Temporal Resolution Enhancement of Human Vocal Tract Image Sequences based on Non-Rigid Image Registration

Ana L. D. MartinsNelson D. A. MascarenhasCláudio A. T. Suazo

Dynamic magnetic resonance imaging (MRI) is anemerging technique for studying speech production. In general,vocal tract image sequences are acquired during thespeech of a word or phoneme. Sequences allow the identificationof shapes taken by the vocal tract during speechproduction. However, there is no prior knowledge aboutthe spatial and temporal resolution requirements, which areexpected to vary for different speech tasks. Available approachestry to enhance the resolution of the images by empoweringthe acquisition devices, which can be expensive.In this paper, we propose an alternative approach to enhancetemporal resolution based solely on the observed imagesequences. Our continuous motion compensated interpolation(MCI) method uses a previous non-rigid image registrationmethod, which provides an intuitive backgroundfor temporal resolution enhancement. We consider motionalong four neighboring images instead of only the enclosingneighbors. Therefore the speech articulators movementis more accurately represented. Results indicate the effectivenessof our approach

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: