In this paper we address the problem of extracting face components in image sequences. We allow face's rotation, changes of illumination and addition of noises. The coarse position of a face is first tracked and predicted by simplified kalman filter. The simplified Kalman filter is very effective for tracking face areas. The actual position fo a face and face components is searched for within the predicted area by a template matching method. The face template consists of skin-color distribution and statistical face size. Experimental results show that suggested awpproach may keep tracking face area and face components very sucessfully.]
展开▼