Expressive deformation profiles for cross expression face recognition

Expression based face recognition has been gaining more and more attentions recently. Most traditional expression based face recognition can perform recognition where the probe and gallery have same expressions. In this paper, we propose to use different expressions for recognition. Our proposal exploits the temporal order in the video and extracts the identity signature from deformation and motion separately. This is significantly different from the traditional approaches where temporal consistency is hardly used and motion and deformation are mixed. We conduct our experiments on Cohn-Canade database and the experimental results demonstrate the improvement of the proposal in terms of both accuracy and efficiency. We are pushing the state-of-the-art cross expression based face recognition in this paper.