Exaggeration quantified: an intensity-based analysis of posed facial expressions

Posed facial expressions are characterized by deliberate and often exaggerated behaviours that usually fail to generalize to the complexity of expressive exhibition of human emotion in real-life. The detailed understanding of exaggeration and its quantification that defines each facial expression as a feature vector in the face space will provide deep insight to the growing interest in analysing posed facial expressions. In this paper, an attempt to quantify the intensity of facial expressions via estimating exaggeration as the deviation in the combined relationship between correlated landmark points that characterize the deformable face model, is made. Such relationships that underpin the discrimination of different facial expressions using exaggeration measurements is based on novel geometric morphometric inspired feature descriptors together with enhanced appearance models that seed a cascaded Support Vector Machine (SVM) classifier. In addition to demonstrating the superiority of the proposed method by applying it to images containing a wide range of expressions using standard datasets; results also distinguish different important facial landmarks for classifying that expression from every other expression.