Uma abordagem de classificação de impressões digitais utilizando redes neurais convolucionais profundas

• Ano de defesa: 2021
• Autor(a) principal: Dantas, Diogo Ventura
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Redes neurais convolucionais; Classificação de Impressões digitais; Biometria; Identificação; Convolutional neural networks; Fingerprint classification; Biometrics; Identificacion; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/24475

Society is increasingly integrated and connected, and many of the most mundane tasks are already being done remotely. This scenario brought several benefits but also caused problems such as fraud and data leakage. The use of authentication mechanisms based on biometric features, such as fingerprints, face, and iris, has gained increasing rele vance in the search for security. The problem of fingerprint identification is not trivial. It is still an open problem, especially with the emergence of artificial intelligence techniques, since no algorithm is error-free. The growing mass of data creates a need for reliability and speed. This work proposes to use deep machine learning techniques to perform the classification of fingerprints. This process can be used in a recognition system to speed up the search and reduce the number of comparisons with the database. This filtering process helps to obtain a lower error rate and a faster speed in the identification process. Several approaches to solving this problem can be found in the literature. Still, most of them are dedicated to solving a simplification of the original question, using four classes instead of the five classes initially defined. This approach is used to increase the method’s accuracy rate and decrease the complexity of the problem. The results of this work re vealed that it is possible to achieve results similar to those found in the literature using the broader definition of the problem. The proposed method reached an accuracy in the classification of 95,273% in the database NIST-4.