Inexact graph matching for facial feature segmentation and recognition in video sequences: Results on face tracking

Abstract

This paper presents a method for the segmentation and recognition of facial features and face tracking in digital video sequences based on inexact graph matching. It extends a previous approach proposed for static images to video sequences by incorporating the temporal aspect that is inherent to such sequences. Facial features are represented by attributed relational graphs, in which vertices correspond to different feature regions and edges to relations between them. A reference model is used and the search for an optimal homomorphism between its corresponding graph and that of the current frame leads to the recognition. © Springer-Verlag Berlin Heidelberg 2003.