Full-Automatic Glottis Segmentation With Active Shape Models