한국해양대학교

Detailed Information

Metadata Downloads

SVM을 이용한 다중 생체 인식 시스템의 설계 및 구현에 관한 연구

Title
SVM을 이용한 다중 생체 인식 시스템의 설계 및 구현에 관한 연구
Alternative Title
A Study on Design and Implementation of Multi-Modal Biometrics Recognition System using SVM
Author(s)
김정훈
Publication Year
2007
Publisher
한국해양대학교
URI
http://kmou.dcollection.net/jsp/common/DcLoOrgPer.jsp?sItemId=000002174450
http://repository.kmou.ac.kr/handle/2014.oak/8477
Abstract
This study intends to design and implement the multi-modal biometric system by applying the RFID recognition system presently has been greatly issued in the ubiquitous society to the multi-modal biometric system using the fingerprint and the speech widely used in the biometrics, in order to improve FAR and FRR of the existing single biometric system.

The recognition system is designed with TI’s DSP TMS- 320C32, TMS320VC5509 and AVR so that it can be implemented as embedded subsystems and is connected to the RFID system in serial while the fingerprint recognition system and the speech recognition system are connected in parallel.

The hardware is built with the algorithms excellent in real-time for efficient real-time implementation. Among them, SVM algorithm, which is the most essential in this study, distinguishes the registered from the impostors by using the distance value obtained two recognition systems. This integrate system improves FAR and FRR. In this study, we performed the SVM performance test the number of vectors within the support vector is the least, which is 27(10.3%), with the Bspline kernel function among 7 kernel functions
therefore, it is verified that the kernel function of this study gives the best efficiency. We compared with other algorithms(weight sum, fuzzy integral, decision tree) in order to evaluate the performance of the multi- modal biometric system using the linear Bslpine function, and the result shows that it is much more efficient with maximum 12% difference.
Appears in Collections:
전자통신공학과 > Thesis
Files in This Item:
000002174450.pdf Download

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse