Frédérique Garcia, Kevin G.Munhall, Mark K. Tiede, Eric Vatikiotis-Bateson
Applying Energy-Minimization Splines To X-Ray Vocal Tract Images
Abstract:Snakes are energy-minimization splines, which have been applied successfully to a variety of visual image recognition tasks. We have implemented a version of this physical model for the analysis of sequences of X-ray ciné images of the vocal tract and video images of the face recorded during speech. After introducing the basic theory of the snake, and describing its implementation, the software interface developed for the Apple MacintoshTM platform is discussed. Then, some results achieved with these two types of images are demonstrated. It is shown that the snake can follow the tongue surface automatically through a sequence of images even though image quality is quite poor, and that it can follow the complex profile of the lips and chin in the video images. Finally, we discuss several limitations of the method, both temporary and inherent, as well as the sort of improvements we expect to implement in the future.