Dissertation > Excellent graduate degree dissertation topics show

Research on Face Recognition Methods with Subspace Linear Projections

Author: MaRuCheng
Tutor: LiuZunXiong
School: East China Jiaotong University
Course: Applied Computer Technology
Keywords: Face Recognition Principal Component Analysis Linear discriminant analysis Independent Component Analysis Binary tree support vector machine Non - negative matrix factorization
CLC: TP391.41
Type: Master's thesis
Year: 2009
Downloads: 151
Quote: 1
Read: Download Dissertation

Abstract

Computer face recognition technology has broad application prospects of computer face recognition method has become a hot research topic in the field of pattern recognition at home and abroad. Linear subspace projection method as one of the mainstream methods of face recognition, descriptive, computational cost, easy to achieve, it is to receive widespread attention. Paper focuses on face recognition the linear projection mainstream of subspace methods, theory and algorithms, a more in-depth study, concluded to deepen the understanding of the face recognition problem, the proposed algorithm in resolving good performance when the face recognition problem. The work of this study include the following aspects: (1) In this paper, a bilinear interpolation method of image scaling, the method is simple and effective; addition to the images mean variance normalization process, the method to overcome the light image adversely affected; method of principal component analysis for face recognition, these two methods combined image preprocessing, and achieved good results. (2) principal component analysis (Primary Component Analysis, PCA) and linear discriminant analysis (Linear Discriminant Analysis, LDA) is a linear projection method, this paper these two typical face recognition method detail. The principal component analysis as determined by the projection direction has a good ability to express, and the projection direction of the determined by linear discriminant analysis conducive to distinguish samples. ORL database as experimental subjects, the use of these two methods for feature extraction, the experimental results confirm the LDA feature extraction is better than PCA. (3) The depth study of the basic principle of independent component analysis and support vector machine analyzed their advantages and disadvantages. Of multi-class support vector machine classification algorithm based on improved binary tree support vector machine algorithm. Independent component analysis and improved binary tree support vector machines applied to face recognition combined ORL database simulation using independent component analysis for feature extraction, and then take advantage of the improved binary tree support vector machines for classification experiment to obtain a good recognition effect. (4) non-negative matrix factorization method is based on the characteristics of the local feature extraction method has been successfully applied to face recognition. The paper detailed analysis of the performance characteristics of the non-negative matrix factorization method, and its two improved algorithms - the local non-negative matrix factorization and sparse non-negative matrix factorization carried out in-depth research. The experimental results show that the algorithm of the above improvements not only feasible, but to a certain extent to effectively improve the correct rate of the face recognition.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile