Dissertation > Excellent graduate degree dissertation topics show

Dimensionality reduction algorithm based on subspace biometric Applied Research

Author: YangXiaoChao
Tutor: ZhouYue
School: Shanghai Jiaotong University
Course: Pattern Recognition and Intelligent Systems
Keywords: Biometrics Dimensionality reduction Video Surveillance Gait Recognition Face Pose Estimation Face Recognition
CLC: TP391.41
Type: Master's thesis
Year: 2008
Downloads: 55
Quote: 2
Read: Download Dissertation

Abstract

With the advancement of technology and the improvement of safety awareness, rapid and effective authentication actual demand increasingly urgent. Many systems require reliable identification mechanism to confirm or determine the need to provide services to an individual's identity. Authentication mechanisms aim is to ensure the provision of access to the service is available to legitimate users, but not others. For example, information management systems, notebook PCs, mobile phones, ATMs, building import and export. Traditional authentication systems by identifying \In this thesis, to study biometric based on individual physiological and / or behavioral characteristics determine the identity automatically. The technical processing corresponding to the nature of identity question: \Biometric systems can be seen as a signal with signal detection and pattern recognition architectures. It from the original input signal to extract the biometric characteristic having an identification information, and the detection of these special characteristics of the database comparison, the final identity of the owner of the signal determined. The properties of high-dimensional signal data processing and application tend to be an obstacle, which is reflected in the associated computational complexity is high and the result is not optimal. Dimensionality reduction is made from high-dimensional data about the process and reduced low-dimensional nature of the data used to reveal low-dimensional structures. As to overcome the \Accordingly, the paper studies the subspace-based biometric identification dimensionality reduction method, the main contributions are as follows: 1) an overview of developments in the field of biometric technology, summarizes the various biometric some of the advantages, strengths, application limits, and about privacy attention. 2) introduction of a variety of commonly used dimensionality reduction algorithm theory and relationships. This paper focuses on the algorithm based on spectral analysis, that is to solve the matrix eigenvalue decomposition algorithm. We introduce an algorithm including traditional algorithms: PCA and LDA; manifold learning algorithms: LLE, LPP, ISOMAP, LE, LTSA and HLLE. 3) dimensionality reduction method based on subspace Gait Recognition. Proposed a dynamic information through statistical gait walking pattern recognition method. On behalf of each class period average silhouette image (GEI) analysis of variance obtained dynamic weights mask (DWM). GEI through DWM dynamic and original shape information to obtain a new enhanced gait characterization EGEI. To increase identifiable information, the use of a set of Gabor wavelet convolution on EGEI then used to identify common vector analysis (DCV) will result in high-dimensional convolution low-dimensional space representation; propose an efficient gait energy image based on the identification method. First of generating synthetic GEI rich training set number of samples. Then use the previously neglected in the literature with good recognition performance Gabor phase information as the identity and learning algorithm using flow ECIC insinuate (LPP) this high-dimensional data in low-dimensional space representation. By using a simple classification scheme on the database at USF gait comparative experiments prove that the two proposed methods to improve the effectiveness of the recognition performance. 4) Video-based real-time face identification method. The face detection and tracking, pose estimation and recognition into the video surveillance system. The system first detects the candidate face and achieve a healthy track, then right to pre-judge the face region, to identify the face orientation, if it is positive towards the human face is the identity of judgment. Compared to traditional static identification scheme, the system can be achieved without initialization, without actively cooperate, automatic real-time recognition. Greater efficiency and flexibility.

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