A fast and accurate algorithm for facial feature segmentation

A facial feature segmentation algorithm for head and shoulder sequences is proposed. The method is based on simple spatial and temporal heuristics techniques for the purpose of practical implementation. Facial features are segmented using two common cues (i) regular and high image intensity variation in the temporal domain due to various motion like talking and moving, and (ii) high edge density around the feature regions. A simple motion detection method and a generic spatial operator are applied to achieve the required segmentation. The paper addresses the issue of how to combine both information to obtain final results and also assesses the segmentation performance