A carregar...
Publicação
Sistema biométrico baseado em eletrocardiografia e impressão digital
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.39 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O reconhecimento biométrico estabeleceu-se como uma alternativa mais segura em relação a métodos tradicionais como palavras-passe, cartões ou tokens de utilização pessoal. Contudo os sistemas unimodais apresentam algumas falhas significativas no que diz respeito à taxa de falsas aceitações, presença de ruído e suscetibilidade a tentativas de falsificação. Embora os sistemas biométricos multimodais ofereçam diversas vantagens relativamente a estes anteriores, a sua adoção ainda é limitada uma vez que se estima que apenas cerca de 30% dos sistemas de autenticação utilizem duas ou mais modalidades biométricas. Esta dissertação propõe a utilização de uma abordagem bimodal, combinando dois sistemas biométricos distintos para melhorar a precisão e segurança do utilizador. O primeiro sistema baseia-se na impressão digital, uma das formas mais comuns e amplamente utilizadas de identificação biométrica. O segundo sistema consiste numa modalidade de biometria que tem o potencial de complementar as abordagens existentes devido à sua natureza intrínseca, a utilização do eletrocardiograma (ECG) como fonte biométrica. O sistema desenvolvido inclui a aquisição automática dos sinais de ECG, extração de minúcias a partir das imagens de impressões digitais, e um módulo de fusão a nível de score que pondera os contributos de ambas as modalidades. A implementação foi realizada em Python num Raspberry Pi 5, com suporte para registo, autenticação e identificação de utilizadores. Foram realizados testes experimentais com um conjunto de 12 participantes, avaliando métricas como a Taxa de Erro Igual (EER), a Taxa de Aceitação Indevida (FAR) e a Taxa de Rejeição indevida (FRR). Os resultados demonstraram que essa abordagem atingiu um melhor desempenho quando comparada com as metodologias de forma isolada, reduzindo significativamente as taxas de erro e aumentando a robustez perante potenciais ataques de spoofing. A integração deste sistema mostrou-se uma solução viável e promissora para sistemas de autenticação biométrica de baixo custo.
Abstract Biometric recognition is currently established as a more secure alternative to traditional methods such as passwords, cards or personal tokens. However, unimodal systems have some significant flaws in terms of false acceptance rates, noise presence and susceptibility to forgery attempts. Although multimodal biometric systems offer several advantages over the former, their adoption is still limited, as it is estimated that only about 30% of authentication systems use two or more biometric modalities. This dissertation proposes the use of a bimodal approach, combining two distinct biometric systems to improve accuracy and user security. The first system is based on fingerprinting, one of the most common and widely used forms of biometric identification. The second system consists of a biometric modality that has the potential to complement existing approaches due to its intrinsic nature, the use of the electrocardiogram (ECG) as a biometric source. The developed system includes automatic acquisition of ECG signals, extraction of minutiae from fingerprint images, and a score-level fusion module that weighs the contributions of both modalities. The implementation was carried out in Python on a Raspberry Pi 5, with support for user registration, authentication, and identification. Experimental tests were carried out with a group of 12 participants, evaluating metrics such as Equal Error Rate (EER), False Acceptance Rate (FAR) and False Rejection Rate (FRR). The results showed that this approach achieved better performance when compared to the methodologies in isolation, significantly reducing error rates and increasing robustness against potential spoofing attacks. The integration of this system proved to be a viable and promising solution for low-cost biometric authentication systems.
Abstract Biometric recognition is currently established as a more secure alternative to traditional methods such as passwords, cards or personal tokens. However, unimodal systems have some significant flaws in terms of false acceptance rates, noise presence and susceptibility to forgery attempts. Although multimodal biometric systems offer several advantages over the former, their adoption is still limited, as it is estimated that only about 30% of authentication systems use two or more biometric modalities. This dissertation proposes the use of a bimodal approach, combining two distinct biometric systems to improve accuracy and user security. The first system is based on fingerprinting, one of the most common and widely used forms of biometric identification. The second system consists of a biometric modality that has the potential to complement existing approaches due to its intrinsic nature, the use of the electrocardiogram (ECG) as a biometric source. The developed system includes automatic acquisition of ECG signals, extraction of minutiae from fingerprint images, and a score-level fusion module that weighs the contributions of both modalities. The implementation was carried out in Python on a Raspberry Pi 5, with support for user registration, authentication, and identification. Experimental tests were carried out with a group of 12 participants, evaluating metrics such as Equal Error Rate (EER), False Acceptance Rate (FAR) and False Rejection Rate (FRR). The results showed that this approach achieved better performance when compared to the methodologies in isolation, significantly reducing error rates and increasing robustness against potential spoofing attacks. The integration of this system proved to be a viable and promising solution for low-cost biometric authentication systems.
Descrição
Palavras-chave
Impressão digital Eletrocardiograma Biometria Multimodal Fingerprint Electrocardiogram Biometrics IPL/IDI&CA2024/M-IA-RCH_ISEL