Facial identification problem : a tracking based approach

This paper presents a method for face identification using a query by example approach. Our technique is suitable for use within Ambient Security Environments and is robust across variations in pose, expression and illuminations conditions. To account for these variations, we use a face template matching algorithm based on a 3D head model created from a single frontal face image. Thanks to our tracking-based approach our algorithm is able to extract simultaneously all parameters related to the face expression and to the 3D posture. With these estimates, we are able to reconstruct a frontal, neutral and normalized image on which dissimilarity analysis for identification and anomalies detection is performed. Our tracking process combined with dissimilarity analysis was tested on Kanade-Cohn database for expression independent identification and several other experimental databases for robustness.