Segmentation technique based on object movement for Speech Production simulation

Abstract

In this paper, we propose the segmentation technique by using Snakes for 4dimentional magnetic resonance imaging (MRI) data of the movements of the oral tract shapes when pronunciation is performed. In the segmentation of the time-sequence MRI images, there is the specific problem that depends on a tongue shape quick deformation. We found that the Optical Flow of image sets is useful criteria for the decision of the geometry of control points in Snakes. Compared to the normal Snakes, our original method which modified Snakes by utilizing Optical Flow demonstrated superior accuracy of the segmentation of 4D MRI data.