Mi SciELO
Servicios Personalizados
Articulo
Indicadores
Citado por SciELO
Links relacionados
Similares en SciELO
Compartir
Ingeniería Electrónica, Automática y Comunicaciones
versión On-line ISSN 1815-5928
Resumen
MENDEZ, Nelson; NICOLAS, Miguel A. y MENDEZ-VAZQUEZ, Heydi. Profile Face Image Frontalization based on landmark points and 3D Generic Elastic Model. EAC [online]. 2019, vol.40, n.3, pp. 72-78. Epub 08-Sep-2019. ISSN 1815-5928.
The first step in most face recognition systems is the alignment of the detected faces. When the faces present large pose variations, the alignment process should be able to generate frontal face images. Different methods have been proposed for face frontalization, but in general, they are not able to reconstruct a frontal image from a complete face profile image. In this paper we extend a face frontalization method based on the 3D Generic Elastic Model (3DGEM), in order to be able of recovering frontal faces from profile images. First, it is determined if the image is a right or a left profile. We train a method based on Active Shape Models (ASM) in order to detect landmark points on profile faces. Then a relationship is established between the profile landmark points and the landmarks located in a 3D face model, which is efficiently adjusted to the image and further frontalized. Face symmetry is taken into account for projecting the appearance of the frontalized face image. The proposal was evaluated by frontalizing the facial images in ICB-RW and CFPW databases. We show the importance of the frontalization process in the classification accuracy.
Palabras clave : face frontalization; profile faces; profile landmarks detection.