A computational model of facial expression

This report details a procedure for generating a computational model of facial expressions. This is a growing and relatively new type of problem within computer vision. One of the fundamental problems when trying to develop a computational model of facial expression in previous approaches is the lack of a consistent method of measuring expression. This report solves this problem by the computation of the Facial Expression Shape Model (FESM). This is a statistical model of facial expression based on an anatomical analysis of facial expression called the Facial Action Coding System (FACS). The FESM allows for the generation of a uniform mapping function, i.e. a mapping function that is not person specific. The FESM provides a robust means for upholding the rules of the FACS and is flexible enough to describe subjects that are not present during the training phase. The second contribution of this report is expression synthesis. We use the FESM for generating a synthetic expression of a subject once a neutral image of that subject is present. This is achieved using an Artificial Neural Network (ANN) in conjunction with the FESM to generate mapping functions that map contours depicting neutral expressions to contours depicting alternative expressions. Finally, the third contribution of this report is classification of expression. We detail two types of machine learning algorithms that provide the greatest potential for classification of facial expressions within the expression space. These techniques are formally known as Radial Basis Function (RBF) networks and Support Vector Machines (SVM). The report is largely concerned with statistical models, machine learning techniques and psychological tools used in the generation of a computational model of facial expression. This model can be used for synthesis and expression classification once a neutral image of a subject is present. This provides a means for a level of interaction with a computer through facial expression synthesis and recognition and is a significant step forward in human-computer interaction.