Anatomy based features for facial expression recognition

In this study we propose a set of anatomy based features for facial expression recognition. The muscle forces that constitute an expression are solved by tracking carefully selected facial feature points. These points are initialized in the muscular regions of influence on the first frame of the video. They are tracked using the optical flow algorithm. The displacements of facial feature points are used for estimation of 3 dimensional head orientation and deformations due to expressions. We model human face with springs as an over-determined and linear system of equations. This system is solved under the constraint of facial anatomy for muscular activities. We use sequential forward selection to determine the most descriptive set of features for classification of basic expressions.