Aplicação de rede neural neocognitron para reconhecimento de atributos faciais

• Ano de defesa: 2003
• Autor(a) principal: Hirakuri, Marcelo Hiroshi
• Orientador(a): Saito, José Hiroki
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• País: BR
• Palavras-chave em Português: Reconhecimento de padrões; Redes neurais; Rede neural neocognitron
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/319

This work presents a methodology for predetermined facial attributes recognition preestablished. This work realizes the recognition of facial attributes previously established: right eye, left eye, nose and lip, using an alternative implementation of Neocognitron Neural Network, robust for distortions and shifts on input patterns. The implementation of the neural network (NEOFAR - Neocognitron for Facial Attributes Recognition), consists of two sub-networks: Neural Network for Detecting of Control Points (NNDCP), which is used for edge detection, line detection, and line end detection, during the training phase; and Neural Network for Facial Recognition (NNFR), which is used for complex features manipulation, during the recognition phase. The performance evaluation showed the viability of project for many applications with excellent future perspectives. The performance tests showed rightness taxes over 85%.