Extraction des expressions faciales en presence de rotation hors plan

摘要

This article presents a new method for facial content analysis and face deformation extraction on video sequence. Rigid movements and inner deformation of the face are tracked with a classical AAM. The fit model is consequently projected onto a set of vectors representing the head rigid movement only; a new model private from inner deformation can then be created. The extraction of the image visual content on the very position of this model allows to create a new image sequence showing the righted inner face with its expressivity conserved. The method proposes an original solution to perform face expression extraction in presence of out-of-plane rotation. We will eventually introduce a new method of expression recognition used for testing the method.