Improved Tracking of Facial Features in Head, Shoulder Video Sequences

This paper describes work which is being undertaken to develop a low bit-rate switchable video codec for videophone type applications. It describes improvements to the tracking of important facial features, such as the eyes, the nose and the lips in head and shoulder video sequences. Results of investigations into a reference set updating strategy for PCA based tracking are presented. These show that adoption of a suitable updating method can significantly improve the performance of this tracking technique where significant head movement, including rotation, is present. Our proposed technique also uses the relative positions of tracked features to allow detection of frontal or non-frontal head position for use in switching between model based or enhanced feature and waveform based coding.