Monitoring Human Faces from Multi-view Image Sequences

The paper proposes a vision-based monitoring of human faces based on the automatic pose-invariant face detection and generation of detailed parametric descriptions of dynamic changes of face mimics. The monitoring approach involves the extraction and the classification of face features. The approach is designed especially for the monitoring of faces which captured directly during the Human Computer Interaction. The use of multi-view images provides sufficient information for the automatic head pose estimation and face mimic changes. The results of the automatic facial image analysis are described by the so-called Face Mimic Graph. The Face Mimic Graph provides the quantitative and qualitative information of face mimic changes captured during the face monitoring.