Modeling dynamic 3D face images using Hidden Markov Models


Face recognition seems to be one of the most natural and effortless biometric techniques in commonly used scenarios. Still, face recognition systems are sensitive to conditions under which the face images are acquired: pose, expression and light condition changes. Currently a rapid development of devices capturing 3-dimensional images can be observed. They become smaller, more precise and more affordable. Some of them are able to stream spatial data in real time. Including depth-seeing devices in face identification system is no longer prohibited by a high cost.
This paper proposes a method bases on modeling spatio-temporal changes in 3-dimensional image sequences. Hidden Markov Models approach appears to be the most appropriate for classifying the dynamic data. A similar approach was proposed before but mostly in context of modeling facial expression , which is in a sense orthogonal to the face recognition task. We present a method of processing 3-dimensional face images to create observations for Hidden Markov Model. We show that the presented analysis of dynamic 3-dimensional face images can improve current recognition systems.


Author: Weronika Gutfeter
Conference: Title