Automatic glottal tracking from high-speed digital images using a continuous normalized cross correlation

The present work describes a new approach for the automatic tracking of the glottal area from high-speed digital images of the larynx. This approach involves three processes: Firstly, the frame with the maximal glottal opening is found automatically, by detecting the frame for which the sum of pixel intensities is minimum. Secondly, a segmentation algorithm based on active contours is used to detect the glottal space and build an initial template. Finally, using the normalized cross correlation the surface that represents the best matching between the initial template and the next frame is obtained. The surface area corresponds, at the same time, to the glottal space and to the new template. The process is repeated for each frame in order to obtain the new template and the new best cross correlation surface. This process is done iteratively until the last frame of the sequence has been reached. The performance, effectiveness and validation of the approach is demonstrated even in high-speed recordings in which the images present an inappropriate closure of the vocal folds.