다중 바이오 인식 시스템에서의 개인 인증을 위한 RBF 기반 패턴 분류 알고리즘 연구 = (A) Study of RBF based pattern classification algorithm for person authentication in Multimodal Biometrics

다국어 초록 (Multilingual Abstract)

Multimodal biometrics is a technology for person authentication and verification which employ multiple types of biometrics data. Multimodal biometrics is expected to compensate the limitation of unimodal biometrics. In particular, the score-level fusion approach has gained lots of attentions which combines matching scores from unimodal systems to make final decision. In this thesis, We investigated the RBF based score-level fusion approach where a pattern recognition algorithm seeks the optimal decision boundary to classify score feature vectors obtained from several unimodal biometrics system for each sample. Furthermore, we considered the qualities of matching scores in the process of RBF based score-level fusion, unlike most conventional score-level fusion methods which assume that all the matching scores are of the same quality. Such assumption may cause the problem not to reflect such situation that the qualities of the matching scores from certain unimodal systems are relatively low. To deal with this problem, the RBF based score-level fusion approach is proposed which incorporates the quality information of the scores in developing classification models. Specifically, the quality information is employed in a separated way to choose the centers in learning phase and compute trade-off coefficients in authentication phase. For performance evaluation, we carried out the experiment using NIST BSSR1 data. The experiment results show that our proposed method is superior to the typical RBF based score-level fusion without quality as well as the unimodal biometrics.

목차 (Table of Contents)

• Ⅰ. 서론 = 1
• 1.1 연구 배경 = 1
• 1.2 기존 연구 = 4
• 1.3 논문의 목적 및 구성 = 6
• Ⅱ. 관련 연구 = 7
• 2.1 다중 바이오 인식 Fusion 방법 = 7
• 2.2 스코어 정규화(Score Normalization) = 11
• 2.3 Quality 관련 연구 = 14
• 2.4 성능 평가 척도 = 16
• 2.4.1 Confusion matrix를 이용한 평가 척도 = 16
• 2.4.2 ROC 및 DET curve = 19
• Ⅲ. RBF 기반 Score-Level Fusion = 22
• 3.1 RBF 모델 및 학습 방법 = 22
• 3.2 다중 바이오 인식 실험 및 평가 = 30
• 3.2.1 커널 개수(m)및 너비(σ)선정 방법에 따른 성능 평가 = 34
• 3.2.2 정규화 방법에 따른 성능 평가 = 37
• 3.2.3 성능 비교 = 39
• Ⅳ. Quality 고려한 RBF 기반 Score-Level Fusion = 42
• 4.1 Quality 고려한 다중 바이오 인식 = 42
• 4.2 Quality 고려한 다중 바이오 인식 실험 및 평가 = 44
• 4.2.1 실험 데이터 = 44
• 4.2.2 성능 비교 = 45
• Ⅴ. 결론 = 47
• 참고문헌 = 48