studied. Four thresholding algorithms are presented, these encompass thresholding of digital images based on the maximum correlation criterion, thresholding of digital images using two-dimensional correlation, maximum entropy thresholding algorithm based on gray level-gradient co-occurrence matrix, minimum error thresholding based on two dimensional histogram. The first one is computationally inexpensive and performs well. The others also perform well because these methods utilize both the gray level information of each pixel and its spatial correlation information within the neighborhood in an image. The problems of Feature extraction are analyzed and discussed deeply. A system of color face recognition based on eigenfaces method is proposed which color face image is transformed into gray level image using K-L transform proposed by Ohta. The techniques of face recognition base on Fisher linear discriminant are analyzed and studied. An approach of color face recognition using principal component analysis based on Fisher linear discriminant is presented. The methods of face recognition based on singular value feature vectors are discussed deeply and extended to the problems of face recognition using multi discriminant vectors. In order to reduce the dimensionality of the space spanned by singular value feature vectors, some algorithms of face recognition based on singular value feature vectors are proposed. These algorithms are called face recognition using principal component analysis based on singular value feature vectors, face recognition using Fisher linear discriminant analysis based on singular value feature vectors, face recognition using DKL transform based on singular value feature vectors respectively.
Keywords/Search Tags:Face Recognition, Edge Detection, Threshold, Feature Extraction, K–L Transform, Eigenface, Singular Value Feature, Fisher Linear Discriminant
PDF Full Text Request
Related items