Smart anonymization

A new application of image processing is explored: the fine anonymization of faces. The objective is to anonymize the persons present in images %so that they can be broad-casted while keeping the relevant information allowing facial expressions and emotions analysis. It is a difficult problem. Indeed, obtaining the anonymity implies deteriorating the faces. At the opposite, allowing the analyses requires keeping enough information on the faces. We propose a novel method based, on one hand, on a face detection and landmarks extraction and, on the other hand, an adaptative filtering based on a minimization of a variational functional is applied, in order to anonymize the face while preserving facial expression and emotions.

Anonymisation fine de visages avec préservation des expressions faciales
G. Letournel, A. Bugeau, V.-T. Ta , J.-P. Domenger, M. C. Manes Gallo
RFIA 2014