Statistical glottal segmentation of videoendoscopic images using geodesic active contours

The use of high-speed videoendoscopy aims at facilitating a diagnosis of voice pathologies. In this work, the challenge is to track the motion of the vocal folds to detect a possible paralysis. To this end, locating the glottis in each image by a suitable segmentation turns out to be a crucial step. In this respect, we propose a segmentation method based on geodesic active contours. The novelty of our approach lies on the use of a geodesic metric that controls the evolution of the active contour according to the statistical distributions of the image regions. Experimental simulations carried out on glottal images taken at different stages of the glottis cycle indicate the outperformance of the proposed approach over several reported glottal segmentation techniques.