Abstract
Most of facial animation applications, such as face detection and face recognition are sensitive to the different lighting conditions and background environment. Head tracking and expression analysis are important in the virtual conferencing system and also sensitive to the lighting conditions. The virtual conferencing system makes communication for people from different places. It is convenient for people to join a virtual conferencing system wherever they have a camera and a computer with internet. Some key techniques in the virtual conferencing system are sensitive to the lighting conditions. For a participant in the system, it is not easy to control the lighting conditions. Worse results may be caused by worse lighting conditions. In order to have a good performance on expression analysis, we need to have a better extraction of facial features. Having a better extraction of facial features is the goal of our approach.The first step is to extract the facial feature blocks which we desire. We extract the blocks by a projection-based method. After those blocks are extracted, a Canny edge detector is applied to each block. Then we can have binary edge images of those extracted blocks. We want to extract some points which are meaningful to the features for each block and we can determine the positions of feature points by the information of those binary edge images.