Re-identificação de pessoas em imagens através de características descritivas de cores e grupos

• Ano de defesa: 2015
• Autor(a) principal: Salamon, Nestor Ziliotto
• Orientador(a): Musse, Soraia Raupp
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Porto Alegre
• Palavras-chave em Português: INFORMÁTICA; PROCESSAMENTO DE IMAGENS; RECONHECIMENTO DE PADRÕES
• Link de acesso: http://hdl.handle.net/10923/7212

The re-identification of people and groups in real environments is still a difficult task due to several changes in illuminance, viewpoints, image resolution and many other challenges. In this work we propose a person re-identification approach applied as a softbiometric tool. The goal is to recognize a person by combining low-level and high-level (when possible) description of him/her, such as color appearance of clothes or objects he/she is carrying on (low-level) and group belonging (high-level).The input features for each person are manually informed by a user using sample patches from any source (a gallery repository, a picture taken or a color palette selection) and semantically organized in a 2D body model. In addition, groups of two persons (both defined as 2D body models) are specified. Finally, each 2D model is then confronted with databases using a color distance based metric, extended through an adaptive threshold and, if applicable, the color signatures of both persons into the group is used to search for a group composition with such characteristics.