A Novel Shape Transformation Approach for Quantizing Facial Expressions

In this paper, a novel methodology for facial expression rating is proposed, as the intensity of an emotion evolves from neutral to high expression. The face is modeled as a combination of sectors and their boundaries. An expression change in a face is characterised and quantified through a combination of non-rigid deformations. After elastic interpolation, this yields a geometry-based high-dimensional 2D shape transformation, which is used to register regions defined on query-faces. This shape transformation produces a vector-valued deformation field and is used to define a scalar valued Sector Volumetric Difference (SVD) function, which characterises and quantifies the facial expression. A two-stage expression classification is used with first stage detecting low, medim and high levels of expressions, and the second stage involving a HMM-classifier for recognizing six different facial emotions- anger, disgust, fear, happiness, sadness and surprise. Further, the proposed shape transformation approach is compared with marker based extraction method for extracting facial expression features. The performance evaluation done on a Italian audiovisual emotion database DaFex comprising facial expression data from several actors eliciting five different emotions - anger, disgust, fear, happiness, sadness and surprise at different intensities (low, medium and high), shows a significant improvement in expression classification for the proposed shape-transformation approach.