“An application of facial animation techniques to expression normalization for robust face recognition” by Wiratanaya, Hahnel and Kraiss

The human face is subject to many variations which complicate the successful identification of an individual. In addition to pose and lighting variations the recognition of a face across different facial expressions has been identified as a major problem [Kraiss 2006]. Existing algorithms handling facial expressions either enlarge the existing set of training images by generating artificial views of every individual [Jiang et al. 2004] or use statistical a-priori knowledge about possible facial expressions to normalize the face (e.g.  Active Appearance Models (AAM) [Cootes and Taylor 2004]). To obtain a satisfying registration of an AAM face graph in the presence of a wide range of facial expressions, the statistical model has to be built from a large database covering many different facial expressions and individuals. Additionally, all images have to be precisely annotated which can be tedious and time-consuming. Creating artificial variations of existing database images partially solves this problem. However this approach still only generates preset variations of the database images instead of being adaptive to the actual facial expression present in the input image.  

1. Cootes, T., and Taylor, C. 2004. Statistical Models of Appearance for Computer Vision.
2. Jiang, D., Hu, Y., Yan, S., and Zhang, H. 2004. Efficient 3D Reconstruction for Face Recognition. Journal of Pattern Recognition, Special Issue on Image Understanding for Digital Photographs.
3. Kraiss, K.-F., Ed. 2006. Advanced Man-Machine Interaction – Fundamentals and Implementation. Signals and Communication Technology. Springer.
4. Waters, K. 1987. A muscle model for animating three-dimensional facial expressions. Proceedings of Siggraph’87, 17–24.