Face analysis using curve edge maps

Abstract

This paper proposes an automatic and real-time system for face analysis, usable in visual communication applications. In this approach, faces are represented with Curve Edge Maps, which are collections of polynomial segments with a convex region. The segments are extracted from edge pixels using an adaptive incremental linear-time fitting algorithm, which is based on constructive polynomial fitting. The face analysis system considers face tracking, face recognition and facial feature detection, using Curve Edge Maps driven by histograms of intensities and histograms of relative positions. When applied to different face databases and video sequences, the average face recognition rate is 95.51%, the average facial feature detection rate is 91.92% and the accuracy in location of the facial features is 2.18% in terms of the size of the face, which is comparable with or better than the results in literature. However, our method has the advantages of simplicity, real-time performance and extensibility to the different aspects of face analysis, such as recognition of facial expressions and talking. © 2011 Springer-Verlag.