Reconhecimento facial através de limiares individuais por padrão de textura e sub-região facial

• Ano de defesa: 2016
• Autor(a) principal: D'Amore, Luiz Angelo
• Orientador(a): Marengoni, Mauricio
• Banca de defesa: Martins, Valéria Farinazzo , Omar, Nizam , Marana, Aparecido Nilceu , Gonzaga, Adilson
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Programa de Pós-Graduação: Engenharia Elétrica
• Departamento: Escola de Engenharia Mackenzie (EE)
• País: Brasil
• Palavras-chave em Português: reconhecimento facial; sistemas biométricos; padrões binários locais; visão computacional; reconhecimento de padrões
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO
• Link de acesso: http://dspace.mackenzie.br/handle/10899/24277

In this work are presented the development and results of a new method of facial recognition based on the use of individual thresholds by texture pattern and facial sub region, with applicability in biometric systems of access control in controlled environments. Among the biometric systems, facial recognition systems have certain prominence in relation to other systems, because not being intrusive. The acquisition of the facial image is made with a distance between the camera and the subject. The goal of the proposed method is to reduce the computational cost, compared to other existing methods, without prejudice for facial recognition rate. The proposed method uses a descriptor composed of four sub-descriptors, each of formed by texture patterns. Each of the sub descriptors have size defined by the number of sub regions facial. In the comparative process between descriptors, the method uses the individual thresholds of texture pattern and facial sub region. This work presents three experiments, each subdivided into specific tests. The first experiment was conducted to evaluate variations related to the use of uniform local binary patterns (LBPu2), the results obtained were used as a comparative basis for the other experiments of this work. The second experiment aimed to evaluate the proposed method for reducing the size of the standard facial descriptor LBPu2 of 59 textures. The third experiment was conducted to evaluate the proposed method based on the use of individual thresholds by texture pattern and facial sub-region, method that aims to improve the classification process of facial recognition. For the experiments was used the frontal facial images of "The Facial Recognition Technology Database" (FERET) and "FEI Face Database" (FEI) sets, with light variations, expressions and no restrictions on the use of beard, mustache or glasses. After the experiments, it was possible to observe that the proposed method reduces considerably the computational cost for the facial recognition systems, besides improving its recognition rate, surpassing the academic results already published.