3D-Aided Face Recognition from Videos

Baptiste Chu 1 Sami Romdhani Liming Chen 1
1 imagine - Extraction de Caractéristiques et Identification
LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
Abstract : The deployment of cameras for security control allows for video stream to be used as input for face recognition (FR). However, most state of the art FR SDKs are generally specifically tuned for dealing with frontal and neutral face images, whereas expression and pose variations, which typically occur in unconstrained settings, e.g., video images, are still major challenges for reliable FR. In this paper, we aim to endow the state of the art FR SDKs with the capabilities to recognize faces in videos. For this purpose, given a video sequence of a person, an extended 3D Morphable Model (3DMM) is used to generate a novel view of this person where the pose is rectified and the expression neutralized. We present a 3DMM fitting method specifically designed for videos to take into account the temporal properties, making use of multiple frames for fitting. Moreover, some constraints of smoothness are used to get a better estimation of its 3D shape and to separate its expression component from its identity component. Finally, we evaluate the proposed method on the Prison Break TV serial and demonstrate its effectiveness using a standard commercial FR SDK.
Type de document :
Communication dans un congrès
2014 5th IEEE European Workshop on Visual Information Processing (EUVIP), Dec 2014, Paris, France. pp.1-6, 2014
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01313196
Contributeur : Équipe Gestionnaire Des Publications Si Liris <>
Soumis le : lundi 9 mai 2016 - 16:12:05
Dernière modification le : mardi 10 mai 2016 - 01:05:55

Identifiants

  • HAL Id : hal-01313196, version 1

Collections

Citation

Baptiste Chu, Sami Romdhani, Liming Chen. 3D-Aided Face Recognition from Videos. 2014 5th IEEE European Workshop on Visual Information Processing (EUVIP), Dec 2014, Paris, France. pp.1-6, 2014. <hal-01313196>

Partager

Métriques

Consultations de la notice

48