A hybrid resampling framework for facial shape alignment

The modeling of human faces from single or multiple images can be solved by fitting an active appearance model (AAM). However, fitting such models with unknown prior estimation is a challenging task since shape and appearance are closely linked. In this paper, we address the efficiency of the sampling technique in regard to the shape alignment accuracy. The hybrid method we propose is based on a non-uniform barycentric resampling of the face model. The selected subsampled locations are improving the representativity of the generic texture reference