Avaliação de modelo de misturas aplicado à classificação de impressões digitais segundo a arquitetura pcasys

• Ano de defesa: 2010
• Autor(a) principal: Atencio, Anibal Cotrina
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: PCASYS (Sistema classificador de impressões digitais); Impressão digital (Computação); Distribuição (Teoria da probabilidade); Algoritmos de expectativa de maximização; Análise multivariada; Classificação; Biometria; Eletrônica Industrial, Sistemas e Controles Eletrônicos; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9601

Currently, biometric recognition systems have achieved great development and has been used in various fields including trade, finance and security. In the specific case of fingerprint recognition, automatic fingerprint identification systems (AFIS) are able to identify individuals located in large databases, dealing with tens of millions of samples, giving rise to a very high computational cost. Thus, the classification of fingerprints is an important problem to be treated as it reduces the search time by an individual, reducing the search space to a particular subgroup. In this study we assessed if, for the architecture of fingerprint classification known as PCASYS (Pattern-Level Classification Automation System), it is possible to assume that classes can be modeled using Gaussian probability density function and can achieve classification results that can be compared with those obtained with more mature techniques such as those based on neural networks. For this, it is used the Gaussian mixture models (GMM), with supervised and unsupervised approaches in parameter estimation task. It was evaluated the results comparing the rates of misclassification of the proposed technique with the results achieved with other techniques, anyway, we analyzed statistically the estimated parameters of Gaussian mixture models using hypothesis testing and mutual entropy. It was tested using the image bank fingerprints of number 4 of National Institute of Standards and Technology (NIST) considering the 4 class problem and 5 class problem considering the natural distribution and balanced distribution.