3D Tongue Reconstruction from Two Orthogonal Ultrasound Images

dc.contributor.authorAhadipour, Ava
dc.description.abstractTongue is the most important articulator of speech. Given the dominant role of speech in human interactions, a large proportion of the research on tongue movement focuses on its role in speech, specifically vowel production. A three-dimensional (3D) tongue model could provide important visual information that would help an individual to correct the tongue position during speech. Researchers have been tried using different imaging techniques such as MRI, CT and Ultrasound imaging to capture the tongue information. Among them ultrasound imaging is, recently, getting popular in the field of linguistic, phonetic research and medical imaging. Having access to real-time display of the movement of the tongue, non-invasive measurement with relatively cheap equipment and also avoiding risks of x-rays or other types of radiation made us use this method for capturing set of required data. We introduce a new method of reconstructing the 3D generic tongue model. The novelty of our method is employing only two ultrasound images for tongue reconstruction purpose. We presented a feature-based deformation method that could represent the different shapes of the tongue during speech with no computationally expensive requirements. Tongue is a difficult target for deformation due to its muscle complexity. The fact that tongue is concealed in the oral cavity can also provide additional challenges to capture the whole image of it using available methods such as MRI, X-Ray and ultrasound imaging. We overcome these difficulties by: (a) Analysis of tongue surface data using ultrasound imaging; Capturing tongue shape and tracking the tongue surface data for different sets of pronunciations from obtained ultrasound images (b) Deformation of 3D generic tongue model using deformation method named Radial Basis Functions(RBFs) based on two orthogonal (side and front view) ultrasound images. We tested our methodology on a set of Ultrasound images received from Department of linguistics. We selected different sample of Ultrasound frames from a set of mid-coronal and corresponding mid-sagittal images. The Ultrasound data is collected while a male person was pronouncing 1 to 10 and then sustained "i", "u", "e", "o", "a", "s", "sh", "l", "n", and "r". Our model is capable of representing tongue shapes during the production of vowels and also the general motion of tongue. A set of control points of the 3D tongue neutral model, allows each tongue segment to be easily captured. In order to test the capability and sufficiency of the proposed method in terms of deformation, we have compared our results (deformed tongue models based on arbitrary set of control points on mid-sagittal and the mid-coronal line of the 3D tongue model) with the results from an open-source tongue simulation system named “Artisynth” (deformed tongue models obtained from different activations of tongue muscles).
dc.publisherUniversité d'Ottawa / University of Ottawa
dc.title3D Tongue Reconstruction from Two Orthogonal Ultrasound Images
dc.faculty.departmentScience informatique et génie électrique / Electrical Engineering and Computer Science
dc.contributor.supervisorLee, WonSook
dc.degree.disciplineGénie / Engineering
thesis.degree.disciplineGénie / Engineering
uottawa.departmentScience informatique et génie électrique / Electrical Engineering and Computer Science
CollectionThèses, 2011 - // Theses, 2011 -

Ahadipour_Ava_2013_thesis.pdfAhadipour_Ava_2013_thesis.pdf7.59 MBAdobe PDFOpen