Upper face description by comparative analysis of gray-scale image pairs

In addition to normal speech, inter-human communication relies on other means such as facial expression. In this paper we develop an automatic method for discriminating between permanent upper face features, namely eyes and brows and transient features in nearly frontal view gray-scale image pairs. Initially, the face box is determined and the upper face features regions are extracted using a geometrical criterion. Following, precise eye localization is performed via template matching and transient features are extracted by the joint use of the Local Binary Pattern (LBP) histogram and Integral Projections. The selected methods, having the advantage of simple implementation, make feasible an interactive real-time application.