Fully-automatic glottis segmentation with active shape models