Person identification using behavioral features from lip motion

Traditionally lip features have been used in speech recognition, but lately they have also been found useful as a biometric identifier. Since this application of lip features is relatively recent, several limitations can be observed. Firstly, majority of the research has focused on using physical attributes of the lips, such as shape and appearance. Secondly they have experimented on comparatively short length, text dependent videos. We believe that visual speech, along with the physical attributes also contains certain behavioral attributes that can be better learned from a text independent, long duration database. Therefore in this paper we have analyzed a specifically designed database and extracted behavioral lip features, to study their utility for person recognition.