Abstract: Face recognition is a difficult visual representation in large part because it requires differentiation among human faces, which vary subtly from each other. The objective of this study, is to use the information that outlines the facial features using the curvature scale space. Pointwise curvatures, which are �Natural Signatures�, of facial features are well suited for facial feature recognition. The accuracy of such feature-based recognition is very high because the value of curvature at a point on the surface is viewpoint invariant. A novel method for extraction of the facial feature signatures from the curvature map of the human face is presented in this study. Comparison between two faces is made on their relationship in feature face using ART neural network. Satisfactory results using diverse probe data prove that curvilinear facial feature signatures provide vital clues in distinguishing and identifying human face.
Mary Metilda and T. Santhanam , 2007. Face Recognition Using Curvilinear Feature Signatures . Asian Journal of Information Technology, 6: 771-777.