GLOTTAL AREA DETECTION IN VIDEOKYMOGRAPHIC IMAGES USING WATERSHEDS

C. Manfredi, L. Bocchi, V. Osma-Ruiz, J.I. Godino-Llorente, N. Sáenz-Lechón

The observation of the dynamic behaviour of vocal fold vibration is of great diagnostic relevance for the clinical specialist. Nowadays, one of the most prominent observation methods is videokymography (VKG). This technique, unlike classical stroboscopy, allows the visualization of a-periodic movements of the vocal folds, as it does not need any optical illusion to be used to show the vibration. VKG allows retrieving a single line per photogram in a high-speed video recording, and builds up an image appending lines in a vertical sequence. Lines can be extracted from previously recorded videos or can be selected by means of a footswitch during the recording.
There are several parameters that can be extracted from VKG images (ratio of: amplitude, period, opening-closing phase, etc …), that are helpful for the medical specialist to objectively quantify the vibratory movement of the selected line of the vocal folds. To obtain reliable and automatic measures from VKG images, it is necessary to use numerical tools that allow discriminating zones corresponding to glottal areas from other parts of the image. This process is rather involved because, although the glottal areas always correspond to dark zones of the image, the variation of the grey levels in those areas can be very high.
In this work, we present a new method for the automatic extraction of glottal areas from VKG images. Differently form other approaches, the proposed method, based on the watershed transform, does not require any kind of initialization. Results obtained for 6 pathological and one normal subjects are compared with another thresholding technique, used in previous works. The error made by both methods with respect to manual detection of glottal areas is also measured and compared.

L. Bocchi ^a, V. Osma-Ruiz ^b, C. Manfredi ^a, J.I. Godino-Llorente ^b, N. Sáenz-Lechón ^b

^aDept. of Electronics and Telecommunications, Università degli Studi de Firenze, Via S. Marta 3, 50139 Firenze, Italy
^bDept. of Circuits and Systems Engineering , Universidad Politécnica de Madrid, Cra. Valencia Km 7, 28031, Madrid, Spain

Corresponding author: C. Manfredi
Tel.: +39 055 479 6410, fax: +39 055 494 569, e-mail: manfredi@det.unifi.it