We demonstrate a novel method of generating a mapping function from the shape of a neutral face to the shape of one showing an expression. It is proposed that this mapping function can be used to automatically generate facial expressions from still images of never seen before faces or classify which expression a person is portraying. This new technique draws on the work of Ekmans [8] Facial Action Coding System (FACS), Cootes [4,5] Active Shape Model (ASM) and Artificial Neural Networks (ANN).
To build an ASM it is required to have a training phase, where each image in the training phase is ‘scored’ by the FACS system and a neural network is used to generate a mapping function as a face moves from a neutral expression to an alternative expression. We describe this method in detail and give results indicate the effectiveness of the technique.