Feature tracking method for motion parameter estimation in a model-based coding application

We present a method by which the dynamics of facial movement may be parameterised for application in a model-based image coding scheme. A tracking algorithm is described whereby the bounding boxes of the eyes, nose and mouth of the subject are initially located and then tracked over subsequent frames using both block matching and code-book search techniques. The six degrees of freedom required to define the position and orientation of the head are derived from the tracked box positions by means of a motion parameter estimation algorithm. Implementation of the algorithm involves interpreting the spatial distribution of the box positions and relating them to a simplified topological three-dimensional model of the face.