Radial Basis Network for Facial Expression Synthesis

| Many multimedia applications require the synthesis of facial expressions. We demonstrate the synthesis of di erent degrees of various 2D grayscale facial expressions using the Radial Basis Function (RBF) neural network. The RBF network is used to generate spatial displacement of a set of facial feature points as in multidimensional density interpolation. We have implemented two RBF networks for synthesizing facial expressions. One network generates the spatial displacement of facial feature points for di erent degrees of expressions and the other generates the spatial displacement of facial feature points of mixed facial expressions. The predicted facial landmark displacement information is then fed into our image warping algorithm along with the expressionless facial image to produce the nal synthesized facial image. We discuss the method used and demonstrate the results.