Reconhecimento facial usando descritores locais e redes complexas

• Ano de defesa: 2016
• Autor(a) principal: Piotto, João Gilberto de Souza
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Percepção facial; Representações dos grafos; Percepção de padrões; Face perception; Representations of graphs; Pattern perception; CNPQ::CIENCIAS EXATAS E DA TERRA; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/2568

The search for biometric scanning methods has grown a lot due to government, military and commercial needs. Researches indicate the face recognition market will move billions of dollars in next years. Thus, finding methods to specific situations drives new advances in this area. Each application face recognition requires a particular solution. There are cases the response time is the most important factor; others require that face must be classified even if partially. In all these situations, accuracy and robustness may be the most important attributes. However, in most cases, these features behave as inverse greatness: increasing the confidence level of the results the method performance will be affected. Therefore, create the method which balances these factors is essential for construction of acceptable solutions. This paper presents a new face recognition algorithm based on local descriptors and complex networks. The method is able to concentrate the information before distributed by various point descriptors, in a unique feature vector. It makes the classification step faster and more efficient. Furthermore, another focus of the method is reduce pre-processing steps, avoiding unnecessary processes. The experiments were conducted with faces datasets well known in the literature, revealing accuracy rates of up to 98.5%. The technique also showed good results when there was noise in the samples, often derived from objects present in the composition of the scene. For additional analysis, classical facial recognition algorithms were subjected to the same data set, generating comparative results between both methodologies.