Reconhecimento de faces 3D com Kinect

• Ano de defesa: 2014
• Autor(a) principal: Cardia Neto, João Baptista [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Computação; Processamento de imagens - Tecnicas digitais; Reconhecimento facial (Computação); Biometria; Imagem tridimensional; Sistemas imageadores
• Link de acesso: http://hdl.handle.net/11449/127666; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/31-08-2015/000844097.pdf

For person identification, facil recognition has several advantages over other biometric traits due mostly to its high universelly, collectability, and acceptability. When dealing with 2D face images several problems arise related to pose, illumination, and facial expressions. To increase the performance of facial recognition, 3D mehtods have been proposed and developedm since working with 3D objects allow us to handle better the aforementioned problems. With 3D object, it is possible to rotate the face around any axis, generate illumination that matches the one in the enviroment and even correct the deformation in the model due to facial expression. The mais problems with 3D facial recognition are: the high cost of the 3D cameras that have been generally employed, and intrusive way that such devices work. Some of them require that the subject remais completely still for several minutes while scanning, limiting, therefpre, the application deployment for uncontrollable enviroments. One alternative to those expensive cameras is the Kinect, a device developed by Microsoft to enchance gaming in the Xbok 360 console. Due to its capacites to generate depth images, Kinect is candidate device to be use for 3D face recognition, replacing the traditional 3D cameras. The mais problem with the Kinect is that it generates low-resolution images, making difficult the ask of precise facial recognition. The mais objective of this dissertation was to ptoposed some mehtods that have been proposed recently for 3D face recognition and to propose a neu method that combines 3DLBP and HAOG features. Experimental results obtained on the EURECOM 3D face database show that when 3DLBP and HAOG features are combineted the results can be better than they are used alone. We Have also proposed a method that increase the facial recognition performance when the faces present partial obstructions, by utilizing a symetric filling approach